Hi, One note that hasn't been mentioned is that the future UI for consuming information will be "agentic," where the tool used will search for and collect the latest information on the topics. This means systems will prioritize content where the information source is described (i.e., references), as accuracy can be validated from there. So references are one of the strengths of Wikimedia services compared to almost all other sites on the internet. This doesn't really solve the eyeball problem since the content is still used externally, but it should be good for something.
Br, -- Kimmo Virtanen, Zache On Fri, Jul 10, 2026 at 3:14 AM Steven Walling via Wikimedia-l < [email protected]> wrote: > EU data provenance regulations do not require foundation model > providers to link to the constituent sources of AI model knowledge. In fact > the disclosure rules don’t even apply to foundation models themselves yet, > only services. And they only require disclosing that text or other media > were generated by AI (via the interface and metadata). It’s not going to > help fix brand awareness and loss in referral traffic, only transparency > that a particular piece of content is or is not AI generated. > > Google already links to Wikipedia when it is used as a source in search > answers, and we’re still seeing a traffic decline. We shouldn’t wait to > assume that EU regulations or walled gardens are going to save us from > being disintermediated. > > The drop in enwiki traffic in particular is likely indicative of a > structural shift considering that in just two years (2024-2026) we have > gone from 33% of Americans using chatbots and AI summaries to 49% ( > https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/ > ) > > Steven Walling > > On Thu, Jul 9, 2026 at 7:39 AM Ilario Valdelli via Wikimedia-l < > [email protected]> wrote: > >> James, thank you for grounding this with real numbers rather than >> impressions — the Similarweb figure and the cross-language health content >> analysis are exactly the kind of evidence this thread needs, and they're >> worth taking as the starting point. >> >> That said, I don't think there's much left to gain from debating what >> should have been done differently. What happened, happened — the past >> doesn't come back, and no amount of hindsight changes the trajectory from >> here. >> >> What matters more now is the immediate ahead of us, and looking at it >> with a different mindset: not only as a problem, but as something that >> might open an opportunity — because very often what looks like a threat to >> us is somebody else's opportunity, and figuring out which side of that >> split we're on matters more than assigning blame for the past. >> >> A few things, connected, that point in that direction (even not a >> complete picture). AI infrastructure spending is running above $1 trillion >> globally this year, with investors increasingly nervous about the return on >> it. [1] At the same time, open-weight models that run locally — largely >> out of China, but not only — are now within a few points of frontier >> benchmarks at a fraction of the cost.[2] Regulation is starting to require >> actual, checkable data provenance rather than declared compliance (EU AI >> Act for instance). And large parts of the web are moving toward walling off >> AI crawlers and charging for metered access by default.[3] >> >> Put together, I think James's numbers might not just describe a decline — >> they might be the leading edge of a market starting to separate "content >> that's cheap to grab" from "content whose origin can be verified," and >> pricing them very differently. If that's right, it matters most exactly >> where James's data points: health content is where the cost of being wrong >> is highest, and where verifiable sourcing should be worth the most, not the >> least. >> >> kind regards >> >> Ilario >> [1] >> https://about.bnef.com/insights/data-centers/ai-data-center-build-advances-at-full-speed-five-things-to-know/ >> [2] >> <https://about.bnef.com/insights/data-centers/ai-data-center-build-advances-at-full-speed-five-things-to-know/> >> https://computingforgeeks.com/open-source-llm-comparison/ >> [3 >> <https://about.bnef.com/insights/data-centers/ai-data-center-build-advances-at-full-speed-five-things-to-know/> >> https://techcrunch.com/2026/07/01/cloudflares-new-policy-pushes-ai-companies-to-pay-for-publishers-content/ >> >> <https://about.bnef.com/insights/data-centers/ai-data-center-build-advances-at-full-speed-five-things-to-know/> >> >> On Thu, Jul 9, 2026 at 10:39 AM James Heilman via Wikimedia-l < >> [email protected]> wrote: >> >>> The question regarding what percent of our traffic comes from search? >>> Similarweb says about 80%, so basically we could lose another 80% of our >>> current traffic. >>> >>> https://www.similarweb.com/website/wikipedia.org/#traffic-sources >>> >>> Is the fall in traffic happening in just big languages? No it is >>> happening in many languages of Wikipedia, at least for health content, >>> There were some exceptions like Persian and Chinese that has seen a small >>> degree of growth >>> >>> >>> https://github.com/nethahussain/Pageviews-in-Medicine/blob/main/analysis/covid_analysis_output.txt >>> >>> J >>> >>> >>> >>> On Thu, Jul 9, 2026 at 10:18 AM Yaroslav Blanter via Wikimedia-l < >>> [email protected]> wrote: >>> >>>> I do not think Google Zero will be (is) experienced by anyone except >>>> for us as a problem. >>>> >>>> Sure, the current AI search is not perfect, but it will be much better >>>> on a horizon of months. It is unfortunate that the end users take the >>>> results uncritically rather than checking them clicking on the source >>>> links, but it is what it is. May be this culture would change, but this >>>> timescale is way way longer than the time to improve the AI search quality. >>>> >>>> Except for obvious financial implications (which are very important >>>> though) it looks like Wikipedia will be mainly the base for training AI. It >>>> is ok, in real life most people do not read encyclopaedias. It is important >>>> to be the training base as well, but we should also realize that if this is >>>> our role then some priorities must be shifted. For example, AI do not care >>>> whether the article looks nice or not, they do not care about usability. >>>> They probably do not care about categories and many other things we spent >>>> years trying to make them perfect. They do care about content though. >>>> >>>> On the other hand, there are clearly things AI can not do, which were >>>> already mentioned here. They can not read offline material, and much of the >>>> paywall material. They can not produce pictures out of nothing - which >>>> means Commons might have a very different role from what it has now. It >>>> might even become the flagship project (unlikely though with the current >>>> Commons community). But all these things require attention, and many of >>>> them will be resisted by the community in the first place. >>>> >>>> Best >>>> Yaroslav >>>> >>>> On Thu, Jul 9, 2026 at 9:33 AM Anders Wennersten via Wikimedia-l < >>>> [email protected]> wrote: >>>> >>>>> I agree we ought to analyze better our situation, before we run to >>>>> conclusions and discuss remedies >>>>> >>>>> Swwp has seen a decrease in traffic like others but in June we have >>>>> seen an increase even if we include only accesses from mobiles (at least >>>>> for June) >>>>> >>>>> >>>>> https://stats.wikimedia.org/#/sv.wikipedia.org/reading/total-page-views/normal|bar|2-year|(agent)~user|monthly >>>>> >>>>> What we have seen more clearly though is that accesses of the most >>>>> read articles per day in decreasing. >>>>> >>>>> We have also more subjectively from media, and social media seen a >>>>> growing skepticism toward AI-generated results and an increase in the >>>>> trust >>>>> of Wikipedia results >>>>> >>>>> Our general conclusions this far are >>>>> >>>>> *We are doing well in our core encyclopedic articles (like of >>>>> administrative divisions within a country) , but less well in newslike >>>>> articles (which the internet readers asks more frequently, but where >>>>> newswebsites always been stronger). And we should concentrate our efforts >>>>> where we are strong >>>>> >>>>> *We are in a late stage in our general product life cycle and must >>>>> accept this. We will never have the increase in editors etc like we hade >>>>> 15 >>>>> yeas ago, but must do what we can do thrive in a this mode. For us - less >>>>> angry internal argument, partially done by avoiding articles of infected >>>>> subjects like Israel-Palestine >>>>> >>>>> And for Google Zero. I believe it is them who has a problem not us. >>>>> Early on, and still, all new PC user were forced into Microsoft Edge and >>>>> Bing which lead us to Firefox and Google. Google zero as I see will lead >>>>> to an exodus från Google, there are many other search engines that works >>>>> fine >>>>> >>>>> Wikipedia is great >>>>> >>>>> Anders >>>>> >>>>> >>>>> >>>>> >>>>> Den 2026-07-09 kl. 08:35, skrev Kimmo Virtanen via Wikimedia-l: >>>>> >>>>> Hi, >>>>> >>>>> Secondly, I'd love to see initiative outside of enwiki. If you look at >>>>>> all the ~1000 Wikimedia projects, almost all of them are seeing an >>>>>> increase >>>>>> in traffic ; only a few big old wikipedias are seeing a decrease (which >>>>>> are >>>>>> also the ones with the most traffic, hence why it's more visible but it >>>>>> obfuscate at the big picture) >>>>> >>>>> >>>>> Based on stats total pageviews of human users of all wikis are dropping >>>>> - >>>>> https://stats.wikimedia.org/#/all-projects/reading/total-page-views/normal|bar|all|(agent)~user|monthly >>>>> >>>>> >>>>> Br, >>>>> -- Kimmo Virtanen, Zache >>>>> >>>>> On Thu, Jul 9, 2026 at 9:13 AM Nicolas VIGNERON via Wikimedia-l < >>>>> [email protected]> wrote: >>>>> >>>>>> Hi y'all, >>>>>> >>>>>> First could someone remind me how much traffic comes from Google (I >>>>>> have a vague memory from last year that this is already very low - maybe >>>>>> 10 >>>>>> or 20 % - and that LLMs are already bringing more traffic to Wikipedia, >>>>>> if >>>>>> true we essentially are at Google Zero already). >>>>>> >>>>>> Secondly, I'd love to see initiative outside of enwiki. If you look >>>>>> at all the ~1000 Wikimedia projects, almost all of them are seeing an >>>>>> increase in traffic ; only a few big old wikipedias are seeing a decrease >>>>>> (which are also the ones with the most traffic, hence why it's more >>>>>> visible >>>>>> but it obfuscate at the big picture). >>>>>> I think we should start by looking at and learning from the >>>>>> "successful" projects instead of focusing on the declining ones. For >>>>>> instance, Wikisources always have been badly to not-at-all referenced by >>>>>> Google and yet there is a strong increase in pageviews recently. >>>>>> >>>>>> Cheers, >>>>>> Nicolas >>>>>> >>>>>> Le jeu. 9 juil. 2026 à 03:38, Michael Snow via Wikimedia-l < >>>>>> [email protected]> a écrit : >>>>>> >>>>>>> On 7/8/2026 12:52 PM, Luis Villa via Wikimedia-l wrote: >>>>>>> > - Scale of impact is important, but it isn’t everything. Sometimes >>>>>>> > we’ll have to build things without knowing if they’ll scale, or >>>>>>> even >>>>>>> > perhaps knowing that they won’t scale (eg Depths of Wikipedia is >>>>>>> one >>>>>>> > of the best community-builders we have even though it has “only” a >>>>>>> few >>>>>>> > million followers). “That’s good but it won’t scale to enwiki” can >>>>>>> > never be allowed to be a blocker for anyone. If the experiment is >>>>>>> > actually great, then we’ll figure out a way to get it onto enwiki. >>>>>>> Or >>>>>>> > not! That shouldn’t stop people. >>>>>>> >>>>>>> At this point, no single initiative (whether community, chapter, or >>>>>>> WMF-led) will have an inherent capacity to scale to enwiki. To do so >>>>>>> would require corporate-style resources with a corresponding >>>>>>> top-down >>>>>>> mandate for implementation, and risk (rightly) triggering resistance >>>>>>> to >>>>>>> the point of launching a true durable fork, for once. Instead, the >>>>>>> best >>>>>>> approach would be to find ideas that are good enough to propagate, >>>>>>> encourage them, and when they have propagated enough, they will >>>>>>> scale >>>>>>> with relatively little intervention needed. >>>>>>> >>>>>>> --Michael Snow >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Wikimedia-l mailing list -- [email protected], >>>>>>> guidelines at: >>>>>>> https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and >>>>>>> https://meta.wikimedia.org/wiki/Wikimedia-l >>>>>>> Public archives at >>>>>>> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/6EQOFSG34E3BC6FIYL55HYBTQBAW4XFN/ >>>>>>> To unsubscribe send an email to >>>>>>> [email protected] >>>>>> >>>>>> _______________________________________________ >>>>>> Wikimedia-l mailing list -- [email protected], >>>>>> guidelines at: >>>>>> https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and >>>>>> https://meta.wikimedia.org/wiki/Wikimedia-l >>>>>> Public archives at >>>>>> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/4WPCSTZNDMBDMYHWHV27BN2EZLGY5LXB/ >>>>>> To unsubscribe send an email to [email protected] >>>>> >>>>> >>>>> _______________________________________________ >>>>> Wikimedia-l mailing list -- [email protected], guidelines >>>>> at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and >>>>> https://meta.wikimedia.org/wiki/Wikimedia-l >>>>> Public archives at >>>>> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/W24PTVEX2A7YSR4YNHIUQTCGMAIMF2S3/ >>>>> To unsubscribe send an email to [email protected] >>>>> >>>>> _______________________________________________ >>>>> Wikimedia-l mailing list -- [email protected], >>>>> guidelines at: >>>>> https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and >>>>> https://meta.wikimedia.org/wiki/Wikimedia-l >>>>> Public archives at >>>>> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/XEKWEBKCXK2V4B3M2JSVFF4E53CWOKI3/ >>>>> To unsubscribe send an email to [email protected] >>>> >>>> _______________________________________________ >>>> Wikimedia-l mailing list -- [email protected], >>>> guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines >>>> and https://meta.wikimedia.org/wiki/Wikimedia-l >>>> Public archives at >>>> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/6N356SKKS2KZEDCV2PFC4GSCERTB7D3E/ >>>> To unsubscribe send an email to [email protected] >>> >>> >>> >>> -- >>> James Heilman >>> MD, CCFP-EM, Wikipedian >>> _______________________________________________ >>> Wikimedia-l mailing list -- [email protected], guidelines >>> at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and >>> https://meta.wikimedia.org/wiki/Wikimedia-l >>> Public archives at >>> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/WC6L3CIEKYWOXNR6FG343RJOX6YGTNRS/ >>> To unsubscribe send an email to [email protected] >> >> >> >> -- >> Ilario Valdelli >> Skype: valdelli >> Tel: +41764821371 >> http://www.wikimedia.ch >> _______________________________________________ >> Wikimedia-l mailing list -- [email protected], guidelines >> at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and >> https://meta.wikimedia.org/wiki/Wikimedia-l >> Public archives at >> https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/HIZMAIRB76D2DSGKNC2IHQJQGGB56HPW/ >> To unsubscribe send an email to [email protected] > > _______________________________________________ > Wikimedia-l mailing list -- [email protected], guidelines > at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and > https://meta.wikimedia.org/wiki/Wikimedia-l > Public archives at > https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/3J6IX424Z27UWSX2PZUPA55OBQY5EU4P/ > To unsubscribe send an email to [email protected]
_______________________________________________ Wikimedia-l mailing list -- [email protected], guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/[email protected]/message/GS2KQ7KDJ6X74TX62QR2RKTPGOYADCDS/ To unsubscribe send an email to [email protected]
