English Wikipedia Got 8.1% Fewer Page Views in the Past 12 Months Than in 2022, and 14.8% Fewer Than the Year Before AI Overviews Launched

Summary

Picture a plumber in Ohio who writes a solid guide on how tankless water heaters work. Two years ago it brought in a steady trickle of visitors from Google. Now the same question gets answered at the top of the results page, or inside a chatbot, and the click never happens. Wikipedia is the biggest reference site on the web, so it is a useful gauge for how much of that reading has moved elsewhere.

We added up every English Wikipedia article view from September 2021 through September 2026. In the 12 months ending September 2026, English Wikipedia recorded 79.24 billion page views. That is 8.1% below its 2022 total of 86.22 billion and 14.8% below the 93.00 billion it recorded in the year before AI Overviews reached US searchers.

The drop was not immediate after ChatGPT launched in November 2022. Readership actually rose in 2023 and held through 2024. The decline shows up from 2025 onward, after AI answers became a default part of Google search.

What we measured

We used Wikimedia's public hourly pageview logs, loaded in Google BigQuery, and rolled them up to monthly totals for the English desktop site (en) and the English mobile web site (en.m). Page views were summed across all articles, excluding the Main Page and non-article pages such as Special, User, Talk and File pages.

We compared four windows: calendar year 2022 (the pre-AI baseline), the 12 months before ChatGPT launched, the 12 months before AI Overviews launched in the US, and the latest 12 complete months (October 2025 to September 2026). We also compared January to September of 2022 and 2026 so that the same months are set side by side.

The past 12 months were the lowest of any comparable period in the data

Annual totals show a rise in 2023 and 2024, then a fall. The latest 12 months are the lowest 12-month total in the five years we measured.

The 2023 and 2024 rise came almost entirely from mobile web views. Desktop views were lower in every year after 2022. Part of the mobile bump in 2023 and 2024 is likely automated traffic, which Wikimedia filtered more aggressively from 2025, so the 2022 baseline is the cleaner comparison.

Since AI Overviews launched, views are down by about one in seven

Measured against the two AI milestones, the post-AI Overviews period shows the larger gap.

The timing lines up with AI answers moving into the search results page itself. The data cannot prove that AI answers caused the drop. Changes in bot filtering, search features and reading habits all overlap with the same period.

Desktop readers fell a bit faster than mobile readers

Both the desktop and mobile sites lost readers between 2022 and the latest year.

Desktop is where people do most of their work research and where AI chat tools see heavy use. Mobile web still carries close to two thirds of English Wikipedia reading, at 64.4% of views in the past year.

What this means

If the largest reference site on the web is getting 8.1% fewer views than in 2022, a small business blog that answers the same kind of factual question should expect the same pressure, and probably more. Explainer pages that once pulled first-time visitors now compete with an answer that shows up before anyone clicks. Any AI SEO agency looking at a small business blog today will usually find the explainer pages losing ground first.

The practical move is to make sure your business is the source those AI answers draw on and name. That means clear, specific pages about your services, prices and service area, written so a machine can quote them. Before you rewrite anything, run a free AI visibility audit to see what Google's summaries say about you now, which is where any work on SEO for AI Overviews should begin.

Limits

Wikimedia pageview counts include some automated traffic that its filters do not catch, and Wikimedia changed how it detects bots in 2025, which lowers counted views from that point on. Some of the change in any period may reflect better bot filtering rather than fewer human readers.

We count views of titles that had at least 5 views in a given month on a given site, which drops a very small tail of rarely read pages.

English Wikipedia is read worldwide, so these totals are not US-only.

Mobile app views are not included; only desktop and mobile web sites are counted.

The data shows timing, not cause. AI answers, search design changes and bot filtering all changed during the same years.

Frequently asked questions

Related studies

• Wikipedia readership in the AI answer era'What Is' Wikipedia Pages Lost a Median 41.9% of Readers
• Wikipedia readership in the AI answer eraProgramming Pages Lost a Median 57.9% of Wikipedia Readers
• Wikipedia readership in the AI answer eraSpanish Wikipedia Views Fell 36.3%; English Fell 8.1%
• Wikipedia readership in the AI answer eraPopular Wikipedia Pages Lost 24.5%; Niche Pages Gained 12.2%