The Wikimedia Foundation (WMF), the non‑profit organisation behind Wikipedia, has taken a firm stand: it is asking artificial‑intelligence (AI) firms to stop scraping Wikipedia’s content indiscriminately and instead access it via its paid product, Wikimedia Enterprise. According to the WMF’s published financial data for fiscal year 2023‑24, total revenue reached approximately USD 185.4 million and total expenses around USD 178.5 million. The move strikes at the heart of how knowledge is shared in the era of large‑language models and raises questions about sustainability, fairness and the shifting economics of open content. In this article we analyse the underlying causes of this shift, its implications for platforms, content creators and AI firms, and explore what may follow
Earlier in November 2025 WMF published a blog post and followed with media communications urging AI companies to stop automated scraping of Wikipedia pages and instead use Wikimedia Enterprise’s paid API. The organisation highlighted that in recent months there has been a surge in bot traffic from what appear to be AI‑systems mimicking human users, particularly in May and June 2025. This automated access is placing strain on servers and possibly undermining the model of how Wikipedia sustains itself. WMF emphasises that it is not currently threatening legal action against all scrapers, but seeks a transition to “responsible” large‑scale reuse through its enterprise product
This development is significant for three key reasons
1. Economic sustainability: Wikipedia’s funding model has traditionally depended on donations and the assumption that high traffic equates to donor willingness and volunteer engagement. With AI systems increasingly embedding Wikipedia content in responses often without requiring users to click through to Wikipedia WMF warns the “traffic‑flywheel” is under threat
2. Infrastructure and cost pressure: The spike in automated access has both increased bandwidth and server loads, especially from requests that appear human but are from bots. WMF positions this as a cost and operational risk
3. Content licence, attribution and fairness: While Wikipedia content is under open licences (e.g., CC‑BY‑SA), the business of large‑scale AI training using that content raises questions about reciprocity and attribution. WMF’s request signals that even open content ecosystems may require new business models in an AI‑driven world
Wikipedia was founded on the principle of free, open access to human‑curated knowledge, sustained by volunteers and donations. Over the past decade the Wikimedia ecosystem has grown significantly in both cost and scale. According to WMF’s FAQ, for FY 2023‑24 total revenue was USD 185.4 million, total expenses USD 178.5 million Meanwhile, the AI training ecosystem has charted rapid growth. Large‑language models rely on massive datasets, and Wikipedia has long been a key source due to its structured, multilingual, human‑edited content. But the invitation to reuse at no cost has now collided with the reality that users may not visit Wikipedia at all if AI systems answer questions directly. Indeed, reports suggest Wikipedia page‑views by humans declined about 8% year‑on‑year, with much of the traffic shift being due to bots rather than genuine human users In response, WMF launched Wikimedia Enterprise in March 2021. The goal was to create a commercial channel for high‑volume re‑use of Wikimedia content especially by entities requiring scale, guarantee of service, attribution, and perhaps commercial terms. As of FY 2023‑24 the revenue from Wikimedia Enterprise was only about USD 3.4 million, representing 1.8% of the total revenue of the foundation. Thus, the present demand should be seen as part of a longer‑term strategic shift: from pure donation funding and volunteer‑based traffic, toward diversified revenue streams in an era where open output is increasingly leveraged by commercial AI systems.
On one hand, WMF’s demand is understandable. The non‑profit warns that free, un‑regulated scraping by AI firms is: (a) increasing operational costs, (b) reducing the incentive for human volunteers since visitors (and potential donations) decline, and (c) weakening the open‑knowledge ecosystem by enabling others to derive commercial value without contributing back
From the perspective of AI firms, some counter‑arguments can be made. First, Wikipedia operates under permissive licences precisely to allow broad reuse without barriers; demanding payment for reuse seems tension‑filled with the open‑knowledge ethos. Second, in many cases AI models use Wikipedia content in aggregate rather than serving verbatim pages the relationship between “scraping” and “model training” is more complex than simple copying. Third, smaller developers may find paid access expensive or impractical, risking a barrier to innovation.
Of deeper concern is the question of precedent: If Wikipedia secures payment or commercial terms for reuse of its content, will other open‑content platforms follow? Will this shift fuel a “paywall” of open data when used commercially? The broader philosophical question is: what does “free knowledge” mean when commercial AI models derive revenue using that knowledge
Furthermore, WMF’s own metrics show that Wikimedia Enterprise is a small part of the revenue, so this step may be as much about signalling value and deterrence as about immediate commercial uptake. The 8% drop in human page views may be interpreted as an early warning but the actual scale of impact on revenues or volunteers is still speculative.
What about the smaller players and volunteer editors ?
Much of the commentary focuses on large AI firms and Wikipedia but a less‑explored angle is how this shift impacts smaller innovation players and volunteer editors. If Wikipedia content becomes monetised when used at scale, smaller companies or independent researchers may be forced into commercial agreements or shut out. That could reduce experimentation and diversity in the AI ecosystem For volunteer editors, the visible signals matter: fewer human viewers may mean fewer donations and fewer new contributors. When Wikipedia states that “open content is free, infrastructure is not,” it hints at a longer‑term risk: the volunteer model may fray if contributors feel their work is simply being commercialised without reciprocal benefit.
What does this mean for the AI ecosystem ?
AI firms need to take note: when using large‑scale human‑edited content like Wikipedia, they may face operational and reputational costs. Becoming a “consumer” of scraped content may increasingly shift toward a “subscriber” or licence-holder model. This could influence training datasets, partnerships with open‑knowledge platforms, and perhaps licence‑compliance monitoring. But, this may accelerate an industry trend where content owners open or otherwise seek revenue from the AI models that are consuming or relying on their content. This could lead to a stratified data‑economy where models trained on publicly‑funded volunteer content still require payments or attribution, changing assumptions about “free for training
Looking ahead, a number of scenarios are plausible
Wikimedia may begin enforcing stricter access controls, limiting scraping or filtering bot traffic, or shifting more aggressively to the paid Enterprise model.
AI firms might negotiate bulk access deals, or new open‑licence arrangements may emerge that include commercial‑use fees, attribution and royalty‑type models.
Smaller data‑providers and volunteer‑based platforms may follow Wikipedia’s lead and demand compensation introducing more complexity into data sourcing.
Volunteer participation on Wikipedia could decline if traffic and donor engagement drops; conversely, WMF may invest more in infrastructure, moderation and community support to compensate
The dynamic between “open knowledge” and “commercial AI” may become a battleground of policy, ethics and business models raising questions about who gets value when volunteered content is used commercially
Wikipedia’s public request that AI firms stop free scraping and instead use a paid access channel represents a watershed moment for open‑knowledge infrastructures. The financial figures speak for themselves: Wikipedia’s revenue for 2023‑24 was roughly USD 185.4 million, expenses about USD 178.5 million. What happened? A surge in bot‑generated traffic and declining genuine human visitors prompted the Wikimedia Foundation to recalibrate its stance toward large‑scale commercial reuse. Why is it happening? Because the assumptions underpinning free access (volunteers, visitors, donations) are under threat in the AI era. What does it mean? It signals that even “free” content platforms may need new business models when commercial AI systems build on them — and that the AI ecosystem’s reliance on openly volunteered content may come with hidden costs. What comes next? The next 12–24 months may see new commercial licences, evolving data‑access models, and possibly policy debates on how open‑knowledge and AI commerce coexist