Optional: We can build a strategy about AI that doesn't just center our loss of readers; there are still plenty of humans to write the Encyclopedia and work on diverse global knowledge, but we need to invest in that diversity.

Lately, I have wanted to write a love letter to humans creating knowledge. I wanted to (belatedly) celebrate 25 years of the Wikimedia movement! The evolution of the weird and wonderful activity of editing an encyclopedia, into a global phenomena .

I spent the last 20 years editing English Wikipedia, becoming an admin both here and on Commons, organizing editathons, leading GLAM partnerships, designing campaigns and learning how to batch edit Wikidata. For me, my best friends and greatest collaborators are part of my life because of the Wikimedia movement. Wikimedia took a weird teenager, with a tendency to dive deep into rabbit holes, and gave him a space to grow into a confident, citizen of the world.

...  but  in thinking about this grand and ambitious project to create the sum of all human knowledge that has consumed more than 20 years of my life-- I am extremely sad about the way we are having conversations about the future right now because it underestimates our strengths (i.e. on Wikimedia-l, or the Foundation's Annual Plan) . The conversations assume the  decline of Wikipedia, because of Google’s  and other AI companies seizing our “reading” public.

Wikipedia's readers have always been a happy accident based on the good content by our uniquely organized contributors: we are the weirdest, most wonderful, and argumentative collections of humans interested in curating public knowledge. Our relationship and dependency on Google was beneficial to our growth but doesnt need to be a dependency; we need to be building a story that plays to the strength of humans who want to create knowledge, not the weaknesses of our distribution channels.

Let me try to offer a better vision, than the old cliches we keep falling back on: for example, the annual plan from WMF depends on funneling readership to new editors. This theory has been repeatedly been unevidenced for at least the last 15 years. The "Reader-to-Editor funnel will disappear" framing of the future does not reflect the movement of human knowledge creators we hope to build for the next 25 years, but instead is rehearsing a failing business model from the last 25.

Wikipedia is not alone in experiencing a fundamental shift in the internet. Almost all publishers and non-profit communicators are experiencing an upending of the knowledge economy, shifting traffic away from knowledge producers and their source of income towards unprofitable AI based interfaces. For example, almost all news websites are seeing a decline in traffic.

Commercial, mass market AI models that are taking over the search space are built to serve user needs, tending to be people pleasing, and epistemologically convergent.

Very few of the AI systems attempt neutrality or accuracy or editorial objectiveness very well or have a balanced process for creating knowledge -- but the vast majority of the public will be consuming these non-critically, and consuming it like they did ten years ago for more trustworthy platforms like Wikipedia.  This week Google announced that search is going to be a “customizable app” system designed to trap readers , not simply a path to users interacting with websites. The internet is no longer an internet of readers of websites.

We have entered an era of tools that mimic reliability, while amplifying the editorial decisions of the commercially and bad actor motivated web (and likely are persuading most people to change their opinions and understanding of critical real world topics and giving bad advice that puts vulnerable people like children at risk).

But the substitution for “reliability” is extremely uneven, affecting different topic areas extremely differently, and publishers are building strategies around that. As Luis Villa, a former staff and long time supporter of the Wikimedia Foundation, recently analyzed: our readership decline is also highly topic specific, more severe in certain languages and somewhat slowed by updated content.

The topics that are declining quickest are not surprising:  maths, sciences and computer sciences are all topic areas where our content tends to be too expert, too technical or meant for users already partially familiar with the topic. These also tend to be the topics that AI tools have strong benchmarks and are being optimized to logically navigate.  

Whereas we are seeing readers decline significantly less rapidly on topics that (historically) have been the focus of encyclopedias: humanities, biographies, social sciences and culture more generally.

AI models concentrate citations in a cluster of “top” sources for a domain. For every grouping of prompts that I am tracking for public interest questions about climate and energy, 10 sources are responsible for 1 in 4 of citations, and the top 100 sources for a topic area are responsible for 3 in 5 citations. Wikipedia is always in the top 50 of the sources, usually the top 10.

A recent study by Muck Rack, a PR advisory and services firm, found sector specific evaluations of public interest topics important to journalism regularly include Wikipedia as a top 5 citation alongside other knowledge repositories like the medical Pubmed and the INGO  International Energy Agency.

What does all of this suggest to me? We are experiencing a market correction in our readership stock, where Google was overvaluing us as a click-through destination for readers on a wide range of topics that we probably weren’t ever well positioned to serve (such as general interest explanations of complex math terms; I love you pi, but a video about a pizza probably explains you better).

At the same time, we have more influence than ever: users are relying on our knowledge content in more interfaces than ever. We just aren’t tracking it, and we aren’t going to get paid for that value in the same way.

Market corrections tend to be very good at revealing where companies and economies have strengths, and where their overoptimistic investments weren’t playing out correctly. Clearly we were overindexing the role of reader metrics in our value to the world.

It's very important for us to double down on the thing that isn’t declining nearly as quickly: updating and increasing the relevance of Wikimedia content and finding the gaps in public knowledge that we are uniquely able to serve to these new kinds of interface.

We need to get really serious about how we imagine building content that will be used by AI in the future, and spend less time chasing the kinds of “it's the easiest answer” readers that we got from Google 15 years ago.  In the past we could rely on different kinds of readers, because of our chance relationship to a monopoly. But the long-term decline of readership to other language wikis like Spanish Wikipedia, where auto-translation by Google has substituted for local language content, this was a bit of a devil’s bargain to begin with.

And in the medium-to-long term, Wikipedia and its sister projects is likely not the destination for readers, but a tool for “architecting” the responses of AI tools (both agentic, and directly accessed by humans like chatbots): Wikipedia continues to play an outsized role in grounding the models as a source of citations overall , and by topic/domain. And, more likely than not in influencing the underlying training data for the models, and thus Wikipedia provide a tangible, human centered way of influencing their knowledge. We are more relevant than ever!

However, as a movement, we aren’t adapting quick enough to “see” the impact we are having on the AI tools; instead WMF strategy, and the community conversation are falling back on a motivated reasoning about finding more “readers who will become editors like us” -- doubling down on a past, instead of imagining a different, regenerative future.  

I recently was in a WIkimedia Foundation community call for the Annual Planning process, and board member User:Victoria said something to the effect of “I promise you that there are more editors like us out there”.

To start with: if you talk to anyone who has worked with me for the last decade, I am extremely optimistic about there being 100s of thousands of new contributors in the world. I have spent over half my life advocating for recruiting newcomers (first as a volunteer organizer, and then 11 years working for the Foundation) -- I believe the new editors are there, but as I have argued for most of that time, showing them the edit button has never worked.

However, I also firmly disagree: I don’t think it helps that we all assume there will be more “organic” editors who are “just like us”. The funnel (or at one point the Foundation expanded this to a flywheel which suggested that more editors would lead to more readers) theory of change may feel familiar, and “make sense”, but it tends to lead everyone who needs to make decisions about our strategy down paths of motivated reasoning.

The tendency of the oldest- Wikipedians (like me) is to obsess over recreating detail obsessed editors like themselves who can spend hours of their lives crafting exquisite text, fixing formatting and content errors -- this is unsustainable and not really what is needed from the public in the face of new AI powered tools. Fixing technical problems, and writing smooth text with structure is the main use of large language models. Creating a pipeline of new editors who are basically glorified mechanical turk contributors, while there is a better piece of software to do that work is not a way to create loyal human volunteers.

Volunteering is a privilege of free time; volunteering on information dense content in a world of information overload is a rare personal choice;  and if the assumed pathway to contribution is the same labor being outsourced to AI models in most other parts of our life -- we are narrowing the funnel not growing it.

By the way, at least in high income, English-speaking countries, we probably found all those folks who want to do these kinds of edits in the last 25 years of Wikipedia. Which would explain why the most recent study on successful new editors shows that they are all almost freshly graduated from university young adults..

The tendency of people who try to influence the direction of the Wikimedia Movement who were not originally editors, is to approach it as readers: “If only random phone users read Wikipedia more and could be invited to click the edit button they would stay”.

This is a nice user feature to have, but again this is not sustainable -- the vast majority of humans approached us as passive consumers of Wikipedia for the past 25 years. A shift in phone experiences is not going to suddenly help them become active knowledge creators or critical consumers -- that is exactly why AI tools and short form video have found an audience, convenience and passive consumption. (and if this was the case all the WMF staff and donors aware of Wikipedia editing would be committed editors). We have repeatedly tried this tactic (for example this Donor banner experiment we ran a few years ago) but it doesn't work. I have even advocated for it before: I have been proven wrong.

As someone who regularly dips his toe into New Page Patrol: I am fairly confident we don’t want most of those organic “under 150 edit individuals motivated to create new articles” users to be our next generation of editors anyway. Many of them are clearly here to promote commercial topics such as business and promptional biographies, the same commercialized spam that dominates the rest of the internet, and is tightly paired to the Audience builder type persona identified by the new editors research which tends to have bad faith intentions, and not “stay” on the wikis in constructive way.

If you talk to retained new contributors, and listen carefully to their stories: it's not that “readers will become editors” but more “the people that make good Wikipedians, also enjoy consuming long-form written content like good Wikipedia articles”. Research like the Journey Transitions study by the Foundation, reinforce this: editors see a world of opportunity in our content. As a former Wikimedia Board chair, Christophe Henner said: we created something truly unique in our human-facilitated process for finding consensus across competing intellectual traditions. Becoming a Wikipedia editors is not a very “normal” way to spend your down time.

What we need is a powerful diversity of tactics leveraging social context to gather people still invested in the human centered exploration of knowledge and give them a chance to join our community. We need a diverse, optimistic investment in reaching people already spending most of their lives engaged in intellectual curiosity.

The tendency is to say “but what about the overlap of readers and donors?”: sure the ratio of potential donors from banners actual donations changes, and does make the work more complicated. But the Fundraising team has been more successful than ever at raising money: in part because they understand how to get the most from our banners, but also because they have been working to decouple fundraising from readership over the last 5 years, and it seems to be working!

Let’s decouple the “editor” and editing strategy from readership as well!

The funnel theory of change is best understood through metaphors: we are trying to persuade individuals that see Wikipedia as a utility to be consumed like a service, that they too possess some internal desire to contribute knowledge; advocating for this theory of change, is like advocating for bottom trawling to catch a rare 1 in 10,000 rare underwater snail, sure it works because you caught everything at the bottom of the ocean, but you end up killing other species, destroying the ecosystem, and throwing out some of the snails with the other bycatch.

In order to catch the editors we need for the future: what we need is the equivalent of lobster traps: species targeted, persuasive systems that get filtered by organizers and experienced editors for the healthiest lobsters we want to keep (though please don’t eat the new editors for dinner or serve them to the Cabal). We need each kind of potential group of editors to feel invited, with specialized support for their motivation to act.

Here are just a few of the signals that we could be letting guide bold, evidence based ways to invest in decoupling newcomers from readership:

• Editing doesn’t need to start with Wikipedia, and is sometimes more interesting elsewhere:
  • Wikidata and Commons continue to increase registrations despite never having been good sources of traffic.
  • Efforts like WikiCite or Coordinate Me consistently bring in high quality contributions to Wikidata and Commons from both external partner networks and existing Wikimedians.
  • Some of our most successful global south programs are on Commons (ever heard of Wiki Loves Monuments or Africa or Earth or Folklore?), Wikidata, Wiktionary and Wikisource, especially in new audiences learning how to contribut to the internet for the first time (i.e. Lexemes in Ghana or WikiSource in India)

• Catching new audiences in classrooms and other places of curiosity work better but requires professional investment
  • Wiki Educations’ program has been the most successful at scale intervention for creating high quality content used by readers: by training students in classrooms to fill knowledge gaps.
  • Reading Wikipedia in the Classroom has been the most successful at scale intervention have seen for increasing readership and editors in smaller language Wikipedias
  • Basque Wikipedia’s education program has consistently demonstrated a successful audience-specific increase in traffic, because its used in schools.
  • Most of us who joined the movement during the early 00s and 10s were aware that “anyone could edit” Wikipedia because the news, faculty and teachers argued with us about how reliable Wikipedia was. You see this origin story a lot everywhere retained editors share their story, like the Clovermoss editor reflections survey.
  • Encyclopedia Britannica has even pivoted their entire business model towards educational environments to create a financially viable future. We stopped comparing ourselves to Britannica a long time ago, but I think there is something to learn from their business strategy.
• Catching knowledge stewards and communicators through their professions  addresses movement’ strategy goals of knowledge equity and topics for impact
  • GLAM partnerships have consistently filled gaps in our visual and Wikidata content pipelines.
  • The continued expanse of Wikidata has been in large part due to well-targeted academic and library friendly outreach programs and library funded data projects.
  • Wikimedia UK and other European affiliates have consistently supported Wikipedians in Residence, often funded by external partners and ally organizations, that produce regular volumes of wikipedia pages, editors and filled key critical knowledge gaps.
• Persuading people in their own context matters, and only affiliates have capacity for that (even if some affiliates are ineffective)
  • Wikimedia Nigeria’s advertising campaigns as part of an early pilot on reader awareness -- did significantly move the needle on reader awareness
• Editor decline is more closely coupled with on-wiki cultural problems than readers
  • That Spanish Wikipedia has had an editor and reader decline long before this current boom, and it appears to have been attached to lower quality content leading Google to auto translate content from English Wikipedia. Why is there lower quality content? One good theory is that editors on the wiki report really high feelings of harassment.

If we were to take all of this evidence together, quite the opposite of a funnel seems to drive the success of individual wikis and content editors: context specific recruitment, higher quality newcomers and culturally dynamic on-wiki communities receiving new editors seem to be some of the strongest signals we have of creation of content and subsequent traffic.

Having spent more than a decade at the Foundation, most staff want to build this kind of future as well and are more than capable of doing it. However, the majority of the individual staff are not feeling represented and forming a Union in order to be more involved in organizational decision making processes (please express solidarity).

Moreover, I have been hearing disappointing reports from attendees to closest thing we have had to a movement-wide conversation about the future, The Futures Lab, which only had ~100 community members. If the world is rapidly changing, how are we supposed to spawn innovation at the scale we need to facilitate our diversity with a conversation that only included 100 people?

The problem is that the overall theory of change is premised on this idea that “investing in readers and the platform” is an “investment in editors and donors”.  This “funnel from reader to engaged reader to editors” may seem like a “common sense” thing but I have yet to see an intervention or documented example of a change in reader or registration experience rippling through to “editorship” -- and WMF’s own analysis of the registration decline couldn’t connect it with interface changes for example.

Even if we end up in a world where humans are not the main consumers and curators of encyclopedic style content:  we need the Wikimedia Foundation to support more diverse and potential futures where we are embracing the strength of human curiosity, and imagine a future that includes us as editors. We could be asking questions like:  What does vibe curating of knowledge look like? How do we empower the most inspired humans to fill the quirkiest gaps of the “sum of all human knowledge”, while not trying to compete with AI?

And if we are so concerned about AI for readers: the data highlighted by the foundation so far about AI has only been one of fear of overwhelming the servers -- again why are we so focused on the “reader/consumer metric” as a leading indicator of a threat?

If we sit back and look at our 25 year history: the reader decline is probably a natural correction in Google sending us searchers. Most of these searchers weren’t intending to find a long-form essay on a deep topic (and as User:HaeB recently pointed out in Wikimedia and AI Telegram, Google is responding to a different kind of user that our form of content is not good for serving). That the world wants to consume us in new formats, should be a celebration of opportunities -- after all, how do we expect to reach all humans in their own language?

If we are serious about the “sum of all human knowledge” center of our mission, there is ample signals that could enable editors and readers to benefit more from Wikimedia in the context of AI:

• AI Search is creating a lot of signal, that the Wikimedia Foundation could be giving that to the editors: for example, AI Query fanout, for instance is driving a huge chunk of citations and traffic, and AI generated synthetic queries are some of the easiest data to filter out of any Google Search Console data (4-8 entity strings, that are clearly not written by humans and structured to elicit subtopics matched against user intent)
• Google and Bing are both including signals about AI citations in their analytics data (I am an advisor at Appropedia.org and this has significantly changed our perspective on where they are providing impact for internet users)
• We could be building better partnerships with companies which provide more researcher focused AI research tools  to make the research part of writing a Wikipedia article easier (as has been piloted by Wikimedia Switzerland and Poland)
• When we negotiate contracts with AI companies for Enterprise, there has to be something they could be giving us to super charge editors filling knowledge gaps with low quality results, like we tried to do with Google search data multiple times.

We need deep editors interested in reorganizing public knowledge, with ambitious public knowledge goals, and they need the infrastructure to help them do that.

We probably don’t need to recruit new editors focused only on shallow editing, primarily fixing typos or our backlog -- if we provide good creative spaces for our existing editing community to build their own workflows with AI assisted programming and classification tools. I am sure LLM assisted writing tools and workflows should address these detail oriented problems. In the last few months, AI has been helping me rethink my editing priorities, and my way of “seeing” how to contribute to the movement.

As someone working now in climate and environmental communication, it makes me really sad that the Foundation’s metaphor for its annual plan pivots on a single deep rooted tree that “goes it alone through a drought”: the shepherd's tree (Boscia albitrunca). They are using one of the worst cliche’s in well meaning, but ineffective environmental communications: the single charismatic anchor species “going it alone” which is good for fundraising, but terrible for ecological conservation. Species depend on ecosystems. Sending roots deeper into the soil  doesn’t work in dryland restoration: the groundwater table is retreating and climate change means that it doesn’t have a chance of regenerating in at least the next 1000 years, unless humans intervene to harvest water on the whole landscape.

From an environmental perspective, a single tree is doomed to fail: we are in a world where all of our environmental systems are on the edge of collapse without more holistic thinking-- and the digital ecosystem is no different. The most concrete thing we can do is invest in diverse, multispecies ecosystems, with different tactics for survival, so that we can weather climate change and human-created water bankruptcy which are drying up aquifers -- i.e. the AI extraction that we need to endure. One tree tends to live for a lifetime, doesn’t recharge the water in the soil and can easily die under stress, whereas an ecosystem can be helped to regenerate new and innovative approaches to survival.

Like libraries, schools, museums and other institutions that have lasted more than 100 years because they have adapted their strategies as intergenerational stewards of the knowledge the world needs, we need to reimagine our ecosystem and do it intentionally. We need a regenerative Wikimedia strategy: investing in building tree plantations, and propagating other kinds of life, and re-architecting the water flows so when it does rain we catch it all. What we need is to imagine a knowledge commons 25 years from now, and double down on our strengths: empowering the humans who want to curate human knowledge.

I look forward to contributing to an encyclopedia and all of the seedlings around it for the next 25 years of my life. Let's keep doing it, but let's be more intentional about how we cultivate the knowledge stewardship we need for that future, rather than live in fear of our lost past!