Wikipedia's 8% drop in human readers tests what AI companies will pay for an open encyclopedia

If bots had been inflating the old counts, the size of the drop depends on whether the 2024 months Miller compared were re-counted with the new filter ^[4][5]. The foundation's reading is that a real decline had been masked ^[5]. Before treating 8% as the size of the AI effect, I'd want the same filter run over both years. It is also an old figure. Meehan, who raised it with Axios ^[1], started on January 20, 2026 ^[8], 95 days after Miller's post first put it in public ^[2].

Wikipedia lives on donations and volunteer labor, and both depend on people reaching the site ^[3]. Set the 50% rise in media-download bandwidth ^[6] against the 8% fall in human views ^[4]. Dividing 1.50 by 0.92 gives about 1.63, so media bandwidth per human page view is up roughly 63% ^[1]. The two measurement windows do not match exactly, so treat that as a rough size.

Meehan's request for resources ^[9] has one product behind it. Wikimedia Enterprise sells structured data feeds to companies that ingest Wikipedia at scale, while individuals keep reading and editing free ^[11]. She has kept the site open, with no paywall for ordinary readers ^[10]. So the buyer pays for cleaner, more reliable access to content it can still reach ^[11]. In my view a licence the buyer can route around is priced like a convenience fee, capped at what scraping costs the scraper. The 50% bandwidth rise ^[6] suggests many large users have so far found scraping cheaper.

This can go three ways. AI companies buy feeds because a supported product beats a crawler on quality and reliability ^[11]. Or they keep scraping and donors keep covering the bandwidth ^[3][6]. Or a consistently filtered count shows a human decline smaller than 8%, leaving the case for payment resting on infrastructure cost ^[5][6]. I'd put the most weight on the second, because the free route stays open under the foundation's own policy ^[10]. That view is wrong if Enterprise starts naming AI companies as paying customers at scale.

For anyone valuing AI companies, the dependency is concentrated. More than 90% of Wikipedia's visits have historically come through Google search ^[7]. The answers on search pages and in chatbots that remove the reason to click are often built on Wikipedia material ^[13]. The company with the most say over Wikipedia's human traffic is the one that has sent it more than nine visits in ten ^[7]. By Meehan's account, the AI companies she is pointing at use the content without proper credit or payment ^[2]. Crypto Briefing's report does not include Enterprise's revenue, customers or prices, so there is no licence figure yet to put into anyone's valuation.