Seconding Galder in bringing this important research finding to wider attention (CCing the Wiki-research list as well).
If I may, we also just covered this in the new Signpost with some additional context and detail that some here might find interesting: "Wikimedia Foundation reports 8% traffic drop since last year due to 'the impact of generative AI and social media'" https://en.wikipedia.org/wiki/Wikipedia:Wikipedia_Signpost/2025-10-20/In_the_media#Wikimedia_Foundation_reports_8%25_traffic_drop_since_last_year_due_to_%22the_impact_of_generative_AI_and_social_media%22 .[1]
As mentioned there, the problem of quantifying the impact of generative AI on Wikipedia readership has already attracted considerable attention by academic researchers in the almost three years since ChatGPT launched. I even think it's safe to say that many might regard it as one of the most important research questions about Wikipedia and AI currently.[2]
However, the academic research publications about this - admittedly difficult - topic have been a bit of a mixed bag so far https://meta.wikimedia.org/wiki/Research:Newsletter/2025/March#So_again,_what_has_the_impact_of_ChatGPT_really_been?. (Even what I would regard as the best paper among these, which Mako and I also chose to highlight in our annual "State of Wikimedia Research 2024-2025" talk at Wikimania this year, still has some potential statistical shortcomings https://en.wikipedia.org/wiki/Wikipedia_talk:Wikipedia_Signpost/2025-03-22/Recent_research#%22Wikipedia_Contributions_in_the_Wake_of_ChatGPT%22. And I am not aware of much peer-reviewed insight that goes beyond the early-years ChatGPT.)
So it's great to see that the Wikimedia Foundation has now itself jumped into this breach. It is evidently well positioned to do so. Not just because it has potentially useful internal data that was not available to the aforementioned external researchers (say referrer information https://wikitech.wikimedia.org/wiki/Data_Platform/Data_Lake/Traffic/Pageview_hourly, or now also the data derived from the unique reader cookies WMF introduced earlier this year, which enable better but still relatively privacy-friendly tracking of Wikipedia readers). But also e.g. because of the foundational work https://research.wikimedia.org/foundational.html its own researchers already did some years ago on "Detecting and Gauging Impact on Wikipedia Page Views" (IMO a paper worth looking at for anyone investigating such questions).
Unfortunately though, Marshall Miller's Diff post doesn't contain any concrete information about the methodology used to causally attribute the drop to those two factors, nor any links to further details. Can we expect the underlying statistical analysis to be published soon? It would also be interesting to know e.g. each factor's share in this 8% drop (is it say 7% from AI vs. 1% from social media, or vice versa?) Or, to circle back a little towards Galder's points, whether different Wikipedia languages were affected differently, etc.
Regards, Tilman
[1[ Alongside a somewhat related story https://en.wikipedia.org/wiki/Wikipedia:Wikipedia_Signpost/2025-10-20/In_the_media#Elon_Musk_announces_AI-based_%22Grokipedia%22_to_challenge_Wikipedia about a certain well-known US entrepreneur trying to replace Wikipedia with a new AI-based encyclopedia, also mentioning some similar but less well-known efforts. [2] For some reason though, this central research question didn't make it into the list https://meta.wikimedia.org/wiki/Artificial_intelligence/Bellagio_2024 of research directions about the implications of AI for the knowledge commons that was drafted at the February 2024 Bellagio symposium convened by the Wikimedia Foundation. But as we heard recently on this mailing list https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org/thread/YI5KUYTCCYTTEISGM3APGXYMQUVFAZNM/#5CPAYECG7XQ5NXVUJ3VODJGLWBFV2NBS, the WMF research department is still working on getting a fuller documentation of that event's outcomes ready for publication. (At that point we will hopefully also see an attempt to solicit wider input on this draft agenda, from other researchers and the editor community, the latter appearing particularly relevant as the project's stated purpose is to convey "vital questions volunteer contributors have raised"). And in any case, it or some of its colleagues have evidently now tackled this research question themselves.
On Mon, Oct 20, 2025 at 11:46 PM Galder Gonzalez Larrañaga < galder158@hotmail.com> wrote:
Dear wikimedians, I recommend reading the very interesting diff post by Marshall Miller about new user trends on Wikipedia: https://diff.wikimedia.org/2025/10/17/new-user-trends-on-wikipedia/. The post talks about two main trends affecting our views: AI summaries and video consumption by new generations. We don't have a good solution about the first one, but I would like to talk about the second one, because in the Basque Wikimedians User Group we have been working on this for some years now, and I think our experience can be part of the solution.
Five years ago we detected this trend and saw that, while there was a rise on educative/informative videos for learning, those videos were mainly done in English or hegemonic languages. Students and teachers were using videos more than ever, but those videos weren't free nor in Basque. That's why we created the platform Ikusgela on wiki and external media channels (Wiki platform: https://eu.wikipedia.org/wiki/Atari:Hezkuntza/Ikusgela, other links to social media and video platforms there)
After four years creating videos we have nearly 250 free educative videos in Basque, many of those subtitled in other languages. We cover topics from philosophy to evolution, from literature to basic science. And now we are publishing an average of 2 new videos per week. We are now working on two new series about migrations/rights and linguistics. The videos have received the highest award in the Basque Country for communications and are used now in education by teachers in many schools and MOOCs. More than 540 articles in Wikipedia have a video from Ikusgela available, with many more to come this school year.
Making good quality videos is expensive, and we are making this in our own with our funding and from competitive grants. However, once the videos are made, remaking those in other languages should be cheaper. Building the videos together is also cheaper than doing it alone. If your chapter or user group is interested on that, let me know and we can make things together.
We, Basque Wikimedians, saw how this trend was coming and worked to take advantage from it, instead of seeing how it decreases our relevancy. I hope efforts from the WMF and other affiliates can go in the direction of multimedia soon.
Best,
Galder
Wikimedia-l mailing list -- wikimedia-l@lists.wikimedia.org, 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/wikimedia-l@lists.wikimedia.org/... To unsubscribe send an email to wikimedia-l-leave@lists.wikimedia.org