Wikipedia talk:Citing sources#Citations in a list

{{discussion top|result=There was not consensus to deprecate list-defined references. Some editors find it awkward, especially when editing a specific section, but other editors like that it makes the main wikitext less cluttered.

::There was 2:1 support in favor of deprecating {{reflist|refs=}} and replacing existing instances. I updated the linked documentation pages to do so. Someone will need to write a bot and follow the procedure at Wikipedia:Bots/Requests for approval. At least one editor had concerns about bots making incorrect edits. There was also discussion of whether or not such changes should be bot-flagged so they don't show up on watchlists, and whether it should be required that other changes be made at the same time. The bot approval process is designed to take these concerns into account and balance them against the proposed benefits; that would be the place to raise them. (It might be helpful if whoever makes the requests notifies the editors who participated in this discussion.)

::There is a big difference in the number of changes (and also pros and cons) between eliminating {{tl|reflist}} to the fullest degree technically possible given the limitations of {{tag|references}}, vs. only eliminating {{tl|reflist}} for articles that use list-defined references. The latter was the question asked, but the former was the solution given in the proposed bot changes. From this discussion it is only safe to say there is support for the latter, and I recommend scoping the bot approval request to that. One editor did urge an "all or nothing" approach, but that was not echoed. If the first round of substitution goes well, we can always discuss more widespread deprecation of {{tl|reflist}}. I did not update the linked documentation pages to say that {{tl|reflist}} was deprecated outside the context of list-defined references. -- Beland (talk) 22:47, 30 May 2025 (UTC)

}}

List-defined references are a pain for VisualEditor users. It displays "This reference is defined in a template or other generated block, and for now can only be previewed in source mode." instead of the actual content of the reference when using the VisualEditor. Modifying the references requires switching to the source code editor, but not everyone is familiar with its syntax. I don't know why the VisualEditor doesn't handle them better, it doesn't seem unsolvable from a programming perspective and I would be fine with list-defined references if it did, but unless there are plans to fix this, perhaps we should discourage it? I'm curious to know what more experienced contributors think. Alenoach (talk) 20:44, 1 March 2025 (UTC)

:VisualEditor is crap. It's VisualEditor that should be discouraged. Jc3s5h (talk) 21:10, 1 March 2025 (UTC)

::Two notes on this:

::# The VisualEditor (VE) can preview a list-defined reference. Check out police jury in the VE. When I rewrote that, I used list-defined references, but no templates. In the VE, you can preview, modify, and reuse the list-defined references. You cannot add new list-defined references, delete existing ones, comment out existing ones, or replace existing ones. The VE will treat any template used within another template as just text. I don't think there is anything in the pipeline to fix that.

::# [https://meta.wikimedia.org/wiki/WMDE_Technical_Wishes/Sub-referencing Sub-referencing] is meant to be the official solution to citing different pages and it is meant to be built on list-defined references, although it looks like that is causing problems for the team.

::Rjjⁱⁱⁱ (talk) 21:24, 1 March 2025 (UTC)

:::Ah indeed, your way of doing it outside the "reflist" template works better with the VisualEditor. I still believe that inline references are more beginner-friendly, but your approach is a clear improvement compared to putting it in the reflist, thanks. Alenoach (talk) 21:42, 1 March 2025 (UTC)

:Entirely agree with Jc3s5h. If the problem is that VisualEditor can't hack it, then the problem is VisualEditor. We should not warp our usage of helpful article-source organizational tactics because of bad tooling foisted on us by Wikimedia. —David Eppstein (talk) 21:37, 1 March 2025 (UTC)

:: I'm disappointed that m:WMDE Technical Wishes/Sub-referencing is going off in a weird direction because VE doesn't do LDR well and they don't want to work on fixing that. Anomie⚔ 11:43, 17 March 2025 (UTC)

:::It is so disappointing. I raised the issues that VE would cause for their plans [https://meta.wikimedia.org/wiki/Talk:WMDE_Technical_Wishes/Sub-referencing/Archive#Cart_before_the_horse_(list-defined_references) over a year ago] and they were dismissive about it then, seeming to frame it as beyond the scope of their project. But then like, who is it in scope for? Is there a team or even a person working for the WMF that has a long-term vision for how to improve referencing, or is the long-term plan to just hope we figure it out? There are limitations to what can be done with the current system; that's why using {{tl|sfn}} feels like putting a puzzle together and {{tl|rp}} is so basic. Rjjⁱⁱⁱ (talk) 00:40, 18 March 2025 (UTC)

:I've tried WP:LDR on a couple of articles, and I find it to be inconvenient, especially if you're using section editing. I think we should discourage it. WhatamIdoing (talk) 06:30, 16 March 2025 (UTC)

::LDR is the preferred method for many editors. It has pros and cons, but it should not be discouraged. -- Michael Bednarek (talk) 06:36, 16 March 2025 (UTC)

:::I wonder whether we could find out what percentage of articles actually uses it. There is a cost (in editor's time to learn about yet another different system) to maintaining unpopular arrangements. WhatamIdoing (talk) 02:34, 17 March 2025 (UTC)

::::Template {{t|Use list-defined references}} has >5300 transclusions. Nobody has to learn about it; as with other citations, other helpful editors will convert citations non-conforming to an article's established style. -- Michael Bednarek (talk) 03:08, 17 March 2025 (UTC)

:::::So it's used in less than one in a thousand articles. That's barely any use at all.

:::::Yes, people do have to learn about it – if they want to be able to fix the citation formatting problem that brought them to the page; if they want to be able to remove a citation without getting an ugly red error message on the page; if they want to understand what's going on with the page so they don't have to rely on "other helpful editors", especially the ones whose "helpfulness" manifests in the form of yelling at them for not doing everything perfectly the first time. WhatamIdoing (talk) 03:16, 17 March 2025 (UTC)

:::::@Michael Bednarek, that seems too low? At least one list-defined reference is used in at least [https://bambots.brucemyers.com/TemplateParam.php?wiki=enwiki&template=Reflist 179,000] articles, based on the "ref" parameter in {{tl|reflist}} transclusions. Rjjⁱⁱⁱ (talk) 04:26, 17 March 2025 (UTC)

::::::I suspect that many of those uses occur in articles with mixed citation styles. But that number further clarifies that discouraging LFDs is impractical. -- Michael Bednarek (talk) 04:43, 17 March 2025 (UTC)

:::::::Well... "discouraging" is usually a long-term and largely inactive/passive process. You just write something in some documentation and leave it for five or ten years, and let community members make individual choices. You could write something as strong as "being discouraged but not banned", but you could also write something like "relatively unpopular" or "less popular than shortened footnotes". WhatamIdoing (talk) 05:11, 17 March 2025 (UTC)

::::::That suggests that about 1 in 40 articles is using that, at least partially. That feels like a more plausible estimate to me. WhatamIdoing (talk) 05:08, 17 March 2025 (UTC)

:::::::Just cruising thru, not reading the arguments: I use {reflist|refs= cos you can better read the text in source mode. Putting the refs in the body text looks like spaghetti code and can make a passage almost unreadable. And a key point: it's rare for someone editing to want to change the ref. Editors usually want to do things to the text. If you need to read the ref that's better done in reader mode. You might want to delete the ref; that is different. Wanting to make changes to ref itself are rare and are are usually like to add the date or something -- important but not usually key; you're not going to change the title or the author etc. Sometimes I have to find the ref tags in all that text, do linefeeds to get the refs out of the way to even read the text, then put them back -- not a huge deal but not excellent. Sometimes I'm like "Jeez this's a dog's breakfast, I'll just not do the edit I was intending to do." Herostratus (talk) 02:24, 24 March 2025 (UTC)

::::::::I suggest using "" rather than "{{reflist|refs=", as shown here. The difference is not significant for source editor users, but "" will not make references hidden for the VisualEditor users. Alenoach (talk) 02:37, 24 March 2025 (UTC)

::::::::Huh, it's odd that you describe straight-through wikitext as spaghetti code, because I think that the jumping-back-and-forth style of LDR is much more spaghetti-like. WhatamIdoing (talk) 04:43, 24 March 2025 (UTC)

:::::::::@Herostratus, Are you confident about "And a key point: it's rare for someone editing to want to change the ref. Editors usually want to do things to the text"?

:::::::::A while ago, an article appeared on my watchlist. I hadn't looked at it in years. There were something like 50 edits over five years. Not a single word of text was changed. WhatamIdoing (talk) 01:04, 21 April 2025 (UTC)

::::::::::Well what were all those editors doing, not seeing the connection.

::::::::::Well I would be pretty confident EXCEPT I now realize that adding the archive url etc, is probably pretty common. So I have to back off from that. Herostratus (talk) 02:59, 21 April 2025 (UTC)

:::::::::::All of those editors were fiddling with non-content stuff, including but not limited to ref formatting.

:::::::::::I increasingly wonder whether we could do a decent study about who writes Wikipedia's contents. High-volume editors do a lot of reverting/blanking, and we do a lot of fiddling with wikitext (some of which is actually useful to the occasional person, e.g., adding archive URLs), but I wonder whether newbies add more content. If a new paragraph is added (and sticks), who added that? WhatamIdoing (talk) 17:50, 21 April 2025 (UTC)

::::::::::::Almost 20 years ago, Aaron Swartz made [https://web.archive.org/web/20140803134036/http://www.aaronsw.com/weblog/whowriteswikipedia a study] in which he found that Wikipedia's actual content is indeed largely written by the newbies and non-regulars. Whether that's still the case is an open question, but it sounds plausible to me. Gawaon (talk) 07:52, 22 April 2025 (UTC)

::::::::::::That would be interesting and might even be useful. I know that my own editing patterns tend to shift around a lot, with bursts of content creation interspersed with assorted gnomery of many types. A lot depends on chance. I see something that needs to be fixed, and if I am in the right mood I fix it if I can, and it often leads to something else related or of a similar type. Other times I fixate on cleaning up or improving something on a larger scale, and then there are policy discussion.... Cheers, · · · Peter Southwood ^(talk): 12:26, 22 April 2025 (UTC)

:No, list-defined references are cleaner and more elegant than the mess you get when the reference details are embedded in the article's prose. The Visual Editor issue is best addressed by enhancing the Visual Editor and/or avoiding the {{tl|reflist}} template as suggested below. Andrew🐉(talk) 20:31, 8 May 2025 (UTC)

Per Alenoach above, handles list-defined references in exactly the same manner as {{reflist|refs=, but with the benefit of VisualEditor support. Should we discourage or deprecate {{reflist|refs= in situations where there are no other parameters passed to {{template link|reflist}} and use a bot to replace all such occurrences with ? Dan Leonard (talk • contribs) 12:49, 29 April 2025 (UTC)

:Isn't there, at the very least, a difference in font size? Gawaon (talk) 13:05, 29 April 2025 (UTC)

::Originally that was true, but since 2010 the font sizes have been the same. Compare angle, which uses {{reflist|refs=, versus Sheetz–Wawa rivalry, which directly uses . Both their reference lists have font-size: 90%;, albeit the former is styled by the CSS class .reflist in addition to .references. Dan Leonard (talk • contribs) 13:13, 29 April 2025 (UTC)

:It would be good to have the software engineering perspective, so I opened [https://www.mediawiki.org/wiki/VisualEditor/Feedback#List-defined_references a discussion here]. I hope we will get an answer about whether the VisualEditor can be improved, or otherwise the design rationale. Alenoach (talk) 13:37, 29 April 2025 (UTC)

::Oppose breaking our markup because of limitations in VE. If VE is broken the solution is to stop using VE, not to break more things. —David Eppstein (talk) 17:02, 29 April 2025 (UTC)

:::Might be good to wait for a deeper understanding of the problem before taking a decision. Alenoach (talk) 17:10, 29 April 2025 (UTC)

::::Ok, apparently this is the main page where the software developers handle it: [https://phabricator.wikimedia.org/T52896 T52896]. It's a major issue that has been there for more than 10 years, and the inability to parse references inside templates also seriously impacts translation tools and infoboxes. One software engineer said in 2014 that fixing it would be too hacky and that there is no good and generic solution, and complained about the templates. No one is working on it, so I guess they don't plan to address the issue. Alenoach (talk) 19:57, 29 April 2025 (UTC)

:::This is essentially just substing a somewhat-redundant template. When called in LDR contexts without other parameters, {{template link|reflist}} appears to just call and are already listed in the documentation as equivalent, so I don't see what you think would break. Dan Leonard (talk • contribs) 18:16, 29 April 2025 (UTC)

::Oppose I love love love using the LDR, if it were a kitten I would carry it in my pocket with me everywhere. If visual editor is the problem, then fix visual editor. Sgerbic (talk) 17:21, 29 April 2025 (UTC)

::: {{re|Sgerbic}} Note the proposal in this subsection is to require doing LDR using {{tag|references}} rather than {{tlx|reflist|refs{{=}}...}}, not to deprecate LDR. Anomie⚔ 18:09, 29 April 2025 (UTC)

::::Sorry, all this coding language is confusing to me as I am only a general editor. I love using reflist|refs, all the articles I write use this style and would hate to see something so tidy and easy to use to be replaced with something so messy and awkward. Sgerbic (talk) 18:41, 29 April 2025 (UTC)

:::::@Sgerbic It shouldn't be any more complicated. You can write your LDRs as you would, but instead of wrapping them in the reflist template, add tags on either end. Cremastra _talk 01:34, 1 May 2025 (UTC)

::::::That is very old school, I want to continue using reflist|refs it is so much neater. I'm not understanding why continuing as I have for the last few years is a problem. If there is a problem with visual editor then that should be fixed. Possibly I am not explaining myself well, this is an article I just rewrote a couple days ago Jotham Johnson. Sgerbic (talk) 03:15, 1 May 2025 (UTC)

:::::::The only difference is that where you currently type:

:::::::{{color|green|{{Reflist|refs=}}

:::::::{{cite web |title=Prof. Jotham Johnson, 61, Dies; Chairman of Classics at N.Y.U. |url=https://

:::::::(etc., all the way down, and then ending with a closing }}), you would instead type:

:::::::{{color|green|}}

:::::::{{cite web |title=Prof. Jotham Johnson, 61, Dies; Chairman of Classics at N.Y.U. |url=https://

:::::::(etc., all the way down, and then ending with a closing ). Or, more realistically, you would do the same thing that you're doing now, and every now and again, a bot would replace the unnecessary template with the original wikitext code.

:::::::Do you understand how small the recommended change is? It's literally just a few characters difference in the whole page. WhatamIdoing (talk) 03:37, 1 May 2025 (UTC)

:::::::The problem with every script in recognizing reflist template LDRs has existed and been investigated for over a decade. If you could fix it without hacky tape, that would be nice. Aaron Liu (talk) 10:57, 1 May 2025 (UTC)

::I opened [https://phabricator.wikimedia.org/T392942 a bug report here]. Alenoach (talk) 18:01, 29 April 2025 (UTC)

:Strong support In addition to better visualeditor support, using {{tag|references}} means that most of the citations will still display even if the WP:PEIS limit is exceeded. All the old reasons to use {{tl|reflist}}, such as font sizes or responsive columns, have long since been overcome by the software. Nothing about using {{tag|references|content={{cite foo|...}}}} for list-defined references is harder or less "tidy" than using the template, and in fact I'd argue that bracketing a long list with tags is more "tidy" than encapsulating it into a template parameter. The only thing holding us back is inertia from the days before the raw tag had feature parity with the template. --Ahecht (TALK
PAGE) 18:50, 29 April 2025 (UTC)

:Oppose: visualeditor is broken, not {{para|refs}}. Boghog (talk) 19:37, 29 April 2025 (UTC)

:I worry that deprecating the reflist template, but only for LDR is going to cause confusion. Good faith editors are likely to use reflist, as it's what they will see commonly elsewhere and only after being told of the situation understanding that the common method shouldn't be used in this specific case. LDRs are not common, so many editors could go a long time before coming across this situation. It would seem the better solution would be either to move away from using the reflist template (if it's true that tags now have all the same functionality), or from using LDRs. -- LCU ActivelyDisinterested «@» °∆t° 20:02, 29 April 2025 (UTC)

::There's no reason for anyone to use {{tl|reflist}} in 99% of cases, LDR or not. It should be deprecated across Wikipedia, not just for LDRs. --Ahecht (TALK
PAGE) 20:24, 29 April 2025 (UTC)

:::I think @Ahecht is right. Most of the times, using {{tl|reflist}} instead of is like using a template to make text bold instead of using the ordinary ''' wikitext code. Back in the day, there were several differences (importantly, the template could do columns ~15 years ago, before the wikitext code had that built in), but those differences are few and far between now, and IMO the template should only be used when non-default features are actually wanted.

:::BTW, when we started using the template, we had the same arguments: Using the template is going to cause confusion, because people are used to the wikitext code. It's not actually a big deal. People figure out it. WhatamIdoing (talk) 20:56, 29 April 2025 (UTC)

::::My point about confusion was in regard to a situation where reflist would be used for the vast majority of articles, but not on the few that use LDRs. A situation that would be quite different from when editors started using the template. This wouldn't be the reverse, but a janky halfway solution. -- LCU ActivelyDisinterested «@» °∆t° 08:19, 30 April 2025 (UTC)

:::There is still a reason: if we ever want to add extra coding to references beyond the basic formatting that the references tag provides, having it in a template makes it easy and avoids having to persuade Wikimedia to maybe do it someday if they ever find the interest to listen to us. That is, it is more flexible and more robust.

:::Beyond all that, there is another reason: changing existing reflists to references tags in millions of articles would represent an enormous clog-up of everyone's watchlists for however long it would take to do so. —David Eppstein (talk) 23:44, 29 April 2025 (UTC)

::::Bots are hidden from watchlists by default, I think. Aaron Liu (talk) 03:21, 30 April 2025 (UTC)

:::::Yes, it's quite trivial to hide bot edits, either through preferences at {{Preferences|Watchlist|Changes shown|check={{int:Tog-watchlisthidebots}}}}, or ad-hoc using the filter button on the watchlist itself. Dan Leonard (talk • contribs) 03:55, 30 April 2025 (UTC)

::::::Hiding bots means never seeing all the damage the bots sometimes do. I regularly check edits by bots and report on bad edits by bots. 99% of the time they are ok but that remaining 1% needs checking. I cannot do that if my watchlist is overwhelmed by thousands of bot edits.

::::::Also, this issue goes far beyond list-defined references: it appears to be a general issue with VE not handling templates nested inside other templates. Working around it in this case will merely take pressure off the VE developers to make VE work without doing anything about the broader problem. —David Eppstein (talk) 04:58, 30 April 2025 (UTC)

::::::: Is that pressure having any effect anyway? Anomie⚔ 11:28, 30 April 2025 (UTC)

:::::::Just mark everything as seen right after the fixbot does things; in the worst case you can always just filter out the fixbot (which will probably be a one-off ish bot used to answer Wikipedia:AWB/R). And the pressure with reflist has been on to ten years; no one has found a non-hacky (without {{tq|making up a list of hack templates that each wiki uses, which is a WONTFIX if ever there was one|q=y}} which I agree with) solution. Aaron Liu (talk) 12:27, 30 April 2025 (UTC)

• Support : the problem seems unlikely to be solved any time soon ([https://phabricator.wikimedia.org/T52896 T52896]). Maybe they underestimate how impactful the problem is, or maybe we underestimate the technical obstacles. But the VisualEditor is not a minor feature, so we should do what we can to accommodate its users. I don't see a good reason for using {{tl|reflist|refs=}} instead of . Alenoach (talk) 20:31, 29 April 2025 (UTC)
• Support I use list-defined references as the most convenient way of handling citations. I wasn't familiar with the technicalities which have been presented here but they seem to make good sense. I'll start using to see how it goes.

: Note that this discussion has technical issues because of embedded tags which need considerable effort to decativate. I've just fixed Alenoach's post to turn off template bracketing which was messing up my post. We need some clerking to keep everything well-organised.

: Andrew🐉(talk) 06:51, 8 May 2025 (UTC)

• {{started}} a new draft proposal at the idea lab since this will require a fairly large site-wide RFC if implemented. Dan Leonard (talk • contribs) 22:32, 29 April 2025 (UTC)Moved below per Mike Christie Dan Leonard (talk • contribs) 02:51, 30 April 2025 (UTC)

::Dan, I'm not sure what would have been wrong with leaving the RfC here and simply advertising it. Since I'm commenting I'll mention that I use VE by preference; I never use LDR but would prefer to be able to edit those articles with VE if I come across them. I agree with some of David's points above, though; I don't think anything should be implemented that would flood watchlists, and I don't see any benefit in changing usages of reflist that are not implementing LDR. If you're going to reword this RfC, I think it should be narrowly defined. I'd specify that no bot edits should be done except when the edit accompanies an edit that would have been made anyway, to keep this off watchlists; and it should only affect LDR articles. David, the one point of yours I don't agree with is that we should leave reflist usages in place just in case someone finds them useful for parameter addition in the future. I think present value (to VE users, of whom there are many) is better than some possibly non-existent future value. Mike Christie (talk - contribs - library) 01:24, 30 April 2025 (UTC)

::One of the most basic use cases of {{tl|reflist}} is to allow multiple columns in the reference section on sufficiently wide screens and to control how wide these columns shall be. How does the <references/> tag handle this? Sorry if this is a noob question, but I didn't find it by a quick look at the docs. Gawaon (talk) 06:43, 30 April 2025 (UTC)

:::It doesn't handle it, and in those cases the reflist tag should be left in place. Though as far as I know one would never use the columns parameter with refs=, so it would be out of scope of this proposal. Mike Christie (talk - contribs - library) 11:33, 30 April 2025 (UTC)

:::: That's not entirely correct. {{tag|references|s}} and {{tlx|reflist}} both use 30em columns when there are more than 10 refs. This can be disabled with {{tag|references|s|attribs=responsive=0}} and {{tlx|reflist|1}}. OTOH, reflist has more options: {{tlx|reflist|2}} and {{tlx|reflist|30em}} do columns without the "more than 10 refs" condition, and other widths besides 30em can be passed to {{tlx|reflist}} too. Anomie⚔ 11:45, 30 April 2025 (UTC)

:::::This (having multiple columns without using the template) was implemented in 2017. WhatamIdoing (talk) 17:11, 30 April 2025 (UTC)

::::::Including widths besides 30em? Aaron Liu (talk) 17:23, 30 April 2025 (UTC)

:::::::Aaron, I don't know the answer to your question (regardless of whether your question is a "When was widths besides 30em added?" or "Was support for widths besides the site-defined default ever added?"), but I wonder whether it matters in practice. I've never seen someone combining narrower column widths with LDR, because {{tl|sfn}} is the main use case for narrower column widths, and those aren't put into LDR. Nobody's talking about an absolute requirement to do this without exception. A bot/AWB script that's capable of detecting whether an article is using LDR could trivially be programmed to leave it alone if other parameters are being used. WhatamIdoing (talk) 19:24, 30 April 2025 (UTC)

:What about {{tlx|reflist-talk|2=refs=}}? --Shmuel (Seymour J.) Metz Username:Chatul (talk) 08:26, 30 April 2025 (UTC)

Since I think this would be a very controversial and thorny RFC if proposed, I've started a proposal below for discussion. Dan Leonard (talk • contribs) 02:52, 30 April 2025 (UTC)

{{divbox|1=plain|2=Working proposal (please edit)|3=

Early features of {{tlx|reflist}} included a different font size (90%) and automatic column handling. Today, both of these features are built into the Cite extension. As a result, the template, when called alone, is effectively a basic wrapper for {{xtag|references|s}}. Usage of {{tlx|reflist}} where the {{para|colwidth}} parameter is given are outside the scope of this proposal. Editors are still free to use the template where its features are actually desired.

VisualEditor cannot handle list-defined references when those references are defined inside of a template like {{tlx|reflist}}; however, it can handle {{xtag|references|p|content={{xtag|ref|p}}}} properly due to its closer integration with MediaWiki. Almost 200,000 pages have references defined inside of {{tlx|reflist}} and thus those entire articles cannot be contributed to by VE users ({{Template parameter usage|reflist|lc=yes|label=none}}).

Documentation on this is lacking. Help:List-defined references says VE doesn't support the referencing style at all, and treats both methods as equivalent.

The VisualEditor war has been going on for a decade now and it's now used by a not-insignificant number of editors. While I think we'd all love if it worked well and didn't dump horrible {{xtag|ref|params=name=":0"}} tags everywhere, "WMF should fix it, we shouldn't have to touch our templates" has gotten old.

WMDE is developing a Wikitext-based reference reuse handling they're calling sub-referencing. This should benefit editors of list-defined references by allowing page numbers to supplement citations to such references. Templates like {{tlx|reflist}} were cited as a major hurdle in the development of this tool.

When the WP:PEIS limit is reached within an article before the references section, the reflist template will just render as Template:Reflist, whereas the {{xtag|references|s}} tag will still function (although some citation templates may not render properly).

The Polish Wikipedia editors routinely check if {{Template link interwiki|lang=pl|Przypisy}} includes a list of references and, using pl:MediaWiki:Gadget-sk.js#L-575, replace it with {{xtag|references|s}}. One sticking point to keeping {{Template link interwiki|lang=pl|Przypisy}} without a list of references on that project was the replacement of a Polish word with an English one; however, this isn't relevant on the English Wikipedia.

The following pages will need to be updated to remove the equivalence between {{tlx|reflist}} and {{xtag|references|s}} and encourage use of the latter by new editors:

• Help:Footnotes
• Wikipedia:Citing sources
• Help:List-defined references
• Template:Reflist/doc
• Template:Sfn/doc
• Template:R/doc
• Template:Harvard citation/doc

Uses of the {{tlx|reflist}} template that do not rely on its

• Some parts of the proposal above are impossible to read in dark mode because they have white text on white background. See mw:Recommendations for night mode compatibility on Wikimedia wikis for suggested fixes. -- Beland (talk) 10:07, 30 April 2025 (UTC)
• :This has been fixed; thanks! -- Beland (talk) 17:47, 30 May 2025 (UTC)
• What's the point of this proposal? I doubt it has more than a snowball in hell's chance of making it due to the flexibility of {{tl|reflist}} to be used with or without a colwidth parameter making it too useful to give up without a good reason (which doesn't seem to exist), and spurious RfCs are a waste of editors' time. Why not rather focus on the original idea of deprecating the use of {{tl|reflist}} with refs=, which seems to be the only actual problem anyway? Gawaon (talk) 11:50, 30 April 2025 (UTC)
• :I don't see how specifying per-article column widths is that useful. Aaron Liu (talk) 12:32, 30 April 2025 (UTC)
• ::Well, following the recommendations in Template:Reflist, I tend to use the default setting when most references are "long" and there is no extra Bibliography section, while setting the colwidth to 20em for articles that use mostly short author–year references ({{tl|sfn}} and friends) resolved in a Bibliography section. I also use 25em in articles that mix both styles more or less evenly. It's not essential, but it gives articles a nicer look. Gawaon (talk) 13:01, 30 April 2025 (UTC)
• :::Hmm, that does makes sense. Would a template that just applies a pre-set column width of either 20em or 25em on the fix that? Would not be hard to script if it just applies 20em or 25em every time, though it wouldn't support user-specified custom column widths. Aaron Liu (talk) 13:14, 30 April 2025 (UTC)
• ::::Well it's in any case slightly more annoying and less intuitive to replace <references/> with a template than just adding or modifying a parameter to an existing template. Plus the idea of getting rid of {{tl|reflist}} in favour of a tag is a solution looking for a problem, as far as I can tell. Hence I doubt it's going to fly. Gawaon (talk) 13:50, 30 April 2025 (UTC)
• :::::{{Tq|solution looking for a problem}} the real problems with the template are listed above. {{template link|reflist}} was created specifically to extend the parser tag and met very similar opposition fifteen years ago, it's worth considering now whether it's still needed in most cases. Dan Leonard (talk • contribs) 14:30, 30 April 2025 (UTC)
• :::::I'm not saying that, I'm saying something like this:

{{thincols}}

Aaron Liu (talk) 15:28, 30 April 2025 (UTC)

• ::::::@Aaron Liu, the Visual Editor wouldn't be able to handle that due to a different limitation. Try editing pages with the VE that use {{tl|refbegin}} & {{tl|refend}} for a list of full citations. Rjjⁱⁱⁱ (talk) 00:34, 2 May 2025 (UTC)
• :::::::There's no under refbegin, though. What I propose is just CSS templatestyles—it doesn't even change the DOM. There's no way it would break VisualEditor unless VE uses computer vision. Aaron Liu (talk) 00:56, 2 May 2025 (UTC)
• ::::::::But to get the CSS to apply, won't you have to wrap the with a div or some other element? Rjjⁱⁱⁱ (talk) 01:35, 2 May 2025 (UTC)
• :::::::::No. It's done through a stylesheet that will apply to all invocations of on the entire page. You're thinking inline styles. Aaron Liu (talk) 01:41, 2 May 2025 (UTC)
• ::::::::::[https://www.mediawiki.org/wiki/Extension:TemplateStyles TemplateStyles] can affect classes like reflist and reference-text? Rjjⁱⁱⁱ (talk) 02:01, 2 May 2025 (UTC)
• :::::::::::Yes. Just put the inline style you mentioned below into a selector for those classes. Demonstrated at User:Aaron Liu/sandbox#References using User:Aaron Liu/Refwidth using Template:TemplateStyles sandbox/Aaron Liu/refwidth-20em.css. Aaron Liu (talk) 11:38, 2 May 2025 (UTC)
• ::::::::::::Wow, this is a really cool development and completely changes the whole proposal. I had no idea TemplateStyles was this powerful. Dan Leonard (talk • contribs) 17:05, 2 May 2025 (UTC)
• ::::::::::@Aaron Liu, I was mistaken; I just assumed they did not work that way. I tested something similar out in {{tl|refbegin/sandbox}}. The first real issue I see is that they both do something to the effect of style="column-width:{{{any value}}}" which I don't think can be done with separate CSS. The first workaround that comes to mind is to create CSS classes for each width and maybe round the values down (so that 22em becomes 20em) to avoid having too massive a number. That still seems kind of wonky though Rjjⁱⁱⁱ (talk) 03:17, 2 May 2025 (UTC)
• ::::::::::@Aaron Liu, okay I did this in a way that feels goofy but seems to work? Check out [https://en.wikipedia.org/w/index.php?title=User:Rjjiii/sandbox7&oldid=1288375254 this sandbox] version of Open-source license in the VisualEditor. You should be able to see, edit, add, remove, and replace all references without any issues. You will not any styling affects unless you save the page. Regardless of changes to {{tl|reflist}}, I may propose some version of this for {{tl|refbegin}}. I think reflist deprecation may get limited traction though, because even without the template, the VE can't add or remove a LDR. Rjjⁱⁱⁱ (talk) 07:21, 2 May 2025 (UTC)
• :::This proposal specifically exempts uses of the template where editors feel a need for specific column widths. Many, if not the majority, of our articles use the default (blank or 30em) parameters. {{xtag|references}} now supports 30em columns as the default, so there's no difference between it and the template's default setting. Dan Leonard (talk • contribs) 14:32, 30 April 2025 (UTC)
• As a purely practical measure, I suggest updating Help:List-defined references first. Also, we can probably get numbers on how commonly the visual editor is used here. The total was over 25,000,000 edits a couple of years ago. WhatamIdoing (talk) 17:17, 30 April 2025 (UTC)
• I would strongly support this -- I use the visual editor and while I don't run into list-reference articles much, I wouldn't mind pasting in rather than {{reflist}} when I make an article. Every RfC gets "solution in search of a problem" comments, some more than others, but there definitely is a problem here to be solved. Mrfoogles (talk) 23:25, 3 May 2025 (UTC)
• :(If you're in the visual editor, then you don't have to paste in that code. It's in the menu under Insert > References list.) WhatamIdoing (talk) 00:24, 4 May 2025 (UTC)
• Hi, I'm Johannes from Wikimedia Deutschland's Technical Wishes team. I noticed this discussion mentioned sub-referencing, the feature we are currently working on. Some thoughts on {{tlx|reflist}} and :
• As far as I know one major reason for creating {{tlx|reflist}} was the desire to render reference lists in multiple columns. Some projects (e.g. itwiki [https://it.wikipedia.org/wiki/Discussioni_template:References#Note_in_colonne]) decided to deprecate their local reflist equivalent when responsive reference lists got introduced to MediaWiki in 2017 [https://www.mediawiki.org/wiki/Contributors_(2015-2021)/Projects/Columns_for_references]. The feature works for both inline reference lists () and list-defined references ( ... ).
• We initially based sub-referencing (our upcoming MediaWiki feature for re-using references with different details) on list-defined references, but moved away from the initial concept because we haven't been able to make it work with templates like reflist in VisualEditor. The new approach for sub-referencing is primarily based on in-line references [https://meta.wikimedia.org/wiki/WMDE_Technical_Wishes/Sub-referencing#How_to_use_Sub-referencing_in_wikitext], to avoid issues for VisualEditor users. Using list-defined references and sub-referencing will still be possible, but leads to the existing issues for VE users when using {{tlx|reflist}}.
• We've investigated the use of templates like reflist on several wikis and documented our findings in :phab:T377043, perhaps it's useful for future discussions.
• : --Johannes Richter (WMDE) (talk) 12:59, 9 May 2025 (UTC)
• :I think it would be better if the references tag would automatically minimise vertical space. If you have an article which is essentially all short citations it's too sparse even on default two-column. Ifly6 (talk) 15:55, 9 May 2025 (UTC)
• :I think one of the problems with LDR is for wikitext editors. Adding a sentence and a source requires two edits: the first to the ==Section== where the content belongs, and the second to the ==References== to add the source. WhatamIdoing (talk) 16:38, 9 May 2025 (UTC)
• ::For wikitext editors, I see it as a tradeoff; having to make two edits, versus having to pick out the references from the running text. For me, making the two edits is easier. Jc3s5h (talk) 17:20, 9 May 2025 (UTC)
• ::I exclusively edit wikitext and find LDR particularly suited to the task. LDR massively reduces clutter in the editing window and allows me to just work on the prose. The Visual Editor masks the content of {{xtag|ref}} tags regardless of the referencing style, but when I use wikitext and LDR I don't have to deal with templates and reference text filling up the window. Compare the wikitext of Sheetz–Wawa rivalry with Pope Leo XIV, for instance. I can easily discern the prose in the former but the latter is a complete mess. Having to make a second edit is trivial compared to the massive readability benefit. It's also analogous to writing scholarly papers: I'd add a sentence to my paper and then separately go into my reference management software and add the reference. Dan Leonard (talk • contribs) 17:58, 9 May 2025 (UTC)
• ::Agree with Leonard. Only reason I don't use it is I'm too lazy. It's not like it's super-hard to just edit the entire article or make the edit to /* References */ first. Aaron Liu (talk) 21:00, 9 May 2025 (UTC)

{{discussion bottom}}

There has been insufficient discussion for a change of this magnitude in my opinion. To be more specific the question answered in the close was not the question asked, therefore people did not address that question. This change would require a full RFC at the very least. All the best: Rich Farmbrough 13:57, 4 June 2025 (UTC).

:Rich, do you mean that there was insufficient discussion to prefer "" over "{{reflist|refs=" (the consensus that was actually reached in the discussion), or do you mean that there was insufficient discussion to prefer normal/non-WP:LDR over occasionally using LDR (a possibility mooted early in the discussion that proved to be overkill, and so was abandoned in favor of a much smaller change that met the goal)? WhatamIdoing (talk) 21:10, 4 June 2025 (UTC)

: The solution that ultimately found consensus was proposed early on under the title "New proposal: deprecate {{reflist|refs= in favor of ?" and it was discussed extensively, so this outcome seems fine and fair to me. Gawaon (talk) 07:28, 5 June 2025 (UTC)

Really? Maybe long, long, ago, but isn't now the consensus that citation templates use is best practice? Semi-random ping to @SandyGeorgia - are modern FAs allowed to have no citation templates? _{Piotrus at Hanyang| reply here} 05:53, 1 April 2025 (UTC)

:I prefer citation templates, and I don't know if the requirements at FA are different, but text based references are still somewhat commonly used. -- LCU ActivelyDisinterested «@» °∆t° 08:31, 1 April 2025 (UTC)

: The lack of a uniform citation style is the most glaring style problem in Wikipedia. What we permitted years ago to encourage the expansion of the encyclopedia under the banner of "everyone can edit" now makes us look embarrassingly amateur. We should decide on a preferred style and make plans, with the help of intelligent bots, for adopting it universally in the long term. Zero^talk 10:27, 1 April 2025 (UTC)

:I have used citation templates exclusively for years, so yeah, so I will support anything that moves WP to more use of templates. A plan to gradually adopt templates as the standard for citatons is more likely to reach consensus. Donald Albury 14:29, 1 April 2025 (UTC)

: I would also support a default for switching to CS1 citation templates with default short citation formas in {{t|sfn}} – imo {{t|rp}} is just bad – with explicit proviso that custom anchors are permitted. Custom anchors are needed to deal with sources that don't have years (eg {{tq|Suetonius, Augustus}}) as is common in classical studies. They can similarly can be used if an article benefits from shortened anchors (eg {{tq|CAH² 9}}) or general short cites by title.

:One of the huge benefits of the {{t|sfn}} "ecosystem" is the ability to produce full listings of missing anchors and sources. You simply can't do this with the text-based anchors. A text version of the citation {{tq|Smith 2000, § 3.14}} with no corresponding bibliographic entry for {{!tq|Smith 2000}} is nonsense and we really need ways to track this automatically. Then we can actually go and [https://en.wikipedia.org/wiki/Wikipedia_talk:WikiProject_Classical_Greece_and_Rome/Archive_45#List_of_your_articles_that_are_in_Category:Harv_and_Sfn_no-target_errors,_2025 solve those problems]. {{small|Though, for some certain self-contained corpuses of citations this can be unnecessary. Eg {{tq|Plutarch, Marius}} is evident by convention. A tag here is useful mostly for people who don't know that convention and I usually try to provide it for such sources cited more than once or if translated.}}Ifly6 (talk) 15:17, 1 April 2025 (UTC)

:We should consider the new parameter, details, being developed for the <ref> tag, currently under development. See [https://en.wikipedia.beta.wmflabs.org/wiki/Draft:Moderator_(town_official) my example] on the Beta-Cluster. Also see m:Talk:WMDE Technical Wishes/Sub-referencing#Request for feedback. If successful, this could eliminate the need for {{tl|sfn}} and its ilk. (For the Beta Cluster you might have to sign up for an account. Also, it isn't always working.) I am not a developer; maybe one of these days they will officially designate me as a pest. Jc3s5h (talk) 15:46, 1 April 2025 (UTC)

::That seems a bit restrictive in terms of narrative footnotes which say something like {{tq|Gruen 1995, p. 123, however notes that Dio, 3.14.15, contradicts the narrative in Suetonius, Julius, 1.2.3}} or {{tq|But see Woodman 2021 for alternative views on blah blah}}. While {{t|sfn}} is somewhat inherently restrictive, ref + {{t|harvnb}} essentially solves. One of the developers notes that those notes automatically merge, though, which is a must-have. I'm also not a huge fan of the anchors; imo anchors should match display text. Ifly6 (talk) 15:53, 1 April 2025 (UTC)

:::I'm afraid a talk page is not the right medium for Ifly6 to convey their point. Jc3s5h (talk) 15:57, 1 April 2025 (UTC)

:I would strongly oppose any push to require citation templates, because (1) the citation templates more and more over the years have been pushed into a rigid format that makes it very difficult for human editors to edit by hand and get right, (2) this rigid format makes it frequent that what you want to cite does not fit into that format and should not be distorted to make it fit, and (3) we have bots running rampant over our articles repeatedly massaging templated citations into what they think is the corrected version of the same citation, but the bots often misunderstand citations (especially when the citation is to a review of another citation or to a reprint of another citation) and formatting a difficult citation manually can be a deliberate defense against those bots. —David Eppstein (talk) 16:23, 1 April 2025 (UTC)

::Can you show an example of this for illustration? Ifly6 (talk) 16:36, 1 April 2025 (UTC)

:::Just trawl through the history of User talk:Citation bot and you will find many errors of these types. Often they get fixed, meaning that the exact circumstances that caused this behavior will not immediately trigger the same error. This does not fix the general issue. (The same issue extends to gnomes as well as bots; I had to today revert a gnome who tried to insert repeated fake titles on a collection of book reviews that had no title and were properly formatted using citation templates using title=none, presumably because that parameter value lists the article in CS1 maint: untitled periodical. Manually formatting the book reviews would have avoided that problem and in part because of that I have been manually formatting book reviews more often recently.) —David Eppstein (talk) 23:29, 1 April 2025 (UTC)

::::Making a {{tl|cite book review}} would help bots and people distinguish between an author incorrectly being put in the title vs. the name of the author being reviewed being part of a correct title. Using a template like that would make it easier for downstream machine consumers (like sites that aggregate references to a work or an author across many sources) to parse these weird cases as well.

::::I expect most people prefer to use HTML forms or wizards to make citations rather than raw wikitext, unlike us long-time editors. It is difficult to implement that without machine-writable templates. If templates don't support pretty much the full universe of cases, then we're discouraging a lot of editors from properly citing their work, so we should make an effort to flesh them out. Personally, I find it's a big pain to remember what punctuation to use where; it's much easier to use templates that tidy up after me. It also would be soooo much easier to change the output later across millions of pages if consensus changes about the punctuation and formatting.

:::: -- Beland (talk) 09:43, 30 April 2025 (UTC)

:The current status is one may follow a printed style manual (like The Chicago Manual of Style (CMOS), and other editors will respect that choice. (Actually respect, not just tolerate). If an editor makes up a style just for a certain article, theoretically it's allowable, but in practice other editors can't follow it because they can't read the original editor's mind. The nearest thing we have to a style manual for citation templates is Help:Citation Style 1, but it has problems.

:#It doesn't purport to be complete. On many points it defers to the style in a particular article, such as sentence case or title case for titles of works cited, giving full first names for authors or just initials, etc.. It is only 25 pages long when exported to PDF, compared to 177 pages for the relevant chapters in CMOS 18th ed.

:#There is no policy that the implementation of the citation templates follow the documentation. If a graduate student at a US university submitted a paper that was required to follow a published style manual, but the citation software used by the student flagrantly deviated from the manual, the student would fail the course. In Wikipedia, some comments would be put on some talk pages and nothing would happen.

:#It is absolutely fundamental that a reliable source should never be disqualified because there isn't a citation template to support it. Hand-written citations must always be allowed in this case. But there is no manual to follow when writing such a citation.

:#Since 2020 parenthetical referencing has been deprecated on Wikipedia. As a result, the only acceptable remaining style is endnotes. Respectable published style guides that recommend endnotes or footnotes separate citation elements with the comma, as in "James II of England". But most Wikipedia articles separate them with periods, as in "Nato phonetic alphabet". This should be fixed.

:Jc3s5h (talk) 16:27, 1 April 2025 (UTC)

::If we want to change the "neither encouraged nor discouraged", then we probably need an RFC.

::I suggest keeping it simple and focused. For example, despite what @Ifly6 says, mentioning {{tl|sfn}} will provoke opposition (because it is not used in ~98% of articles and is not wanted in subject areas that rely primarily on short articles instead of books/sources that need to give specific page numbers), and it is largely irrelevant, so it shouldn't be mentioned.

::The simplest is probably to use the "change X to Y" format. For example:

::* Should we change "neither encouraged nor discouraged" to "gently encouraged but not required"?

::* Should we change "neither encouraged nor discouraged" to "preferred but not mandated"?

::* Should we change "neither encouraged nor discouraged" to "by far the most popular choice, but not required"?

::I have, in other areas (e.g., MOS:APPENDIX), had good success with declaring a given option to be "popular" rather than "preferred". Editors tend to choose the popular/normal/usual approach. WhatamIdoing (talk) 17:00, 1 April 2025 (UTC)

:::When wording the RFC, keep in mind that if it succeeds, some editors will try to interpret the new wording as license to change articles to citation templates without seeking consensus, just as one may now change an article from parenthetical referencing to endnotes without seeking consensus. Jc3s5h (talk) 17:14, 1 April 2025 (UTC)

::::Perhaps the entire WP:TEMPLATEREFS sentence (or even the whole/short paragraph) should be in the RFC. The specific sentence currently says: {{xt|The use of citation templates is neither encouraged nor discouraged: an article should not be switched between templated and non-templated citations without good reason and consensus – see "Variation in citation methods", above.}}

::::That could be changed to something like "The use of citation templates is popular, though not required. However, an article that predominantly uses a non-templated style should not be switched without prior discussion – see "Variation in citation methods", above" (example text only; write whatever you think would be helpful). WhatamIdoing (talk) 17:39, 1 April 2025 (UTC)

::(1): I don't see much ambiguity about when to use sentence vs. title case; the CS1 page has guidance for which fields use which. For the "first initial vs. first name" question, it seems to me we should always put the full name, unless only the initial is available, for disambiguation purposes - especially given that Wikipedia citations have machine consumers that correlate authors. Are there only a few remaining questions we could easily answer? Or would we want to pick the third-party style guide closest to general Wikipedia practice as a default? Or provide a short list of third-party guides and let articles pick one?

::(2): Isn't it common sense that if a template does not match its documentation (or the MOS), one or the other should be changed? Making that common sense into a policy wouldn't magically summon volunteer labor to do the implementation work.

::(3): If we decided to go full-template, presumably if there are situations not covered we'd add parameters or additional templates. Situations not covered in the meantime would simply remain non-compliant. We could, if we wanted, designate a third-party style guide as a default or allow an article to choose from a short list of popular third-party styles.

::(4): I think what you are describing is Citation Style 2? I would support merging these two styles so that there is more site-wide consistency, but I have no opinion as to the most "respectable" punctuation. -- Beland (talk) 09:25, 30 April 2025 (UTC)

:FAs are allowed to have any consistent citation style, whether produced by templates or handwritten. Nikkimaria (talk) 00:34, 2 April 2025 (UTC)

::Pinged here ... agree with Nikkimaria. I don't see a need for any change; not broken, doesn't need fixing. SandyGeorgia (Talk) 01:32, 2 April 2025 (UTC)

:I would support changing guidance to say templates are "preferred" and letting people change articles to use them without any further discussion. This makes formatting more consistent because there's less room to be sloppy, automates the process of finding some incomplete or bogus citations, makes it possible to write user-friendly GUI tools that hide the raw wikitext, significantly simplifies the parsing downstream consumers have to do thanks to COinS (e.g. citation aggregation sites, author profile builders, archive.org). Clarifying badly-formatted citations will probably help with the enormous task of fact-checking all our content. This process will probably also shake out some citation styles that should not be used on Wikipedia because they are so radically different from what is done on the rest of the site. And maybe one or two we want to keep but give them their own templates. -- Beland (talk) 10:01, 30 April 2025 (UTC)

::Does the community actually want to be "letting people change articles to use them without any further discussion"? WhatamIdoing (talk) 18:32, 30 April 2025 (UTC)

:::I would support doing an RFC to find out. I often do so for one or a handful of citations at a time, and I don't remember anyone reverting that on the grounds the article doesn't use a template-compatible citation style. (I do remember some confusion about how to cite web pages that are only accessible from archive.org.) Often I'm switching to templates because they handle square brackets in titles without awkward escaping. -- Beland (talk) 19:26, 30 April 2025 (UTC)

::::I suggested a couple of possible alternative wordings above. Do any of those appeal to you? WhatamIdoing (talk) 19:30, 30 April 2025 (UTC)

:::::I would go for the full throated version - change "neither encouraged nor discouraged" to "preferred", not "preferred but not mandated" or the other suggestions which seem to leave a lot of wiggle room for arguments to break out. -- Beland (talk) 19:42, 30 April 2025 (UTC)

::::::I realized another benefit of going full template - just as we have the ability to set "mode=cs2" once for an entire article, we could add "mode=chicago" or "mode=mla" or whatever alternative styles the community can't bear to part with. This would let us change styles for an article very easily (except perhaps if downcasing is needed?) if consensus changes about which articles need which style, and it would also strongly enforce per-article consistency without forcing any particular citation style. I like the idea of having a short list of approved styles, because readers encountering a very rare citation style are likely to be confused or maybe assume it's the result of sloppiness. The fewer citation modes the better in my opinion, but this might be a compromise of the sort you're looking for in order to widen support. -- Beland (talk) 19:48, 30 April 2025 (UTC)

:::::::All I can say is that if we go “full template”, I will no longer be able to contribute to Wikipedia. Perhaps I am too old. I still think in terms of sentences and paragraphs, not data parameters and fields. I never learned how to use templates, and don’t really have any interest in learning now. I still format citations by hand. I am fine with others following along after me and inputting my citations into a template, but I’m never going to create a new citation using one. Oh well… time to catch the early bird special at the Golden Corral and then go watch Matlock. Blueboar (talk) 20:25, 30 April 2025 (UTC)

::::::::{{ping|Blueboar}}{{tq|All I can say is that if we go “full template”, I will no longer be able to contribute to Wikipedia.}} You appear to be assuming that you'd have to type these templates out by hand, which has never been true. Polygnotus (talk) 01:44, 2 May 2025 (UTC)

::::::::There's no reason for you to quit Wikipedia; even if the MOS says templates are preferred, I (and I hope all other editors) will be happy to accept your hand-formatted citations, and leave converting them to templates to a wikignome. -- Beland (talk) 20:55, 30 April 2025 (UTC)

:::::::::I've inquired about the prevalence of citation templates at Wikipedia:Village pump (technical)#Prevalence of citation templates, and it looks like ~80% of articles use the main citation templates.

:::::::::I hope that if the community decided to officially "prefer" citation templates, they would also choose to reiterate the main behavioral goal: {{xt|While you should try to write citations correctly, what matters most is that you provide enough information to identify the source. Others will improve the formatting if needed.}} Or, to put it more simply, do your best. Nobody should get hassled about how they format a citation so long as (a) we have enough information to identify the source and (b) they don't revert if someone comes along after to "fix" it. WhatamIdoing (talk) 23:20, 30 April 2025 (UTC)

::::::::::If the community decides to officially "prefer" citation templates, it should only do so at a time when the citation templates are capable of fully formatting all citations. That time is not now; the citation templates are too inflexible, and too prone to raising errors in common use cases (such as that we wish to cite the original publication of a book source but include the isbn of a reprinted copy of the same book). —David Eppstein (talk) 00:17, 1 May 2025 (UTC)

:::::::::::Isn't that already possible? There's no validation of ISBNs, but you could always do {{cite book |title=Original book |year=1901}} etc. for the original copy, followed by a separate {{ISBN|1234567890}} if you wanted to keep them apart. WhatamIdoing (talk) 00:39, 1 May 2025 (UTC)

::::::::@Blueboar Few thoughts (speaking as someone who remembers time before the Internet, too...). First, I'd expect that "old geezers" from my and older generations are familiar with forms to be filled in. Templates are just that. I don't think filling in a form takes longer than writing a citation by hand. (I am assuming, of course, the use of VE or such, doing this by typing the code is painful, please don't). Second. Templates enable various uses of metadata. They make citations better just like hyperlinks makes text better, or computers enable Wikipedia. They are a step in the right direction. That third, my third point - frankly, conversion of citations from free flowing whatever written format into templates is something that AIs should be able to handle. I don't know when we will have a bot or gadget for that, but just run ChatGPT or such in a window where you run a task telling it to turn it into Wikipedia citation template code, and voila, you should get a well formatted code to paste back into wiki in a second. So, errr, there's no need to leave or such. Learning how to use the better system (and yes, because of metadata, it is strictly better, no ifs and buts) in this case is not hard - just fill in a simple form, or have AI give you a code. Look, I understand the issues (annoyances) of unfriendly new interfaces well, but in this case, it's easy to move from old, inferior output to the new, superior one. Really. Try it. _{Piotrus at Hanyang| reply here} 04:47, 2 May 2025 (UTC)

:::::::::Forms work great for a database… not an encyclopedia. Blueboar (talk) 12:13, 2 May 2025 (UTC)

::::::::::Huh? They work perfectly fine for me and all others who use VE, or tools like TWINKLE... _{Piotrus at Hanyang| reply here} 13:25, 5 May 2025 (UTC)

:::::::There have been similar entertainings regarding other styles in the CS1 module (MLA and Vancouver particularly), and they've either just gotten nowhere or I suspect more commonly were not friendly to integrate with the current structure of the module set and so were given up on. Were something like this to be done, I suppose it would be possible to place them in their own modules and then call those only when a certain parameter is provided to the CS1 module, but even today there are some checks that CS1 makes very early in the execution of the module which may be inapplicable in other older/recognized citation styles. So you might as well start your own module. Module:Cite LSA used to exist as one attempt at this, and there are a few Bluebook style citation templates that have a bare minimum of centralization. For what code sharing might be possible because arbitrary style does ask for a review, I've mused before on the CS1 help talk page, but I suspect those have gone nowhere for time and little or no known potential users. (For example, the ID and access date checking that CS1 does. Of course, then we're imposing some burden both on CS1 and external users of CS1, primarily at our sister and sister language wikis.) Izno (talk) 23:04, 30 April 2025 (UTC)

{{ping|David Eppstein}} you mentioned above that the current set of citation templates are not ready to be preferred because not all works can be cited with these templates. It seems to me they're really not ready for use at all, because at any time a need to add a new citation to an existing article that already has a long list of citations, but no existing template is suitable for the work to be added. The problem is all the existing template documentation is focused on which template to use, and how to set the parameters. It's hard to find examples of how citations should look when they are rendered; any such examples are scattered and disorganized in the documentation.

If proper documentation existed, an editor who had to add a citation for something that isn't supported by any existing template could decide which template is the closest fit, and hand-write a template that generally resembles one of the existing templates. Jc3s5h (talk) 01:05, 1 May 2025 (UTC)

:It's a fair point that existing templates don't handle all use cases, but that may be because there is no guideline pushing people to use them in all cases. My expectation if templates are "preferred" would be that unhandled cases would be left hand-written until someone created a template to handle them. Based on this feedback, maybe we need to say that explicitly. How about:

::The use of citation templates is preferred for situations the templates are designed to handle. Templates should be expanded or created to cover the remaining situations that would otherwise need to be manually formatted.

:"Situations" might include the need to support rare citation styles, though I hope this is not the case. I see templates supporting CS1, CS2, Vancouver, Bluebook, and Harvard. Do we know of any articles that consistently use a style that is not one of these?

:I don't see why extensive documentation is needed, though some basic points are helpful. But if you need to create a new template and you want to see how e.g. the CS1 templates render something close to your use case, just plug the relevant parameters into a template and preview it or put a copy in your sandbox. Adding too much documentation increases the risk that the code and the documentation get out of sync, which will not help someone trying to expand the system.

:In any case, I think the existing templates cover 80-90% of what is needed, and I'm sure we have plenty of work converting those to keep us busy while template builders expand support. -- Beland (talk) 02:23, 1 May 2025 (UTC)

::There are plenty of FA-level articles that use nontemplated styles. Many people who do scholarly work off-wiki can comfortably format citations consistently by hand, so I don't know why we would do "plenty of work" to change them. And it's trivially easy to create inconsistently formatted citations using templates. Nikkimaria (talk) 02:47, 1 May 2025 (UTC)

:::Those scholars can continue to contribute hand-formatted citations, and if you don't want to do any work on this, you don't have to. A good reason to change them is that they are not emitting COinS metadata, and thus are slightly less useful to downstream consumers. It's easier for scripts to validate the contents of individual fields than it is to make sure that all the punctuation and italics and everything in a hand-formatted citation is done correctly. I mean, how would a script be able to tell the difference between a chapter in a book and an article in a journal if the formatting can't be trusted because it's what's being checked? Featured articles are about 0.1% of the overall encyclopedia. The fact that they're nice and tidy should be celebrated, but that doesn't obviate the problem of the millions of untidy articles. -- Beland (talk) 08:18, 1 May 2025 (UTC)

::::I personally think that millions of articles looking untidy is a feature, not a bug. What a focus on compliance with the MoS even for poorly written articles that cite unreliable sources does is put a huge amount of precisely defined lipstick on pigs. Unifying the citation style of articles should not be done before checking the actual content of the citations. —Kusma (talk) 08:42, 1 May 2025 (UTC)

:::::I see your point, and I do actually use poor formatting as a proxy to automatically identify articles with dubious content, though that's usually a pile of unreferenced strings. But it would be awkward to try to preserve this potential signal as long as possible by making a rule that wikignomes aren't allowed to clean up spelling, punctuation, citation formatting, etc. without verifying the claims being made in the prose they are tidying and that the sources are cited accurately. Often that happens naturally, and it's easy to catch glaring problems when doing that, but fact-checking takes so much longer than tidying up, it lags by decades. We also don't have a way of checking which passages have already been fact-checked, which would lead to a lot of redundant work. At the very least I do tag prose I've just made from a pile of dubiousness into a clean, grammatical flow as needing citations if it doesn't have any.

:::::The Guild of Copy Editors does actually reject unreferenced passages; these do tend to change a lot when the first sources are added, which is extremely healthy. But after that, as soon as someone has put in enough effort to make plausible footnotes, the text is considered stable enough to deserve tidying.

:::::If I had to guess, I'd say we have greater problems with claims not matching cited sources with mature citations rather than when the citation is first added. People tend to edit article prose without verifying that the new claim is still supported by the footnote at the end of the sentence, and sometimes sentences get combined or split and footnotes wander around. -- Beland (talk) 09:38, 1 May 2025 (UTC)

:Jc3s5h, can you give me an example of a work for which "no existing template is suitable"? WhatamIdoing (talk) 02:58, 1 May 2025 (UTC)

::{{tl|Cite map}} requires a title. Suppose a map doesn't have a title. Style guides typically say to give a description of the map where the title would usually go, but not use quote marks around the description, and not use italics, so readers can tell it's just a description. Jc3s5h (talk) 04:32, 1 May 2025 (UTC)

:::@Trappist the monk, what would you recommend for a CS1 template that doesn't require a title when the work is untitled? WhatamIdoing (talk) 05:24, 1 May 2025 (UTC)

::::Jumping in randomly here...my first thought would be we could add a "no-title-desc" parameter to {{tl|cite map}}? I would also be tempted to put parens instead of quote marks like, (untitled map of Massachusetts Bay Colony) but perhaps this is not common practice in professional citations. I'm also wondering if this has actually come up or if this is speculative? Text works with no title (as used to be common practice) are named by the first few words; see MOS:INCIPIT. -- Beland (talk) 08:27, 1 May 2025 (UTC)

::::

::::cs1|2 journal templates support {{para|title|none}} which suppresses the rendering of the article title. That was intended to be used for en.wiki articles that followed the citation tradition wherein the title of the cited article is not made part of the citation. I suspect that most if not all uses of {{para|title|none}} are not used to maintain that traditional substyle.

::::

::::I once suggested that cs1|2 might support a {{para|description-in-lieu-of-title}} sort of parameter (in need of a better name) that would render an unstyled description in place of {{para|title}}. That suggestion died aborning.

::::—Trappist the monk (talk) 13:30, 1 May 2025 (UTC)

:::::@Trappist the monk, are you sure about that? This: {{cite web |title=none |url=https://www.example.com}} doesn't look like suppressing the rendering of the article title to me. WhatamIdoing (talk) 20:55, 1 May 2025 (UTC)

::::::{{tq|cs1{{!}}2 journal templates support {{para|title|none}}}}

:::::::{{cite journal |title=none |journal=Journal}} → {{cite journal |title=none |journal=Journal}}

::::::—Trappist the monk (talk) 21:27, 1 May 2025 (UTC)

:::::::Perhaps that feature should be extended to {{tl|cite map}} (or generally; there are webpages with no titles, too). WhatamIdoing (talk) 22:06, 1 May 2025 (UTC)

::::::::Journal templates do not support title=none when there is a url present. In general, webpages are going to have urls and urls are going to block the citation templates from supporting title=none even if that support is extended to non-journal templates. —David Eppstein (talk) 22:28, 1 May 2025 (UTC)

:::::::::That makes sense because we need a title for the link.

:::::::::Not putting the title of the article being cited in the citation to the article...sounds crazy when I say it out loud? Is there an article with an example of this? -- Beland (talk) 01:10, 2 May 2025 (UTC)

::::::::::Sometimes there just isn't a title. Consider a sign: Maybe it will have a title, and maybe it won't. A letter is another source that often doesn't have a title. WhatamIdoing (talk) 01:39, 2 May 2025 (UTC)

:::::::::::Sure, but a sign is not an journal article. What I'm scratching my head over is why a citation wouldn't have the title of a journal article when one exists. I feel like I need an example for context. -- Beland (talk) 02:46, 2 May 2025 (UTC)

::::::::::There's plenty of solutions for this, see Help talk:Citation Style 1/Archive 98#Handle title=none with url better.

::::::::::Why those aren't implemented is beyond me. Headbomb {t · c · p · b} 00:53, 7 May 2025 (UTC)

::Another example that I run into all the time is book reviews, which don't usually have titles, or are labeled with things formatted as titles that are not really titles like "Reviews - Euler’s gem, by David S. Richeson. Pp. 336. £16.95. 2008. ISBN 978 0 691 12677 7 (Princeton University Press)". When the review is published in a journal and has only a doi link, then the cite journal template can handle it with title=none, but most other formats of book reviews cannot be handled by the templates without making up a nonexistent and therefore false title. We should not be putting false information into the encyclopedia, not even in references and not even because the template doesn't work without it. And the bots that run around "improving" citations will often get confused by citations to reviews and mix them up with citations to the thing being reviewed or vice versa (an egregiously bad example from today: [https://en.wikipedia.org/w/index.php?title=Fundamental_theorem_of_calculus&diff=1288133688&oldid=1288020772]). To avoid both problems I've taken to frequently formatting references to book reviews manually instead of with the templates. —David Eppstein (talk) 04:53, 1 May 2025 (UTC)

:::It's unclear to me if the example you link to is a problem with the bot or the human operating it? It's also unclear to me what the thing being cited is. Is it a book or an article or a review of a book or ?

:::If a journal publishes a book review just titled "War and Peace by Herman Melville" then I agree it might be confusing and arguably incorrect to put "title=Review of War and Peace by Herman Melville". It seems better to have output like:

::::"[Review of] War and Peace by Herman Melville". Archimedes Syracuse.

:::or

:::: "War and Peace by Herman Melville" (review by Archimedes Syracuse).

:::or whatever the professional style guide specifies for these situations. It might be useful to have separate fields like "reviewed_title" and "reviewed_author" if we need to fabricate strings but make it clear they are not a word-for-word title the reader should be looking up. Or a separate template like {{tl|cite review}} to take the same fields as e.g. {{tl|cite journal}} but produce different output with "review" in there somewhere. -- Beland (talk) 08:50, 1 May 2025 (UTC)

::::The example I link to was a perfectly good and perfectly normal citation to a book. Until Citation bot got to it. Citation bot somehow discovered the existence of a review of that book in the journal Nature and half-converted the citation into a Frankenstein citation half about the book and half about its review.

::::It is useful to cite things that have reviews. It is also, separately, sometimes useful to cite the reviews of those things (for instance in articles about the things being reviewed). Many humans are capable of distinguishing which kind of citation is intended and keeping them distinct from each other. The bots have demonstrated themselves to be incapable of this. This bot misbehavior makes it problematic to have templated citations to reviews because the bots are likely to misinterpret them and break them. —David Eppstein (talk) 16:47, 1 May 2025 (UTC)

:::::Huh. I would expect citations to books to use {{tl|cite book}}. Still unclear to me if there is a human review step that should have caught this? -- Beland (talk) 01:13, 2 May 2025 (UTC)

::::::If you're referring to the fact that the bot citation damage involved a {{tl|citation}} template rather than a {{tl|cite book}} template: that is one of the key differences between Citation Style 1 and Citation Style 2. In Citation Style 1 editors have to figure out which of many different citation templates to use and the automatic tools frequently get it wrong calling them all cite web. In Citation Style 2, everything uses one template, {{tl|citation}}. The other difference is. That Citation Style 1. Has many periods. That break up. The flow. Of the citation. Citation Style 2 uses commas, instead.

::::::I'm surprised you wouldn't know this already. Am I misinterpreting your reply? —David Eppstein (talk) 07:46, 2 May 2025 (UTC)

:::::::Either template can produce either output style with the "mode" parameter if the default output is not desired. I guess I'm just not in the habit of using {{tl|citation}}; it seems a bit more vague, but of course it's not wrong to use it.

:::::::That really wasn't the important part of my comment. Since no one was answering my question, I went ahead and tested the Fundamental theorem of calculus scenario. Citation bot does not give humans a chance to preview its changes before it makes them, it only gives them a link to the diff afterwords. Though in this case, even if I had manually checked the source, it's unclear I would have noticed that it was a review and not the original work. Both the bot and the humans can be confused because the review has all the same metadata as the original work (with the complication that two authors are usually mentioned rather than one). I can't think of a good way to distinguish the two automatically, so humans just need to look out for this. It's possible looking for key phrases on the page (in this case, "review" isn't used, but "Books Received" is) could be used as a trigger to put up a red flag for the human user. This isn't 100% reliable because e.g. "reviews" would also show up on literature review articles. It's also possible the review is in fact what is being cited, so it's not great to use as an automatic exclusion. For now, I have added a note to User:Citation bot flagging this for humans generally.

:::::::The point of Citation bot is to provide readers with easier access to sources, gets get quite a bit of use, saves a lot of work, and works well in the vast majority of cases - so I would be reluctant to try to revoke its bot approval. Even this error will bring readers to a review of the source they are looking to read, which has a relatively straightforward recovery since they still have access to all the correct metadata once they realize what has happened.

:::::::There is no need to use hand-formatted citations to prevent the bot from altering a citation. Its documentation shows how to exclude the bot from an entire page or from a single citation known to be problematic (I would prefer the latter for ease of long-term maintenance). -- Beland (talk) 22:04, 2 May 2025 (UTC)

::::::::And then the people who maintain the bot go around removing these exclusions when they think they have fixed the very specific issue that caused the bot to misbehave once and be tagged for exclusion. But the problem is not specific bugs; it is that certain classes of issue require human understanding that the bot lacks. We have just this month had a Citation bot user blocked after an ANI thread because they thought the bot could be run without supervision and were blowing off complaints about the resulting bad edits. The bot is usually useful but occasionally causes problems, and needs checking. —David Eppstein (talk) 18:20, 3 May 2025 (UTC)

:::::::::I've not seen people removing nobot exclusions, but if they do, it could help to put in the exclusion comment what to check after removal. But people could just as easily go around switching hand formatted citations to templates and not know that the reason they were hand coded was bot danger, rather than simply laziness. It seems better to explicitly declare bot incompatibility than lay a trap of a secret workaround. -- Beland (talk) 01:50, 4 May 2025 (UTC)

{{outdent}}

I have started Wikipedia talk:Citing sources#RFC on preferring templates in citations. -- Beland (talk) 23:54, 6 May 2025 (UTC)

To what extent is a press release a reliable source? They are mentioned in passing in WP:RS, but I cannot track any direct comment on them.

In the case I have dealt with, I had excised comment on visitor numbers from Vasa (ship) with [https://en.wikipedia.org/w/index.php?title=Vasa_(ship)&diff=prev&oldid=1286715591]. I have since seen statistics on museum visits in Sweden collated by a government agency ([https://kulturanalys.se/publikation/museer-2023/], table 23 in the spreadsheet). But the simplest "headline figure" that seems to encapsulate the number of visitors to the ship since her salvage is a Vasamuseet press release [https://www.vasamuseet.se/om-vasamuseet/pressrum/vasamuseet-okade-antalet-besok-med-9-procent-ar-2024-och-slog-besoksrekord-i-februari-mars-och-december] giving a figure of 45 million to date. (Added with [https://en.wikipedia.org/w/index.php?title=Vasa_(ship)&diff=prev&oldid=1287366912]) To my mind, this figure from the museum is validated by them having to report these numbers to a government agency. Simple arithmetic from the government agency report makes the 45 million entirely believable. For myself, I find the cited source totally sufficient. Clearly other press releases by other organisations may be different. In my specific example I have chosen not to contextualise the visitor numbers as "Scandinavia's most visited museum" which, I understand, is in their marketing material. I don't see marketing material as an RS, whilst a press release may well be.

I am wondering to what extent my decision-making is supportable by guidance on RSs. ThoughtIdRetired _TIR 19:57, 25 April 2025 (UTC)

:A museum's press release wouldn't confer notability under WP:SIRS, but for confirming uncontroversial statistics like visitor numbers, there's no real reason to expect or need an independent source. The relevant section is WP:SELFSOURCE, where organizations' statements about themselves are acceptable in some cases. Your citation seems perfectly fine (although if you went with "most-visited museum" you might run afoul of the rule against {{!tq|unduly self-serving}}); after all, we have {{template link|cite press release}} for this reason. Dan Leonard (talk • contribs) 19:17, 28 April 2025 (UTC)

::Press releases are always Wikipedia:Self-published sources. They are almost always Wikipedia:Primary sources. They are usually not Wikipedia:Independent sources.

::But: That doesn't mean they're WP:NOTGOODSOURCE. WhatamIdoing (talk) 19:29, 28 April 2025 (UTC)

I used the Template:Cite web for a [https://tsdr.uspto.gov/#caseNumber=77072468&caseSearchType=US_APPLICATION&caseType=DEFAULT&searchType=statusSearch webpage on the US Patent and Trademark Office's TSDR], and I was wondering whether there is a special way to cite a US trademark registration in an article. I saw that there's a template for a patent (Template:Cite patent) and was curious if there was something similar for a trademark as well. Appreciate any guidance. (Guyinblack25 ^talk 03:04, 29 April 2025 (UTC))

I've always "known" that direction quotations must be cited immediately after the quote, even if an end of paragraph citation would normally cover it. This leads me to write paragraphs like:

{{tq|In 1916, Abramson designed the Home of the Daughters of Jacob on 167th Street between Findlay and Teller Avenues in the Bronx. The building consists of eight wings arranged radially around a central core, and has been described as "novel in design, being in the form of a wheel".{{Cite news |date=October 30, 1916 |title=Lay Stone for New Home |url=https://www.nytimes.com/1916/10/30/archives/lay-stone-for-new-home-thousands-attend-ceremonies-of-the-daughters.html |access-date=November 11, 2024 |work=New York Times |pages=8}} The property consists of 36 lots which were previously part of Gouverneur Morris's estate; at the time of purchase by the Daughters of Jacob, it was still occupied by Morris's 1812 house which was torn down to make room for the new building.}}

where I put a citation directly after the "novel in design, being in the form of a wheel" quote, even though the exact same reference appears at the end of the paragraph. This has always seemed silly to me.

Looking at WP:INTEXT, I see it says {{tq|In-text attribution may need to be used with direct speech ... An inline citation should follow the attribution, usually at the end of the sentence or paragraph in question}} which sure sounds to me like the extra citation immediately after the quote is not actually needed. Am I just mis-reading this? Can I condense duplicate citations like this into a single one at the end of the paragraph?

{{reflist-talk}} RoySmith (talk) 22:19, 3 May 2025 (UTC)

:Maybe I've been doing it wrong? But I put the cite at the end of the content it supports, even if there's a direct quote in there. Schazjmd (talk) 22:25, 3 May 2025 (UTC)

{{User:ClueBot III/DoNotArchiveUntil|1749600080}}

Should the text of Wikipedia:Citing sources be changed to prefer templates over hand-formatted citations, while welcoming contributions from editors who continue to format manually? 23:53, 6 May 2025 (UTC)

Specific changes proposed:

1. Change "(Note that templates should not be added without consensus to an article that already uses a consistent referencing style.)" to "(Templates are preferred, but contributions with manually-formatted citations are welcome.)"
2. Add "change citations from manually formatted to templates, without admonishing of contributors" to "Generally considered helpful"
3. Change the line starting "adding citation templates to an article that already uses a consistent system without templates" in "To be avoided" to "removing citation templates that are used correctly"
4. Change the first paragraph of WP:CITECONSENSUS to: "Citation templates are preferred in situations where they exist and can be used as designed. They keep citations formatted in a consistent way and are more machine-readable for a variety of purposes. Contributions of manually-formatted citations from editors unfamiliar with or who simply do not care to use templates are welcome, and may be reformatted into templates by other editors without notification beyond a polite edit summary."
5. Add to "Generally considered helpful" the line: "Adding or enhancing templates and modules for recurring situations where citations would be otherwise left manually-formatted due to lack of support"

• Proposer rationale: This proposal would still allow any citation format to be used consistently throughout an article, but would allow interested editors to move from hand-formatted to template-formatted citations for the following reasons:
• Much more consistently formatted output, tolerating variation in human-written input, resulting in a more professional and trustworthy appearance for articles.
• Automatic output of COinS metadata for browser plugins and web spiders that power data aggregators.
• Automatic detection of errors, such as dangling references, incomplete or vague citations, putting the wrong information in the wrong place, or using disfavored formats (such as for dates).
• Automatic improvement by bots (e.g. adding archive URLs, adding missing data and links to full text).
• Much easier to make future changes site-wide to formatting if consensus changes.
• Much easier to change an article's citation format (if consensus finds the wrong one was chosen) simply by substituting templates or (with module support) simply adding a "mode" declaration to the page. This also makes it easier to move citations between articles that have different citation formats. (We can already set "mode=cs1" or "mode=cs2".)
• Inexperienced editors (or those who simply prefer them) can use graphical tools like VisualEditor to add and edit citations without having to know wiki syntax or the formatting details of the specific citation style used by an article. Editors who use the source editor will still be under no obligation to use templates in new citations if they dislike them.
• Manual formatting of citations should not be used as a workaround to avoid mangling by a bot. An explicit bot exclusion is a better way to handle this because it alerts future editors to the bug and prevents them from stumbling into it again. This also facilitates research into bot improvements.

:According to the previous discussion, about 80% of articles already use citation templates, so the result of this guideline change is not much different than simply implementing our existing rule that articles should use a consistent format. Upgrading citations also provides an opportunity to eyeball neglected articles and passages and remove any obvious garbage. -- Beland (talk) 23:53, 6 May 2025 (UTC)

::*Oppose major change to WP:CITEVAR allowing the imposition of a new citation style at the whim of gnomes and encouraging gnomes to perform this imposition, regardless of whether the citations are already in a consistent format. Inconsistently formatted manual citations do not need this change; a consistent style can be chosen for them regardless. The only actual point of these changes is to impose machine-friendly human-unfriendly rigid templated metadata on citations.

::—David Eppstein (talk) 00:42, 7 May 2025 (UTC)

::*Oppose addressing rationale: #1, #5 and #6 are disingenuous, don't care about #2, #3 and #4 are hopelessly naive, #7 is no different from present, and #8 is nonsensical. ~~ AirshipJungleman29 (talk) 01:12, 7 May 2025 (UTC)

• Oppose. In addition to the points made by David Eppstein and Airship, allowing manually formatted citations makes it much easier to copy them from external sources, and much easier to incorporate subject-matter experts into the community. Templates don't magically make citations look professional nor address vague or misplaced citations. And forbidding manual formatting of citations as a workaround to known problems is a non-starter. See also the thousands of words here explaining some of the issues with the concept of this proposal. Nikkimaria (talk) 01:20, 7 May 2025 (UTC)
• :Editors would still be allowed to copy already-formatted citations from other web sites. What problems are you seeing that require manual formatting as a workaround? -- Beland (talk) 03:23, 7 May 2025 (UTC)
• ::Saying that others can clean up after them is not a good solution to something that's not a problem to begin with. As to workarounds, there seem to be several examples of issues provided in the discussion above. Nikkimaria (talk) 03:54, 7 May 2025 (UTC)
• :::I looked into the situations mentioned in the previous discussion, and it seems they can all be dealt with by putting an HTML comment in the citation template that instructs the problematic bot not to edit it. That seems better than laying a trap for someone who later comes along and changes the citation to use a template for whatever reason, and it gets tread on by the bot again. -- Beland (talk) 04:14, 7 May 2025 (UTC)
• Strongly oppose – allowing style variations at authors' discretion and leaving decision to local consensus is one of the best ways Wikipedia avoids pointless conflict and churn. There's not much benefit for every page to be perfectly consistent in every aspect of style, and the potential harms of changing this are dramatic. Aside: every editor should at least read and consider {{slink|User:Jorge_Stolfi#Please_do_not_use_{{(}}{cite{{)}}}_templates}}.–jacobolus (t) 01:53, 7 May 2025 (UTC)
• :If you want to render citations e.g. "issue 10" instead of "(10)", I agree that would be an improvement. It could be done across millions of citations in a single edit because they use templates and not manual formatting. We could also allow articles to easily choose to do this or not by adding a mode selector. -- Beland (talk) 03:21, 7 May 2025 (UTC)
• ::Implementing a mode selector to "easily choose to do this or not" would require people to (a) notice a change in the template's behaviour, (b) figure out how to create the option to override it in the template, (c) get that template change implemented, and then (d) change the article. That's not a particularly easy process compared to just formatting refs manually. Nikkimaria (talk) 03:54, 7 May 2025 (UTC)
• :::It's much easier than going through every single manually formatted citation and manually re-formatting them. -- Beland (talk) 04:09, 7 May 2025 (UTC)
• ::::No, it's really not. First off, if someone has already manually formatted a citation in a way that they feel is appropriate, no reformatting is required regardless of what changes in the template. If there is a desire to change the format, either way you'd have to go through every single citation - but the template approach adds a bunch of extra work to do that, because first you'd have to change the template and then change the citations in the article. Nikkimaria (talk) 05:00, 7 May 2025 (UTC)
• :::::Sure, I guess there are some scenarios where there is more upfront cost. If an article is already consistently formatted manually and wants to change to a different format, it will need to have all the citations re-formatted either way, and if it's done the template way and that style is not already supported, template upgrades will also have to happen. But compared to the number of articles (millions) the number of citation styles is quite small (less than ten?) so templates will seldom need to be upgraded to support new styles. The benefit of that investment in this scenario is only realized if there is a second style change where the entire page can be flipped with a mode setting.
• :::::Maybe that happens with mature articles scoring high assessment grades, but I work on a lot of articles with detected typos, and I often see a mix of clashing citation styles on the same page. For those, most of the citations are going to have to be reformatted regardless of the chosen destination style. My thought is, why not make that destination style a template, so that we never have to do another mass-reformat no matter what changes about the preferred citation style? Also, a common way to fix poorly formatted citations is to use a script like reFill, which outputs templates. If we chose a non-template style, we'd have to do a lot of work to make up for the lack of automation, and then even more work after that to make up for the lack of automation finding archival URLs. -- Beland (talk) 07:10, 7 May 2025 (UTC)
• ::::::There are literally thousands of citation styles that exist. I grant you that most are part of a long tail, but that's still a heck of a lot of template changes. Archival URLs are already automatically added on articles without templates, so that's not a benefit of mandating templates.
• ::::::If you're encountering an article with clashing citation styles, you're already allowed to make that consistent, using templates if that's what you prefer to do. Your proposal doesn't change anything about that use-case; it instead targets articles that already have a consistent style that just happens to not be template-based.Nikkimaria (talk) 22:04, 7 May 2025 (UTC)
• :::::::It does prevent people from changing inconsistent articles to manually formatted citations, which is work that would have to be undone later if in the long term we are moving in the direction of using templates universally. And if that's the direction we're going, we might as well start on the manually formatted articles, too.
• :::::::I can't imagine people creating thousands of style templates for tiny variations; it would be a lot less work just to use templates for a popular style very close to one's preferred style. That also seems like a crummy experience for readers, looking at a thousand different styles and either having to learn to interpret a bunch of different conventions (extra difficult for those who are not native English speakers) or just being annoyed at what looks like sloppiness.
• :::::::I don't know of any bots that can operate on manual citations to validate date formats, find dangling references, create markup for COinS, fill in missing authors, or connect citations to databases like DOI. -- Beland (talk) 23:34, 7 May 2025 (UTC)
• ::::::::But we shouldn't be preventing people from improving citations, even if we might sometime eventually want to change how they've done that.
• ::::::::I agree with you re: {{tq|I can't imagine people creating thousands of style templates for tiny variations}} - that's why I don't think your proposal about template upgrades and mode settings is at all workable. What is more likely to happen is (a) users try to shoehorn their preferred formatting into citation templates and get into edit-wars with well-intentioned bots or gnomes (as already happens!), or (b) users manually format citations to get their preferred formatting, and then get into edit-wars as a result of your proposal. As to {{tq|crummy experience for readers}}, I don't see any evidence that CS1 is a better experience for readers than APA, MLA, or any other format you could name - the page jacobulus linked has already suggested some ways in which a non-CS1 format might easier to interpret.
• ::::::::Bots aren't the panacea you suggest - see Jc's comments. Nikkimaria (talk) 00:42, 8 May 2025 (UTC)
• :::::::::If an article has two citations, and one of them is manually formatted and the other uses a citation template (assuming reasonably equivalent contents: they both have the source's title, a URL, etc.), is removing the citation template actually a case of "improving citations"? WhatamIdoing (talk) 00:52, 8 May 2025 (UTC)
• ::::::::::All else being equal, as much as adding one. Nikkimaria (talk) 02:00, 8 May 2025 (UTC)
• :::::::::::Maybe not?
• :::::::::::Because I think that eventually – maybe after I die, but some day in the future – citation templates will grow from "merely" 80% of articles having some to basically all articles using them for everything, and that means that the process will be:
• :::::::::::* Start with a 50–50 mix.
• :::::::::::* Switch to templates.
• :::::::::::* Done.
• :::::::::::vs
• :::::::::::* Start with a 50–50 mix.
• :::::::::::* Switch to manual formatting.
• :::::::::::* Eventually switch back to templates.
• :::::::::::* Finally done.
• :::::::::::And therefore I think that putting a finger very lightly on the scale in favor of citation templates will save time, net, in the end. WhatamIdoing (talk) 02:27, 8 May 2025 (UTC)
• :::::::::::All else is not equal; templated citations are more useful for browser plugins, data aggregation, and more easily changed en masse. Which is why this proposal would define one direction as an improvement.
• :::::::::::I think it would be reasonable to call a truce among say, the top 5 or so most popular citation styles and support those with templates, for compatibility with various academic fields and major published style guides and what people learned to use at school. It's reasonable to ask people to pick one of those and not to force our readers to learn citation style #534 which they came up with one day while filling a complaint with the local dog catcher. There are hundreds of style rules; generally the way they reduce edit wars is by providing unambiguous answers to arbitrary questions. -- Beland (talk) 02:50, 8 May 2025 (UTC)
• ::That sounds very complicated and controversial, likely to waste vast amounts of effort and attention on style nitpicks. This discussion itself is already doing significant harm, insofar as the participants might otherwise be making productive content contributions. –jacobolus (t) 05:20, 7 May 2025 (UTC)
• :::I accept your opinion that content is more important than style. As a community of editors, though, it seems we have decided that style is important enough to worry about that we do have a Manual of Style instead of willy-nilly formatting, and we have decided that it should be consensus-driven rather than saving a lot of time by handing it over to a benevolent dictator or style committee. I can't think of a way to reconcile those choices with the idea that we can't have this type of discussion because it burns time we should be spending on content. It's a valid concern, though I'm not sure we're spending "significant" resources on it given how many thousands of edits are being committed while we're having this discussion. Also keep in mind that not everyone who enjoys wikignoming also enjoys working on content. Part of the reason I do a lot of wikignoming fixing spelling and style errors is so that other editors can focus more on their area of expertise and interest and don't need to be distracted fixing small things that I could fix more quickly en masse. And some days I'm too tired or stressed to wrangle a lot of prose and I just need to relax by fixing a bunch of malformed punctuation or unconverted units of measure. -- Beland (talk) 07:31, 7 May 2025 (UTC)
• Oppose per David Eppstein and AirshipJungleman29. We want people to leave citations that support the information they add. That is far more important than putting people off by insisting on using a cumbersome or alien format that they do not understand. And what to do with people who can’t work in templates? Are we going to punish them for adding what may be good content with a good source if they do it the old fashioned way? It’s unworkable and unnecessary. - SchroCat (talk) 03:46, 7 May 2025 (UTC)
• :I agree with your concerns. The language is meant to explicitly to welcome non-template contributions and advise against "punishing" contributors who do that. Specifically: "Contributions of manually-formatted citations from editors unfamiliar with or who simply do not care to use templates are welcome, and may be reformatted into templates by other editors without notification beyond a polite edit summary." and "contributions with manually-formatted citations are welcome". Is there some other language that would make that more clear? -- Beland (talk) 04:12, 7 May 2025 (UTC)
• ::I get that you want this to happen, but there’s no need to bludgeon every comment that is made. - SchroCat (talk) 04:20, 7 May 2025 (UTC)
• :::I'm trying not to bludgeon (and was actually hoping to just ignore this RFC for a few days); I'm answering the questions you asked. It was honestly unclear to me if you had read the proposed text which is supposed to solve exactly the problem you raised and upon thinking it through still came to the same conclusion, or if you were mostly just reacting to the title of the RFC? Maybe I'm missing something; do you think editors will feel that manually formatted contributions will be unwelcome when they read that templates are preferred, even though it also explicitly says those contributions are welcome? Do you think editors won't follow the guideline telling them not to complain about those contributions? Do you think just arriving at a page and seeing all the citations already using templates will put off potential contributors?
• :::I'm fine with this not happening - though I think it would be tidier and easier to maintain, I realize a lot of people have a lot of strong opinions about formatting. {{u|Hanyangprofessor2}}'s questioning of the current guideline generated several comments favoring either encouraging templates or going even further and having a single citation style for all of Wikipedia. Seemed to me like it was time to check in and see if consensus on this has changed, but if it hasn't, there's plenty of work to be done cleaning up citations under the existing style rules. -- Beland (talk) 06:58, 7 May 2025 (UTC)
• ::::You really are bludgeoning this, despite saying you're not. Your name is already appearing far too frequently contradicting those !voting against this So far, I think nine people have commented in this thread and you've made nine comments throughout the thread, all to those who oppose it. Please just let please comment without interference. - SchroCat (talk) 07:48, 7 May 2025 (UTC)
• :::::This is supposed to be not merely a tallying up of opinions, but a discussion with the possibility of improvement and compromise. But if you are uninterested in exploring mitigation of whatever problem you are foreseeing and just want let your opinion stand, that's fine; I will sit in confusion given that we seem to be in agreement and disagreement simultaneously. -- Beland (talk) 08:59, 7 May 2025 (UTC)
• ::::::I am NOT tallying up opinions - that's a complete mischaracterisation of what I have written, which is clear to anyone reading it. I was pointing out that nine people have made comments in this thread (that's tallying contributors, not opinions) and you have now made ten comments. To quote from the guideline: "{{tq|If your comments take up one-third of the total text or you have replied to half the people who disagree with you, you are likely bludgeoning the process}}". - SchroCat (talk) 09:31, 7 May 2025 (UTC)
• @AirshipJungleman29, @David Eppstein, @Jacobolus, @Nikkimaria, @SchroCat: I wonder what you would think about a short, simple factual statement, like "While citation templates are the most popular choice, using them is not actually required". (If anyone's curious, we ran the numbers: about 80% of articles contain at least one CS1|2 citation template.) WhatamIdoing (talk) 05:28, 7 May 2025 (UTC)

:*I think the proposal is not just a solution in search of a problem, but a positive backwards stop that will inhibit editors, particularly the less experienced from adding citations. I am a huge fan and use them in 95 per cent of the material I add, but we should not be creating artificial hurdles that stop people from adding sources. That's only going to damage us. - SchroCat (talk) 07:48, 7 May 2025 (UTC)

:*:Based on the discussion above, I don't think that the OP has any interest in "creating artificial hurdles that stop people from adding sources". I think the goal is more like "if you put [https://www.example.com/source.html|Title of source] in ref tags, then let's have an official rule saying I can quietly turn that into {{tl|cite web}} for you".

:*:Because, in practice, that's what happens. The manually formatted citations are almost never beautifully formatted examples of any style guide, they almost never form a "consistent" citation style, and they regularly do get converted to citation templates. We just sort of pretend in WP:CITEVAR that everything's equal, when it's really not equal. WhatamIdoing (talk) 16:29, 7 May 2025 (UTC)

:*::Unfortunately (as I think you've found elsewhere!) once the language is added it really doesn't matter what the OP's intentions were. Nikkimaria (talk) 00:42, 8 May 2025 (UTC)

:*:::True, and I'm not wild about the proposed language. (For one thing, there's a lot of it.) But given the apparent intention, I think it might be possible to find language that meets the goal without making people think that it's a bot free-for-all. (Yes, that's an odd conclusion for text that doesn't mention bots at all, but I've had discussions in which I've told editors that if we moved one section of text out of a {{tl|policy}} to another page, with no changes to the text and with its own copy of the {{tl|policy}} tag on the new policy page, it would still be a policy, and they still thought it would result in the new policy page not being a policy. That's a plural they, by the way: two editors had difficulty with the concept. Wikipedia is one of the few sites on the internet that really does (and values) close reading, but we don't always pay attention.) WhatamIdoing (talk) 02:24, 8 May 2025 (UTC)

• Oppose after reading through above. I'm a big fan of the templates, they help me input without much thought, and they help me understand what each piece of information is in others' sources. However, at its core, "This page in a nutshell: Cite reliable sources." It's gratifying when an IP editor adds a bare url, and if they want to manually add more information in plain text, this page should takes pains not to discourage this. Past a certain point of relevant information being included, the marginal benefit of encouraging gnoming to change manual sources into template sources seems limited. CMD (talk) 05:31, 7 May 2025 (UTC)
• Tentatively Support. As far as I can tell, this is how we operate by default in WP:MED, and I feel that the consistency of citation formatting is a big part of why medical articles tend to be easy to verify and to conduct further reading on. It also is pretty much immediately apparent when there's an "ugly duck" citation that is almost certainly subpar. I understand that not all of Wikipedia can or should be held to the standards of medicine, but at the same time, I think our pages are broadly a good demonstration of why this proposal has merit. That being said, I would like to hear what @Boghog has to say about this, and may well change my mind depending on what the medical citation master says. Just-a-can-of-beans (talk) 06:37, 7 May 2025 (UTC)
• Oppose. We need to be rigid about verifiability but not about how to write references. See the top of my user page which starts with a large quote symbol and {{tq|Don't worry about formatting references; just get all the information in there.}} Effective editors work in different ways and it is a mistake to try to dictate what they should do when it does not affect the reading of articles. I happen to love reference templates, but the hard task is to teach new editors why references are important and how to find the right kind of sources. Lets focus on that. StarryGrandma (talk) 14:34, 7 May 2025 (UTC)
• Oppose Per Nikkimaria, CMD, and StarryGrandma. Ealdgyth (talk) 14:41, 7 May 2025 (UTC)
• Oppose implementation by bots, weak oppose in general. Previous discussion is full of complaints about what a bad job bots do of creating templates. Jc3s5h (talk) 15:05, 7 May 2025 (UTC)
• Oppose, per most of the comments above, particularly StarryGrandma. Mike Christie (talk - contribs - library) 16:31, 7 May 2025 (UTC)
• Mixed
• Support Encouraging use of templates and adding missing facilities. Use of templates makes changes in style easier.
• Strongly Oppose any wording that encourages mass changes via bots. They break things too often. -- Shmuel (Seymour J.) Metz Username:Chatul (talk) 16:34, 7 May 2025 (UTC)
• :I don't see any wording encouraging people to use bots. Are you referring to "They keep citations formatted in a consistent way and are more machine-readable for a variety of purposes."? Would you like to have that sentence dropped? -- Beland (talk) 22:56, 7 May 2025 (UTC)
• ::It isn't sufficient to not mention bots. The proposal should include language that strongly discourages bots. I'll add more on machine readability below. Jc3s5h (talk) 23:06, 7 May 2025 (UTC)
• :::Isn't the bot approval process the right place to make that sort of decision? Normally that's done on a per-bot basis rather than a blanket rule, to take into account different pros and cons. -- Beland (talk) 23:37, 7 May 2025 (UTC)
• ::::The bot approval group is very pro-bot, and will be thrilled to approve citation-related bots with an error rate that many of us think is far too high. When is the last time a bot author was told "fix every mistake your bot made and do nothing else on Wikipedia until you're finished, on pain of a community band"? The prohibition must be in the guideline. Jc3s5h (talk) 23:52, 7 May 2025 (UTC)
• :::::Are you talking about unsupervised bots or those with many human operators who are supposed to verify the output? I used to operate an unsupervised bot and administrators did not hesitate to block the bot if it made mistakes which I then had to clean up or explain as not mistakes. -- Beland (talk) 02:10, 8 May 2025 (UTC)
• ::::::From what I've seen, users who invoke supposedly supervised bots don't do an effective job of making sure the output is correct. A bot that tries to automatically change an article from free-form citations to citation templates would produce a mass of changes all mixed together, and the human thinking process just isn't good at checking that kind of change. Jc3s5h (talk) 02:16, 8 May 2025 (UTC)
• :::::::Nothing in this proposal suggests writing a bot to change citations from manually-formatted to template. The whole idea is that bots have an easier time after the conversion, because the humans have done the hard part of segmenting strings into appropriate semantic fields. -- Beland (talk) 06:59, 8 May 2025 (UTC)
• ::::::::The current rule of not allowing changes from consistent non-templated to templates serves as a restriction against doing this with bots. So it does encourage the use of bots, even though it doesn't mention bots. Jc3s5h (talk) 09:41, 8 May 2025 (UTC)
• :::::::::I suppose that's true, but only on the ~20% of pages (plus or minus those that use only one or two templates that violate an otherwise consistent style) that don't already use at least some templates. On most pages, bots could theoretically change citations from hand-formatted to template-formatted in order to achieve consistency or get rid of a style that violates the other rules of the MOS (like using all caps). Assuming that bot could get bot approval and be smart enough to actually do that. So it seems like that should already be a huge problem if it was something actually likely to happen. -- Beland (talk) 00:18, 9 May 2025 (UTC)

• Support, the technical benefits are overwhelming. But oppose bots; understanding is still needed to avoid errors. Ifly6 (talk) 16:54, 7 May 2025 (UTC)
• Oppose - a one-size-fits-all diktat is not helpful. As long as citations are clearly and accurately presented to our readers it does not matter a rap whether templates are or are not used. Tim riley talk 17:18, 7 May 2025 (UTC)
• Oppose - what matters is that sufficient information is given to allow readers and editors to know what the source is and to find the information. Enforcing citation templates (which this proposal will effectively mean even if it isn't wording that strongly or meant by the proposers) won't help this and will inevitably break some references as data is unthinkingly rammed into fields just to get it into the template - whether correct or not. In addition forcing some "One True Wikipedia Reference Style" will drive some editors away, because it isn't the style that the editors are used to / is standard in that field, and this is just the sort of annoying little thing that gets some people angry enough to quit.Nigel Ish (talk) 17:54, 7 May 2025 (UTC)
• support - "use whatever is easiest for you, and let someone else worry about reformatting it to get the many benefits of structured syntax" doesshould not be controversial (which isn't to say I'm surprised it is - I've opposed a lot of style standardization proposals in the past myself). I'm just having trouble finding a persuasive objection in the opposes as to why we should not have better archive links, why we should get in the way of the many tools that improve accessibility and verifiability, why we should make it harder for users of visual editor to work on citations, etc. — Rhododendrites ^talk \\ 22:44, 7 May 2025 (UTC)
• "They keep citations formatted in a consistent way and are more machine-readable for a variety of purposes." I've seen too much sentiment that being machine-readable and bot-friendly comes first, and the ability of the editor to write a cite for any source the editor has access to comes second. The book in the editor's hand doesn't have an ISBN? Let bots add an ISBN for a similar book. Source has a publication date that's not supported by templates, such as Michaelmas term, 2001? Issue an error message. The guideline should very strongly discourage changing a manual citation into a template if there isn't a citation template that fully supports the source, and if the editor who wants to make the change can't prove the correctness of the revised citation because the templatephile doesn't have the source in hand. Jc3s5h (talk) 00:03, 8 May 2025 (UTC) (fixed in light of comments)
• :One boldtext vote per person please. FWIW I agree any act of implementing a template must retain all of the information in the citation. I cannot imagine that would be controversial, though. — Rhododendrites ^talk \\ 00:54, 8 May 2025 (UTC)
• :Regardless of templates we cannot violate the MOS. Any instance in violation of MOS:DATE needs to be changed whether we use a template for it or not. PARAKANYAA (talk) 00:58, 8 May 2025 (UTC)
• Support per Rhododendrites, Chatul, Ifly6. The proposal doesn't call for the abolition of handcrafted citations or the use of bots to convert those. Some free-form citations are not well-"crafted" and need improving. -- Michael Bednarek (talk) 00:44, 8 May 2025 (UTC)
• Support. I am surprised people have so many issues with the bots, I have never seen them change anything besides doi/hdl and curly quotes and archive urls. But even then, if the bots are the problem that isn't the fault of the template that's the fault of the bots. Reverting people for having a wrong format in their content additions is already prohibited in the policy, so that wouldn't change. The templates have a lot of benefit and inconsistency is a negative. PARAKANYAA (talk) 00:57, 8 May 2025 (UTC)
• :They do sometimes make a mistake. One example of such a mistake is [https://en.wikipedia.org/w/index.php?title=Fundamental_theorem_of_calculus&diff=1288133688&oldid=1288020772 this edit]. The bot added a URL pointing at a book review (whose title is exactly the same as the book) instead of the book itself. That bot can be triggered by any editor, who is supposed to then check the output (the instructions for that bot say "Editors who activate this bot should carefully check the results to make sure that they are as expected"), but not everyone does, and even if they do, we can't guarantee that they'd notice a problem like this every single time. (The editor who failed to catch it has been blocked.) WhatamIdoing (talk) 02:18, 8 May 2025 (UTC)
• ::Has a similar kind of thing happened with the CS1 templates, not {{tlx|Citation}}? I see why the universal one would have that problem because it thinks all documents are the same thing. PARAKANYAA (talk) 02:25, 8 May 2025 (UTC)
• :::Humans make careless mistakes no matter whether they are editing on full manual or semi automated. Given how many IP edits are vandalism, uncited rumors, heavily biased, ungrammatical, and so on, I would expect editors using semi-automated tools are reverted at a much lower rate than rando humans. -- Beland (talk) 03:07, 8 May 2025 (UTC)
• ::::Parakanyaa, you can test it in a sandbox. Just copy that {{tl|citation}} to a template, switch it to {{tl|cite book}}, and trigger the bot for the sandbox page. If the bot makes the same mistake, you'll know that the problem isn't unique to the CS2 template. WhatamIdoing (talk) 03:14, 8 May 2025 (UTC)
• :::Citation bot mixes up the cs1 templates all the time, converting one to the other, often getting it wrong, and even when it gets it right doing a partial conversion that makes the converted template erroneous. And in many cases, the human editors before the bot also choose the wrong cs1 template Having multiple citation types is just one more thing for humans and the bot to get wrong; I don't think it provides much useful bot guidance. —David Eppstein (talk) 05:28, 8 May 2025 (UTC)
• ::::Humans also mess up citations all the time when they type things in, leaving off critical information, misspelling names, mangling the formatting, and leaving a lot of dangling references as pages get edited.
• ::::Whether semi-automation makes humans more or less accurate and whether any mistakes are worth the massive productivity gains seem like questions for the bot approval process. The answer depends a lot on the bot. If you think {{u|Citation bot}}, for example, is doing more harm than good, then request to have its bot approval partially or completely revoked. Someone could easily write a more conservative bot that makes fewer mistakes for humans to stumble over, but which leaves more work undone. I don't hear anyone complaining that {{u|InternetArchiveBot}}, for example, makes mistakes. I wouldn't want to throw the IABot baby out with the Citation bot bathwater with an indiscriminate rejection of automation.
• ::::I also can't imagine bots operating on manually formatted text are going to make fewer mistakes than bots operating on machine-readable templates. If we mandate supporting non-template citations forever, sooner or later every bot task that currently only looks at template citations is going to be attempted for non-template citations. -- Beland (talk) 06:56, 8 May 2025 (UTC)
• Support per Beland's proposer rationale, Just-a-can-of-beans, Ifly6, PARAKANYAA and others. Changing manually written citations to some template format is an improvement and should not be discouraged. Gawaon (talk) 07:24, 8 May 2025 (UTC)
• Oppose as written. Don't mess with people's citation styles. There are not many articles where this proposal would have a large effect (only those with consistent non-templated citations) but on those articles we should respect the WP:CITEVAR choices of the authors. —Kusma (talk) 09:55, 8 May 2025 (UTC)
• Oppose. Citations should be clear and complete. The template formats are often rigid and unhelpful. MOS:CITEVAR gets it right: If an article uses a clear and consistent format, it is a gigantic waste of time for people to come by and change it to their preferred template and start an argument about whether they even did that correctly. And, as others have noted, templates may discourage some users from contributing at all if they feel that they will have trouble using them. -- Ssilvers (talk) 15:05, 8 May 2025 (UTC)
• Oppose The citation templates are too fiddly and fussy and focussed on formatting. What matters most for the reader is not the format but the legitimacy of the citation. A relevant quotation from the source and a URL to link to it are the best value for verification but the citation templates tend to obscure this with bibliographic clutter. Andrew🐉(talk) 21:21, 8 May 2025 (UTC)
• :I suggest the opposite is true. The formatting of a properly constructed manual citation requires special knowledge; filling in parameters of a template doesn't and results in proper and consistent formatting. Many sources don't have URLs and require other identification means – that's not "bibliographic clutter". Even for online sources, a mere URL is clearly not enough, and quotations are rarely required. -- Michael Bednarek (talk) 01:04, 9 May 2025 (UTC)
• ::The purpose of a citation is purely functional – to verify a stated fact. My impression is that editors such as Michael regard formatting of citations as an end in itself – an elegant craft like calligraphy or concrete poetry. But form should follow function and the current templates don't assist the function. A functional template would contain semantic features such as AI verification and a specification of the claim being made. Anyway, it seems apparent that we disagree and so consensus is lacking. Andrew🐉(talk) 22:04, 9 May 2025 (UTC)
• :::Current templates do perform some validation which manual formatting does not, for example with date formats or if some parts of the author seem to be missing or incorrect, or if a web page is being cited but no URL is given. They highlight these errors in categories other editors can find, to help minimize the number of readers who try to use a broken citation. If we had AI capable of reading a web page and determining whether or not the content there supports the cited sentence, we probably wouldn't need templates because we could just have AI fix all the formatting errors and inconsistencies that humans leave. -- Beland (talk) 04:09, 10 May 2025 (UTC)
• ::::My impression is that current AI technology is just about capable of confirming that a citation supports what an article says. And it might go further to assess whether other, uncited sources are in general agreement with the cited source. Automating the fact-checking of our articles would be better than providing selected citations and expecting each reader to repeat the fact-checking themself. Andrew🐉(talk) 16:23, 10 May 2025 (UTC)
• :::::I'm an AI engineer; that's nowhere near true. -- Beland (talk) 18:00, 10 May 2025 (UTC)
• ::::::My impression is based on use of Gemini Deep Research which does a web search and then analyses the results. That seems somewhat different from a LLM and more relevant to validation of citations. Andrew🐉(talk) 09:22, 11 May 2025 (UTC)
• :::::::From what I can tell, Gemini Deep Research is simply a framework that organizes repeated LLM runs against a larger set of web pages than vanilla Gemini. -- Beland (talk) 22:19, 11 May 2025 (UTC)
• ::::::::You could try testing 100 or so statements for yourself, but I expect high false positive and false negative results would make such a system not useful for a variety of reasons. 1.) LLMs aren't reliable even for comparing literal statements. That's compounded when the article is summarizing at a higher level than the sources, using very different words. 2.) If multiple sources are cited for the same sentence, it's difficult to know which parts of the sentence are supported by each source, and if all parts of the sentence are supported by at least one. 3.) Segmentation in general is difficult. Is a citation at the end of a paragraph for the whole paragraph or just for one sentence? If there is a citation in the middle of the sentence and one at the end, which words are covered by the first cite? We don't need AI to point out unreferenced paragraphs or sentences, and we already have a backlog of hundreds of thousands of manual tags for that. 4.) I don't know of any way to ensure that the AI hasn't been trained on Wikipedia content or web pages that reuse Wikipedia content; it could easily report all facts in Wikipedia are true simply because they appear in Wikipedia.
• ::::::::There's also the practical considerations of cost and execution time. It currently takes over a day just to run a spell check of Wikipedia with a donated server. I haven't yet solved the engineering problem of running a full grammar check in less than a year of calculation time. It might be feasible to run an AI fact checker on a single page, but even that would require a large number of queries and someone would need to pay for a subscription beefy enough to handle that. We would need infrastructure to track which articles and statements had been checked, and I haven't even gotten that built for spell check yet. -- Beland (talk) 00:11, 12 May 2025 (UTC)
• :::::::::I just asked Gemini Deep Research {{tq|"Please fact-check the Wikipedia article https://en.wikipedia.org/wiki/George_Wendt"}} as this is an article that I worked on today. Its action plan was quite sensible

{{cot|Here's my plan to tackle that topic.}}

Research Websites

1. Access and thoroughly review the content of the Wikipedia article on George Wendt (https://en.wikipedia.org/wiki/George_Wendt).
2. Identify the main factual assertions within the article, including but not limited to: (a) Biographical information (e.g., birth date, place of birth, education). (b) Key career achievements and timelines (e.g., notable roles, particularly in "Cheers", filmography, theatre work). (c) Awards, nominations, and recognitions. (d) Significant personal life details mentioned (e.g., family, marriage).
3. For each significant factual assertion identified, conduct targeted web searches using multiple reputable and independent sources (e.g., established news organizations, biographical databases, academic publications, official entertainment industry resources) to find corroborating or contradictory information.
4. Compare the information found from external sources with the claims made in the Wikipedia article, noting any discrepancies, inconsistencies, or unsupported statements.
5. Investigate any conflicting information by consulting additional reliable sources to ascertain the most accurate and widely accepted facts.
6. Evaluate the neutrality of the Wikipedia article's content, looking for any potential bias, unsourced opinions presented as facts, or overly promotional/critical language.
7. Examine the references and citations provided in the Wikipedia article. Assess their quality, reliability, and whether they directly support the specific claims they are attributed to.
8. Synthesize the findings to provide an overall assessment of the Wikipedia article's factual accuracy, highlighting specific claims that are well-supported, those that are inaccurate or misleading, and areas that lack sufficient credible sourcing.

• Analyze Results
• Create Report
• Ready in a few mins

{{cob}}

• :::::::::The results took about 5 mins and were quite voluminous -- pages of output. The details and analysis seemed consistently rational and to the point. I was especially interested in its comments on a passage that I had added myself. {{tq|"Wendt, playing the character Norm, made a prominent entrance to the Cheers bar in every episode. He would be greeted by a cheer of "Norm!" and make a wisecrack as he walked to his barstool. This regular bit of business was a highlight of the show. Disparaging references to the character's wife, Vera, and the wretched state of his life were other running gags."}} I'd written this as a paragraph with a citation at the end. Another editor had subsequently merged this paragraph with others to make a wall of text. And another editor had cited a YouTube video in the middle of my paragraph which caused confusion.
• :::::::::The issue in this case is that Wikipedia citations don't clearly specify what they are citing. The reader has to make assumptions from the proximity of the citation and the surrounding text and it's easy for this to become unhinged as the text and citations are moved around. A citation should capture the text that is being cited when it is added so that any subsequent drift can be understood.
• :::::::::The overall level and quality of the AI analysis was debatable but seemed comparable with what one would get from an average Wikipedia editor. The advantage of the AI is that it can do it all mechanically and doesn't tire. There is clearly big potential here for such a tool to make systematic checks and highlight issues for investigation. This would be comparable with WP:EARWIG which is routinely used to check for copyvio.
• :::::::::Andrew🐉(talk) 21:49, 21 May 2025 (UTC)
• ::::::::::Without access to the results of this automated evaluation, I can't say whether it was accurate or inaccurate. For example, you didn't quote what it said about the paragraph on Norm's entrances.
• ::::::::::It seems like many pages of textual output is not helpful in automating fact-checking, especially if it contains mistakes that can only be detected by human fact-checking. What would be helpful is having the system flag which sentences are and are not verified by the given sources. Having it go off and consult web pages which are not cited and thus have no bearing on the question being asked seems like a lot of wasted work, which may cause the human interpreting the results to have to do more work to sift through that. -- Beland (talk) 02:11, 22 May 2025 (UTC)
• :::::::::::The details are not important as it was a proof of concept. Running this tool is not as difficult or expensive as you seem to suppose so I encourage you to try it yourself.
• :::::::::::The point is that the current system of citations is quite weak as a form of verification. Whether they use templates or not, the burden is currently on each reader to read and make sense of the cited works. A system of verification and fact-checking which was performed for the reader -- either on-demand or as an offline process would be better.
• :::::::::::Andrew🐉(talk) 11:42, 22 May 2025 (UTC)
• ::::::::::::I would argue the details are what make such a system feasible vs. infeasible. You only ran the AI on one article, but there are almost seven million articles, and it would by no means be feasible to do that if the goal is for editors to know what changes articles need to become 100% accurate.
• ::::::::::::If the goal is for readers to fact-check a single Wikipedia article they are interested in, that's a different problem of a different scale. Readers cannot trust the output of an AI like the one you ran to be accurate or to detect errors in Wikipedia articles, especially since the AI may have been trained on bogus websites or erroneous Wikipedia articles. They would have to fact-check that output by checking its sources. For establishing truth there is no way around applying the traditional techniques of critical thinking, tracing citations, and evaluating the reliability of sources.
• ::::::::::::Readers who want to do this anyway can already do it just like you did, and I don't see how their ability to do that has anything to do with whether or not our citations use templates. -- Beland (talk) 23:04, 22 May 2025 (UTC)
• Support. Metadata is useful; that's pretty much all there is to it. One day we should just use AIs to convert all citations we have into one uniform style, and do it dynamically. --_{Piotrus at Hanyang| reply here} 05:42, 9 May 2025 (UTC)
• :If we have metadata, we should not use it to convert all citations into a uniform style, we should use it like a BibTeX database and allow displaying of the citations in any suitable style. There are massive differences in citation styles and practices between different academic disciplines and one size does not fit all there. —Kusma (talk) 07:38, 9 May 2025 (UTC)
• Oppose. Wikipedia needs to remain accessible. The majority of contributors don't use citations at all. The most important thing for increasing Wikipedia's reliability is to go from no-citations to relevant citations-of-any-format, not to format citations we already have perfectly. Doing this requires the barrier of adding a citation to be as low as possible - someone dropping off plaintext, simple citations needs to be encouraged. I use the citation templates myself usually, but the plain text ones are still fine and not a problem - the value add of microformats is way too small to outweigh scaring off contributors who want to use plain text and would find having their citations forcibly turned into the templates off-putting. (For the scenario where an editor would explicitly like some passing help in formatting their citations with the templates but isn't sure how, no problem with having some noticeboard to drop such requests off, as long as there is some "I am a major contributor" checkbox to avoid passing editors from dropping every single article with plaintext refs in there.) SnowFire (talk) 14:00, 12 May 2025 (UTC)
• :Why not provide tools that make it easier to add citations in a standard format than to manually type in wikitext? There would be no need to say that they have to use the tools, just say that they might save the editor some time. -- Shmuel (Seymour J.) Metz Username:Chatul (talk) 15:27, 12 May 2025 (UTC)
• ::These tools already exist. Have you ever been to an edit-a-thon aimed at new editors? One of the words of wisdom is to run away from these citation tools as fast as you can and say "yeah if you become an expert you can come back and worry about this later." About the best case scenarios are tools which are "drop a raw URL in, get a formatted citation" but anything more complex than that is asking for trouble. If we make it a "mistake" to stop at a plaintext citation, some people will react by not adding citations, which is far worse a loss than the exceptionally minor gain of the auto-formatting. SnowFire (talk) 16:06, 12 May 2025 (UTC)
• Oppose as per various points mention above. In addition (citation) templates imho are often a pain for editors working across several language wikipedias, as each wikipedia tends to have its own non-standardized template zoo.--Kmhkmh (talk) 16:31, 12 May 2025 (UTC)
• Support. I've been using computers to edit and format text longer than most folks here. The idea that people want to hand-format references just blows my mind. Bibliographic citations are structured data and should be managed in ways which preserve that structure for all the reasons described above. On wiki, that means {{t|cite}} templates. And yes, I'm one of those "inexperienced" users who prefers the Visual Editor. The citation tools built into VE are infuriatingly clumsy, but still better than hand-formatting citations. RoySmith (talk) 13:21, 14 May 2025 (UTC)
• :{{tqq|The idea that people want to hand-format references just blows my mind.}} The idea that people want to hand-format anything just blows my mind. I refer to the usual GUI software as What You See Is All You Get{{efn|In the sense that you can't tell what references and white space are fortuitous and what will persist across edits.}} (WYSIAYG) and strongly prefer markup languages such as SCRIPT and {{Stylized LaTeX}} that allow automating complicated layouts. Will a GUI makes simple tasks easier, unless it provides a mechanism to expose and edit markup, it makes more complicated tasks inordinately more complicated. I believe that the best short term strategy is for VE to create the initial template but make it easy to edit the underlying wikitext. -- Shmuel (Seymour J.) Metz Username:Chatul (talk) 18:24, 14 May 2025 (UTC)
• ::There's two distinct things being conflated here: GUI (i.e. Visual Editor) vs source editing, and using citation templates vs hand-formatting citations. Those are orthogonal issues. Anyway, I'll see your LaTeX and raise you "bib | tbl | eqn | troff" :-) RoySmith (talk) 19:21, 14 May 2025 (UTC)
• :::Some editors prefer source editing and others prefer VisualEditor. Either type of editor might be the first to make a citation, so the other system always needs to be able to cope with the result, whether or not templates are used. -- Beland (talk) 19:51, 14 May 2025 (UTC)
• :::Those came late to the game; I started with punched cards and IBM Administrative Terminal System. SCRIPT was a huge jump forward. -- Shmuel (Seymour J.) Metz Username:Chatul (talk) 20:21, 14 May 2025 (UTC)
• Oppose as one of the most zealous CS lovers on Earth. Of all the times I've migrated a mixed-CITEVAR article or mostly debilitated manual CITEVAR article to templates, I've never had real pushback, because no one actually cares in articles they didn't personally contribute to and moreover polish considerably. That makes it obvious to me there's no reason to enshrine a thumb on the scale within site policy. If I ever for whatever reason preferred manual citations for an article, this is a potential headache it is simply needless to conjure. Remsense ‥ 论 16:07, 14 May 2025 (UTC)
• :I think the desire for a rule is because some editors agree with you that badly formatted refs should usually be migrated to citation templates, but feel less bold than you. They're looking for written permission, in a model that Everything that is not permitted is forbidden. WhatamIdoing (talk) 18:00, 14 May 2025 (UTC)
• Support per @WhatamIdoing, @Just-a-can-of-beans, @Chatul, @Ifly6 and @Michael Bednarek. I feel like a lot of opposers are conflating "use of citation templates" with "imposing one particular citation style", when this doesn't have to be the case. Just because there aren't currently (to my limited knowledge, at least) widely-used non-CS1 citation templates, doesn't mean that non-CS1 templates could never exist.I'm certain there would be template editors/coders out there who would be willing to put together Template:Cite APA, Template:Cite CMOS, or similar. (These could have, for instance, some kind of {{code||medium=}} parameter to allow different citation formats for {{code|book}}, {{code|web}}, etc. Additionally, there could be mode parameters for expanding abbreviated formats like volume/issue/page, as discussed in {{slink|User:Jorge_Stolfi#Please_do_not_use_{{(}}{cite{{)}}}_templates}} mentioned by @jacobolus above). This would allow different disciplines to retain their preferred citation style, while also allowing for easier standardisation, data tracking, and other benefits of templates. This could be accompanied by another set of templates, similar to Template:Use DMY dates/Template:Use MDY dates, which would indicate to editors the citation style to use in the given article (these could be Template:Use CS1, Template:Use APA style, etc). Eventually, like with the date format templates, citations could be inputted using any citation template, and then the parameters could be formatted automatically based on the "Use X style" template at the top of the page. But I think I'm getting ahead of myself.{{pb}}Aside from the issue of citation styles, I would be hesitant to risk patronising new editors by suggesting citation templates are too difficult for them to grasp. The most commonly-used citation templates (cite web, cite news, cite book, cite journal) are already available by default in the editing toolbar via a form interface with prompts for clearly-labelled parameters that, in my opinion, can be easily understood, even by brand new editors.{{pb}}It's also worth noting that, per WP:CIRNOT, it's okay for new editors to make good-faith, constructive edits that don't 100% conform to the MOS; {{tq|Articles can be improved in small steps, rather than being made perfect in one fell swoop. Small improvements are our bread and butter.}} So, if the MOS was updated to prefer templates (which is what this RfC is suggesting, not that we mandate templates and "punish" users for citing manually), that would just give future editors the ability to standardise these citations using templates while retaining the information added manually by the original editor. If anything, such a guideline might reduce edit warring, because it would reduce ambiguity, especially for articles where no citation style has previously been established.{{pb}}Apologies, this turned into a bit of an essay. If I've got anything blatantly wrong, I'm very open to corrections! Pineapple Storage (talk) 13:27, 15 May 2025 (UTC)
• {{ping|Pineapple Storage}} Well, since you offered... you phrased the matter as "patronising new editors by suggesting citation templates are too difficult." Now, there will always be dedicated people who love to grapple with this, but I'm sorry, this is just a factual issue. Please read https://xkcd.com/2501/ . I write this as someone who's been to edit-a-thons attended by smart, dedicated, interested people with college degrees and the like, and I can assure you that yes, messing with writing out citations is just factually difficult for many people. A problem with UI design is that it doesn't matter how much your 20% of power users reassure you that everything is fine, the 80% of quiet occasional users are hard to poll and much worse at handling your app / website / etc. than you think. Citation templates are not trivial. It would be an acceptable price to pay if the matter was very very important, but is it here? I really don't think so. SnowFire (talk) 21:39, 16 May 2025 (UTC)
• :File:Auto-citing_a_source_using_VisualEditor,_small.gif
• :Yes, but it's not hard to use a citation filler, and that's what most new editors do. Most edit-a-thons start by telling people to use the visual editor (older version of which is what's blinking at you here), but there's a citation filler in the 2010 wikitext editor, too. Even a newbie can paste a URL into a dialog box.
• :And the question here isn't "Shall we tell people on their first edit that they should do this thing?" but much closer to "Do I really need to have a full-blown CITEVAR discussion on the talk page before I quietly re-format the citations?" WhatamIdoing (talk) 21:44, 16 May 2025 (UTC)
• WhatamIdoing: Did you read https://xkcd.com/2501/ ? Please do so. Wishes are not reality. Yes, it is hard, no matter how much we can tell ourselves it's easy. We are geochemists talking among each other about how it's actually easy to understand the nature of ionic bonds involving Calcium all day, and among the people already in the group, yes, it very well may be easy. But that doesn't mean it's true of the least-skilled-with-templates/wikitext 20% of Wikipedia editors (who are still a very useful and positive resource to have!), and certainly not true of the misty potential Wikipedia contributor who isn't great at the tech side but might be convinced to join up, if it doesn't seem like there's an insurmountable barrier of policies telling them that they're in error for everything they add. SnowFire (talk) 21:51, 16 May 2025 (UTC)
• :I'm familiar with the comic strip. You, on the other hand, [https://en.wikipedia.org/w/index.php?title=Special%3AContributions&target=SnowFire&namespace=all&tagfilter=visualeditor&start=&end=&limit=250 do not seem to be familiar with the visual editor], which is how newbies mostly get started these days.
• :Imagine a world in which nobody says anything as incomprehensible as "Please use this thing we've called a citation template". Imagine instead that there's a "Cite" button in the toolbar, and when you click it, it has a little box for you to paste your URL in. And then it magically turns your URL into something that looks very nice, and all you have to to is click the blue button to insert it. You never see the "template" and don't have to even know what it is.
• :Try it out. Go to https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&veaction=edit Mash the keyboard to put some text on the page. Click the "Cite" button in the toolbar, and either paste in a URL (https://www.example.com) or put in a DOI or an ISBN if you want to get fancy. See what happens. WhatamIdoing (talk) 22:02, 16 May 2025 (UTC)
• :@SnowFire I totally understand where you're coming from, and I think the xkcd comic is good and probably does apply to some more obscure/backend WP mechanisms, but I really don't think it applies here. I write this as someone who was a brand new editor less than three years ago, and my {{userContribs2|Pineapple Storage|11|eleventh ever contribution|2022|08|12|2017}} was creating a page with 8 template-formatted citations. What spurred me to start editing was (if I remember rightly) seeing a typo on a page and thinking "Hmm, I wonder whether I could correct that...", so I clicked Learn to edit in the left-hand menu and went through the introduction. Help:Introduction to referencing with Wiki Markup/2 says {{tq|To add a new reference, just copy and modify an existing one.}} and that was exactly how I started adding references to my sandbox draft. If anything, the fact that citation templates are so readily available made it far easier for me to start referencing, because I could just copy and paste {{code|{{Cite web |date=1 January 2001 |last=Smith |first=Jane |url=https://www.example.com/example_page |title=Example page |website=Example}}}} and replace the values with my own, without having to figure out how to format the citation myself: {{code|Smith, Jane (2001)."[https://www.example.com/example_page Example page]". Wikipedia. 1 January.}} (Just typing that out was exhausting, let alone having to look up an entry in Harvard MOS etc.) I understand that there will always be new editors who don't want to spend any time reading through tutorials etc, and that's fine; this RfC is just proposing that, if they choose a manual citation format (for instance, one that doesn't comply with any formal citation style, which, in my experience, manual citations often don't) then other editors can come along and standardise them using templates later, without anyone jumping done their throat about WP:CITEVAR. Pineapple Storage (talk) 23:49, 16 May 2025 (UTC)
• ::By the way, for anyone curious about what the tutorials for new editors say about referencing/citations, and citation templates specifically, see Introduction to referencing with Wiki Markup §{{nbsp}}RefToolbar and Introduction to referencing with VisualEditor §{{nbsp}}Adding references. Pineapple Storage (talk) 09:27, 22 May 2025 (UTC)
• :Re {{tq|"there aren't currently (to my limited knowledge, at least) widely-used non-CS1 citation templates"}}, I much prefer the CS2 template {{tl|citation}}. The CS1 templates of {{tl|cite conference}}, {{tl|cite magazine}}, {{tl|cite podcast}}, etc. seem absurdly baroque and confusing. I use the {{tl|citation}} template as a simpler and more universal option. I'm surprised that there hasn't been more pressure to standardise on this. There's a similar issue with infoboxes and there's a lot of pressure to merge and consolidate those. Andrew🐉(talk) 11:57, 22 May 2025 (UTC)
• (de-indent, re WhatamIdoing) Correct, I don't like VE that much, other than messing with large tables. But yes, I have used it anyway when interacting with new editors. And yes, I've even shown off the "transform a URL into a citation" option before! But the very fact we're talking about the toolbar at all says a lot. Again, I'm not making this up, many new editors have trouble with the absolute basics of editing. Telling them "if you switch to VE and find the right button there's a tool that automatically generates well-formatted citations, which are useful because, er, microformats and machine readability" is a lot when someone is learning the basics. And again, there's a lot that newbies need to learn, but why insist on this one in particular? I don't know what to say other than that I'm talking about the 20% least skilled in wikitext contributors + the current non-editors but potential future contributors who are scared off by assuming that editing Wikipedia is very complex. I have multiple talented, smart, educated friends who just ask me to make simple text edits on their behalf, the kind that don't require knowledge of citations at all. Sure, maybe some would never edit WP and thus are irrelevant, but it is hard to overstate how imposing the very basics of editing are on Wikipedia. The people who have trouble with this are not going to find this discussion, but we should keep them in mind regardless and advocate on their behalf. SnowFire (talk) 23:14, 16 May 2025 (UTC)
• :Why insist on this one? Because:
• :* Nobody says the newbie has to do any of this. I know this has been repeated multiple times, but let's be clear: The goal is not to make the newbie figure out how to format any citations through any method at all. If the newbie can get a URL somewhere in the vicinity of the edit, even if it's just in an edit summary or a note on a talk page, the wikignomes will do their best to fix it. It happens that clicking a button in the toolbar (something most ordinary people don't find difficult?) will produce a satisfactory result in both the wikitext and visual editors, but in the visual editor, the newbie will have no idea that there's a template being used, and therefore cannot experience any of the disadvantages of citation templates. (For example, there's no "visual clutter" in the wikitext when you don't see the wikitext at all.)
• :* The goal is for an experienced Wikipedia:WikiGnome to know whether, when faced with a mishmashed mess in an article, whether the community (a) would prefer the mess cleared up using citation templates or (b) would prefer the mess cleared up without using citation templates or (c) still feels that it's necessary to keep pretending that we're all 'neutral' about citation templates and that this stated neutrality will have any practical effect other than the wikignomes "randomly" choosing to use citation templates "on a case-by-case basis". Anyone who's been watching for the last dozen years knows that 'neutral' means citation templates in practice, but sometimes we have social/political reasons to say one thing while doing another.
• :WhatamIdoing (talk) 02:46, 17 May 2025 (UTC)
• Oppose I'm a zealous supporter of templated cites as they are much, much less susceptible to data rot than the text based alternative. I would support a preferred status for templates, a light finger on the scale for their use, but I don't support the current wording - it goes to far. Currently if you find an article you believe would be improved by citation templates you only have to get consensus on the talk page for doing so. These nothing in CITEVAR that says it can't be changed, only that there has to be consensus for it. -- LCU ActivelyDisinterested «@» °∆t° 23:53, 16 May 2025 (UTC)
• :I would be worried that certain WP:OWNers will fight to the ends of the earth on that. Ifly6 (talk) 19:57, 17 May 2025 (UTC)
• ::If one editor tries to own an article, it just takes two others to form a consensus for change. Styles are not set in stone, they are as open to changing consensus as much as any other article content. -- LCU ActivelyDisinterested «@» °∆t° 13:00, 19 May 2025 (UTC)
• :::A majority of 2 users against 1 is not "consensus". Requiring local consensus is a way to make global consensus impossible to implement. Nemo 12:52, 22 May 2025 (UTC)
• ::::My comment was about a single article, not a global consensus. Local consensus about article content is quite normal. -- LCU ActivelyDisinterested «@» °∆t° 18:03, 22 May 2025 (UTC)
• :::::I wouldn't describe a consensus formed on a talk page, affecting a single article, as "local consensus". That's "ordinary consensus". WP:LOCALCON is about Alice and Bob deciding that all of "their" articles are exempt from relevant policies and guidelines. It's not about three editors making an ordinary decision on a talk page. WhatamIdoing (talk) 18:59, 22 May 2025 (UTC)
• Oppose per jacobolus among others. Not everything needs to be filed away into templates. Plain text can be easier to format. Cremastra (u — c) 18:08, 17 May 2025 (UTC)
• Oppose, and I say this as a huge fan of citation templates who uses them almost exclusively. This seems like WP:CREEP to me and does not have any measurable benefit other than allowing scripts and bots to read metadata. It should be sufficient that citations are consistent within an article. Mandating that people use citation templates brings up various problems, not the least of which is that changing citation formats is a waste of editor time when citations are already consistent in that article. The main purpose of a citation is to verify text, not to be visually appealing. Epicgenius (talk) 14:42, 20 May 2025 (UTC)
• :Allowing scripts and bots to read metadata sounds like a good enough reason to me. RoySmith (talk) 14:47, 20 May 2025 (UTC)
• ::@RoySmith, I agree, and if this were a smaller wiki that already used citation templates predominantly, I would support this without reservation. The problem is that mandating this on millions upon millions of articles seems to be very cumbersome, especially if existing hand-crafted citations seem to work fine. To be fair, I could still support this if the wording were toned down. However, as currently written, it effectively gives editors free rein to indiscriminately convert manual citations to cite templates—which could result either in a waste of editor time (due to the amount of time that is required to do this carefully) or sloppy automated conversion of citations (if they use a tool like VisualEditor or reFill). Epicgenius (talk) 19:09, 21 May 2025 (UTC)
• Support There’s a lot to read here (can’t claim to have read all of it), but I don’t see anything about the encyclopaedia reader. It is a lot easier to find where article content comes from with short references if templates are used (mouse over etc). Shouldn’t we be focusing on the reader experience? ThoughtIdRetired _TIR 19:53, 22 May 2025 (UTC)
• Oppose The citation templates are now extremely complicated, and really need a review and simplification before we move even to "encourage". If that were done I'd support some kind of gentle encouragement or at least discouragement from un-templating. All the best: Rich Farmbrough 14:05, 4 June 2025 (UTC).
  

::Complicated is subjective. Writing a reference from scratch is probably easy for someone who has spent a lifetime dealing with academic references. But for the new, younger editors that we are trying to encourage, I would guess that a template is a lot easier. Once they realise that cite book or cite journal are usually automatically populated from an ISBN or a doi, they only need to deal with cases where that doesn't fit. The solution to that is to ask for help. ThoughtIdRetired _TIR 16:16, 4 June 2025 (UTC)

:::It's also frankly not that hard to turn a Google Scholar formatted citation into a template. The parameter names just aren't that difficult. {{parameter|title}} is no Finnegans Wake. Ifly6 (talk) 16:42, 4 June 2025 (UTC)

{{notelist-talk}}

When an article is following a third-party style guide that says, for example, "use title case for books and sentence case for journal articles", which rules for implementing title and sentence case should be used? Wikipedia has guidelines for this at Wikipedia:Manual of Style/Capital letters, Wikipedia:Manual of Style/Titles of works, and some topic-specific pages.

For example, Wikipedia guidelines say the name of the game "Go" should be capitalized, but the name of the game "chess" should not. If a third-party style guide says "go" should be lowercase, which should take precedence? (The specific example is not important; I'm just using it as a neutral illustration.)

The sensible choices I can think of:

• Follow the Wikipedia guidelines defining title and sentence case
• Follow the third-party guidelines defining title and sentence case
• Pick one or the other for a given article and use WP:CITEVAR to arbitrate that choice

My reading of the current Wikipedia guidelines is that they apply to citations regardless of third-party citation styles; if there is consensus against that, I would support adding a note explaining the exception.

-- Beland (talk) 17:59, 14 May 2025 (UTC)

:In practice, the last one is the one that's most likely to 'stick'. WhatamIdoing (talk) 18:01, 14 May 2025 (UTC)

:Considering that we have our own rules for title case, what would be the point in not following them if title case is to be used? Gawaon (talk) 18:39, 14 May 2025 (UTC)

::I agree; just wanted to check because some people are interpreting citation capitalization guidelines in very different ways than I would expect. -- Beland (talk) 18:46, 14 May 2025 (UTC)

:That's the problem with forcing title or sentence case, there's always issues of which capitalization is considered proper... That's why none of these options are an improvement but instead serve to pigeonhole individuals into others' preferred capitalization style. Hey man im josh (talk) 19:10, 14 May 2025 (UTC)

::Pretty much anyone who has opinions about style will find something they don't like about any given manual of style, but the point of having one is to resolve those disagreements in favor of a single style so that presentation to readers is consistent. I don't see any particular reason that personal freedom to capitalize at will should be valued over quality of reader experience. -- Beland (talk) 19:36, 14 May 2025 (UTC)

:::We don't have a single allowed style though, that's the problem. And frankly, I'm not sure how reader experience is affected by the capitalization of references. Really don't think it makes anyone's experience worse. Hey man im josh (talk) 13:33, 15 May 2025 (UTC)

::::I agree with Josh… we allow multiple citation styles, and that isn’t a problem. Most editors don’t care about capitalization in citations - as long as it is clear what book, journal, website etc the citation is pointing to.

::::For those who do care: feel free to conform to your favorite style, but don’t argue about it… and definitely don’t edit war about it. If someone else objects, and reverts your change… move on to another article and leave it be. Blueboar (talk) 14:48, 15 May 2025 (UTC)

:::::I've been doing some edits to fix case errors, and sometimes that includes changing citation style when there does not appear to be a consistent citation style in use. Josh reverted a bunch of those, even though I was careful to first verify that the style was not consistent either internally or with corresponding outside sources. I'm not claiming that I made everything perfectly consistent, but I moved closer to the most common style, which was sentence case. Examples: [https://en.wikipedia.org/w/index.php?title=Jaylon_Jones_(American_football,_born_2002)&diff=prev&oldid=1289426634], [https://en.wikipedia.org/w/index.php?title=Jonathan_Mingo&oldid=prev&diff=1289426626], where the titles I changed included several that were made up, that is, not findable in the source (e.g. [https://www.nfl.com/prospects/jonathan-mingo/32004d49-4e17-0999-c137-d684c1d1ab38 title=Jonathan Mingo Draft and Combine Prospect Profile]). It's annoying. Dicklyon (talk) 05:21, 16 May 2025 (UTC)

::::::If the capitalization matched neither the source nor a style guide, what was the reason for reverting? -- Beland (talk) 05:35, 16 May 2025 (UTC)

:::::::The reason for reverting, in those two examples, was that the articles were not made "entirely consistent", meaning that Dicklyon's changes were not an improvement. Prior to that, the references matched the source capitalization. Hey man im josh (talk) 15:05, 16 May 2025 (UTC)

::::::::No, they didn't: several were over-capitalized with no good reason. I was aiming for consistency; if I came up short, let me know and I'll work on it some more. Dicklyon (talk) 16:36, 16 May 2025 (UTC)

::::::"Jonathan Mingo Draft and Combine Prospect Profile" is from the browser's tab. ~WikiOriginal-9~ (talk) 17:04, 16 May 2025 (UTC)

:::::::It's interesting that some of these titles are from the page metadata, not visible on the page. Some others are not (e.g. [https://draftscout.com/dsprofile.php?PlayerId=1040862&DraftYear=2023 title=2023 NFL Draft Scout Jaylon Jones College Football Profile] appears to be "made up"). Dicklyon (talk) 18:26, 16 May 2025 (UTC)

:If you care about that level of detail, use the Wikipedia MoS. In general just using the right case will be such an improvement over using the wrong case that it is probably sufficient. I often use https://titlecaseconverter.com/ (which has a Wikipedia option) for fixing titles in all caps. All the best: Rich Farmbrough 14:29, 4 June 2025 (UTC).

Adweek posted on Instagram. That's what came up in a Google search for the topic I was trying to find. How do I properly indicate Adweek is responsible?— Vchimpanzee • talk • contributions • 18:13, 16 May 2025 (UTC)

:Have you looked at {{tl|cite Instagram}}? WhatamIdoing (talk) 21:46, 16 May 2025 (UTC)

::How do I find postid?— Vchimpanzee • talk • contributions • 22:08, 16 May 2025 (UTC)

:::I don't know anything about Instagram, but perhaps if you ask your favorite web search engine, there will be a how-to page available. WhatamIdoing (talk) 22:24, 16 May 2025 (UTC)

::::Looking at some random instagram post, the URL is https://www.instagram.com/p/BZl8azcjBf4/. It's a good guess that the last part of that (i.e. "BZl8azcjBf4") is the post id. RoySmith (talk) 09:30, 17 May 2025 (UTC):

:::::Thanks. I tried that earlier and it didn't work. Somehow it worked this time.— Vchimpanzee • talk • contributions • 16:45, 17 May 2025 (UTC)

::::::@Vchimpanzee, a bit late, but I added Template:Cite_Instagram/doc#postid in response to this thread. Glad you were able to get the citation to work the second time, Rjjⁱⁱⁱ (talk) 16:56, 17 May 2025 (UTC)

:::::::For all the good it did me. The entire section was thrown out. However, I put back my contribution because I think it's worthwhile information.— Vchimpanzee • talk • contributions •

:::::::And someone reverted me in another article just because the source was Instagram. Luckily, a source behind a paywall--probably the same source, since it was Adweek's Instagram--had what I needed to verify before the part that was blocked.— Vchimpanzee • talk • contributions • 16:07, 18 May 2025 (UTC)

::::::::There's nothing wrong with citing a magazine's copy of an article that they posted on social media instead of the same one on their website, but I suppose people don't necessarily check what's going on.

::::::::If you need another source, try this one:

::::::::* Knopper, Steve, and Mike Cessario. “2025 Branding Power Players.” Billboard 137, no. 4 (March 8, 2025): 53–63.

::::::::If you put the article title into the main search box for Wikipedia:The Wikipedia Library (within quotation marks) it should be easy to find a free-to-read copy. WhatamIdoing (talk) 16:29, 18 May 2025 (UTC)

To be precise, I am asking about the online edition of The Encyclopedia of Science Fiction, a RS [https://sf-encyclopedia.com/] that had paper editions, but for a decade now has been an online encyclopedia. Its entries do not give a date of creation, although they have a "last updated" date at the very top. From experience I know that the last update can cover both major changes as well as minor ones like adding a single hyperlink (just like on Wikipedia). History function is semi-handled by links to Internet Archive, to which each entry links. Some entries predate the online version and are revised from as far back as the 70s (first paper edition). Is it ok to give the date based on the "Entry updated"? _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 00:42, 18 May 2025 (UTC)

:@Piotrus I'm obviously not as experienced as you but I've cited this source a bit and whenever I do I just leave out the date. But maybe the updated one would be fine. PARAKANYAA (talk) 00:52, 18 May 2025 (UTC)

:I suggest that if you list a |date= (instead of relying on |access-date= alone), then you should use the updated date. If the creation date is later discovered, then it can be put in |orig-date=. WhatamIdoing (talk) 03:41, 18 May 2025 (UTC)

::I would follow {{u|WhatamIdoing}} here: I don't tend to use the {{para|date}} parameter for websites, unless they explicitly state a last updated date, but when they do, that's the "date" we're interested in. Using {{para|orig-date}} where possible seems like a sensible idea too. UndercoverClassicist ^T·C 21:29, 21 May 2025 (UTC)

I'm starting a new section rather further derail the thread above. I tried out Andrew's prompt, "{{tq|Please fact-check the Wikipedia article https://en.wikipedia.org/wiki/Example}}"

I suspect that Gemini will struggle more with topics that are either obscure or surrounded by misconceptions. I tried the prompt above for the Piri Reis article. I plan to nominate this for FAC sometime in the future, and it's a topic that is surrounded by misconceptions and spooky stories. Here are the results:

• "{{tq|A snippet from a scholarly article suggests that both Maximus Planudes (c. 1260-1310) and Piri Reis may have depicted a 'North American Baptistery,' identified as the Newport Tower, on their maps. This implies a much earlier and more detailed knowledge of North America in the Byzantine and Ottoman Empires than commonly understood, which is a notable new insight not present in the general Wikipedia article.}}" This is pseudohistorical nonsense.
• "{{tq|While the Wikipedia article mentions Ptolemy's *Geographia* as a source for Piri Reis, it does not delve into this specific interpretation involving Planudes or the Newport Tower.}}" It seems hung up on this "Newport Tower" angle. It's in Rhode Island, which no reliable source even claims is present on the map.
• "{{tq|The article states Piri Reis's execution occurred in 1553 , while a scholarly source indicates 1554.}}" More recent scholarship makes 1553 almost certainly the correct date.
• "{{tq|This minor discrepancy should be noted, perhaps by presenting a range (e.g., "1553/1554") or acknowledging the scholarly debate surrounding the precise year.}}" Sure, but the article already says, "Some older sources list the year of his death as 1554. Venetian documents from the period have allowed historians to date his execution to some point during 1553.".
• "{{tq|While the article correctly states that the 1513 map is "accurate for its time," it could more explicitly incorporate the nuanced finding from Gregory McIntosh, who concluded that it was not the most accurate map of the entire 16th century, having been surpassed by numerous later works. Adding this context would provide a more precise historical placement of the map's accuracy.}}" This is the best point so far. The Piri Reis map actually does give this context. It's just out of scope for the cartographer's biography. A thought though, the LLM has checked the map's Wikipedia article during this fact check. Is it basing its advice on a different Wikipedia article?
• "{{tq|The article's debunking of pseudoscientific claims, particularly the "ice-free Antarctica" hypothesis, could be more detailed and explicit. Incorporating the geological and cartographic arguments presented in scholarly sources would strengthen the refutation.}}" I'm not sure how much more the LLM is wanting here. Ther article already includes "{{tq|Hapgood's book was met with skepticism due to its lack of evidence and reliance on polar shift.[158] According to geologist Paul Heinrich, the book also did not account for post-glacial rebound, and the 1949 survey initially cited by Mallery could not measure even one percent of the area drawn in the Piri Reis map. Subsequent studies have shown no significant similarities to Antarctica's coast.[159]}}" I think it may be asking of the kind detailed analysis of the Carribean section, but again I think that's probably out of scope for this article and it's covered in the map's article.
• "{{tq|Additionally, a brief mention and debunking of the Newport Tower theory, referencing the claims and counter-arguments from academic analyses , would further enhance the article's comprehensiveness in addressing common misconceptions.}}" I don't know why it is so caught up on the Newport Tower.
• "{{tq|While summarized, the article could elaborate further on the unique breadth of information contained within Kitab-ı Bahriye. Highlighting its inclusion of socioeconomic, environmental, and cultural details, beyond mere navigational data , would more fully convey its significance as a comprehensive strategic document rather than simply a nautical guide.}}" This is really more true of the Italian isolario genre, but a valid point is that I could give some additional detail. Even though I won't take the advice here, this is probably the first point where I'll make some kind of change to the article. I will probably add a couple lines based on Kitab-ı_Bahriye#Subjects_covered to flesh this section out. I would say this comment is useful for an editor who knows the material and harmful for an editor who does not.
• "{{tq|The article could more explicitly discuss the limited contemporary circulation and appreciation of Piri Reis's works during his lifetime, contrasting it with their significant posthumous recognition and their role in fostering national pride in modern Turkey.}}" The first paragraph of Piri_Reis#Legacy covers this in some detail already.
• Now we're at reccomendations: "{{tq|Update the execution date to reflect the scholarly debate, potentially using a range like "1553/1554" or adding a brief note on the differing accounts.}}" This is useless.
• "{{tq|In the "1513 World Map" section, perhaps within the "Significance" or "Characteristics" subsection, add a sentence or short paragraph clarifying that while the map was accurate for its time, it was subsequently surpassed by other 16th-century maps, citing the work of Gregory McIntosh.}}" This not good advice, and I suspect it is drawing from the map's Wikipedia article. A potential point is that maybe like one sentence could be added to the end of the section, but more detailed coverage of the map should be in the map's article and not duplicated in the cartographer's article.
• "{{tq|Enhance the "Pseudoscientific claims" section to include a more detailed refutation of the "ice-free Antarctica" hypothesis, drawing on geological evidence, cartographic practices of the era, and the lack of historical evidence for pre-1770s Antarctic voyages.}}" Again, I think this is not great advice and it feels like it is trying duplicate the map's article into the cartographer's article.
• "{{tq|Consider adding a brief subsection or paragraph to the "Pseudoscientific claims" section addressing the Newport Tower theory, outlining Shekleton's claims and presenting the scholarly counter-arguments regarding carbon dating and the interpretation of early maps.}}" Why does this keep coming up?
• "{{tq|Enhance the "Kitab-ı Bahriye" section with more specific examples of its non-navigational content, drawing from the detailed descriptions in the provided research. This would emphasize its comprehensive nature as a strategic intelligence document.}}" I noted above that I won't follow this exact advice, but it is a good point about giving some kind of tangible example.
• "{{tq|Strengthen the "Legacy" section by explicitly contrasting the limited contemporary appreciation of Piri Reis's work with its profound later national and international recognition, linking this resurgence of interest to the 1929 rediscovery and its significance for modern Turkish identity.}}" Again, this is the first paragraph.

So after reading that, I would say that it raises one good point for an editor with knowledge of the subject. I think it would be overall harmful for an editor with no prior knowledge of the subject.

One significant problem that I did not think about beforehand is how much Gemini relied on other Wikipedia articles. I'm the primary author for the articles on his cartographic works, the Kitab-ı Bahriye and Piri Reis map, and I noticed a bunch of places where it was clearly plagiarizing/scrambling material from those articles. According to Gemini's source list, it was able to access [https://ia804603.us.archive.org/6/items/gregory-c.-mc-intosh-the-piri-reis-map-of-1513/Gregory%20C.%20McIntosh%2C%20Norman%20J.%20W.%20Thrower%20-%20The%20Piri%20Reis%20Map%20of%201513-University%20of%20Georgia%20Press%20%282000%29.pdf Gregory McIntosh's book] so I am not sure why it relied so much on the Wikipedia article:

;Gemini

:"{{tq|The distinctive and somewhat peculiar arrangement of the Caribbean on the 1513 map, which combines features of Central America and Cuba into a single landmass, is attributed to Columbus's original belief that he had reached Asia. Furthermore, Hispaniola is depicted as merged with Marco Polo's description of Japan, reflecting the geographical understanding of the time.}}"

;Wikipedia

:"{{tq|The northwestern coast combines features of Central America and Cuba into a single body of land. Scholars attribute the peculiar arrangement of the Caribbean to a now-lost map from Columbus that merged Cuba into the Asian mainland and Hispaniola with Marco Polo's description of Japan. This reflects Columbus's erroneous claim that he had found a route to Asia.}}"

;McIntosh

:"{{tq|The Columbian conception of the transatlantic lands and islands (as recorded in Columbuss writings and the writings of his contemporaries) and the Toscanelli-Martellus-Rosselli-Behaim conception of the East Asian coast are combined with the geography of the West Indies and the Caribbean to produce the configurations of the Piri Reis map—configurations that are copied from Columbus's map.}}"

I guess it's a bizarre complaint to criticize who the LLM plagiarizes, and in this case I do think the Wikipedia article is solid, but shouldn't it try to plagiarize the best source it can find? Why didn't it center on the book? Does the length or some other aspect of a book PDF make it too difficult? Does it consider a 25-year-old book to be outdated? The Kitab-ı Bahriye article makes more sense because pretty much all the other good sources for this book are offline or pay-walled; the LLM may have no way to access any of them. I tried running the prompt again on the book's article, and although Gemini did give me a result it was very bogus advice and the LLM could cite no sources, so I'm not sure how it did any fact-checking.

I also tried running the prompt on Roswell incident which is a Featured article about a topic heavily linked to misconceptions and conspiracies. Gemini had very little suggestions on this one, but did say, "{{tq|Initiatives should be developed to improve media literacy and critical thinking skills among the general public.}}" Which, okay, that's valid, but how am I supposed to do that from a Wikipedia article? It also gave the advice, "{{tq|Ongoing efforts to digitize and make such records easily searchable for both researchers and the public are crucial to ensure that factual information is readily available to counter persistent myths.}}" Which seems to assume that I work for the Air Force?

I realize that I am kind of focused on the topics where an LLM will struggle. Perhaps there are other areas where it is more effective, Rjjⁱⁱⁱ (talk) 04:53, 22 May 2025 (UTC)

:{{ping|Andrew Davidson}} This is why I say the details and quality of the response matter; this sounds much worse than a manual fact check against cited sources. -- Beland (talk) 17:35, 30 May 2025 (UTC)

:Thanks for the examples and comments. Trying such tools out to see what works is a sensible way forward. I see that the WMF have published their plans for using AI tools and their priority is to use them to support human editors, helping them to check content integrity, for example. This sounds similar but we shall see... Andrew🐉(talk) 18:57, 30 May 2025 (UTC)

:About a year ago I tried to get ChatGPT to even acknowledge that something Wikipedia had wrong was wrong. Eventually I gave up. However the systems are getting better. All the best: Rich Farmbrough 14:33, 4 June 2025 (UTC).

::Right, technology moves so quickly that this discussion could look outdated in five years. To use machine translation as a comparison, I remember a point where it often created gibberish. Now, editors regularly use machine translation to check the hard facts (names, dates, places, and so on) in cited sources written in other languages. With LLMs, I have used prompts like, "Make an outline of this article to write three introductory paragraphs"; that lifts the burden of fact-checking from the LLM and lets it just crunch the article down. I've also had some luck in technical areas with asking an LLM to go through and add comments; I've never added these AI-generated comments, but it helped me read and update templates. When it comes to fact-checking, I'm not sure. Perhaps it could skim a high-level article for missing facts rather than bogus ones. Rjjⁱⁱⁱ (talk) 03:01, 5 June 2025 (UTC)

:::If you have an AI write an article intro, you need to manually verify all the claims made in that intro against the body, because LLMs can randomly inject false claims that sound truthy. Likewise, any "missing facts" would need to be manually fact-checked against cited sources. -- Beland (talk) 13:03, 5 June 2025 (UTC)

::::(On mobile right now.) I agree. I am using "outline" very literally above. In the case of the lead, if an editor has already written the body content from sources then the human editor has the facts. An ai-generated outline could straight up invert a fact, but still provide an overview of the topics covered for a human editor to reference when seeking due weight in the lead. That is how an outline "lifts the burden of fact-checking from the LLM". The AI technology we have now can't deal in facts; people can. I am still unsure of the best uses for AI, Rjjiii (ii) (talk) 14:49, 5 June 2025 (UTC)

:::::There's no particular guarantee that an LLM will respect due weight, either in the abstract importance of a given fact that can only be assigned by reasoning, or in terms of the number of words devoted to it in the body. The intro may instead reflect the weighting of the LLM's training data, or simply be random because it's not calibrated for this purpose. -- Beland (talk) 15:17, 5 June 2025 (UTC)