Ahrefs 600,000-page study: AI content and Google rankings

Most claims about whether Google punishes AI-written pages rest on anecdotes. Ahrefs published one of the larger measurements: a sample of 600,000 top-ranking pages, each scored for how much of its text a detector flags as AI-generated. This write-up reports that study by Si Quan Ong, published on 7 July 2025. SEO Madman did not run it.

How did Ahrefs measure AI content in Google results?#

Ahrefs drew 100,000 random keywords from Keywords Explorer and collected the top 20 URLs for each, giving 600,000 pages. Each page was scored with Ahrefs’ own detector, part of Page Inspect in Site Explorer, which reports the share of the page it judges to be AI-generated. Pages were then grouped into pure AI, pure human and mixed, and the AI share was correlated with ranking position.

What did the study find?#

Mixed content dominated: 81.9% of pages, against 4.6% pure AI and 13.5% pure human. Inside the mixed group, 13.8% showed minimal AI use (1 to 10%), 40% moderate use (11 to 40%), 20.3% substantial use (41 to 70%) and 7.8% dominant use (71 to 99%). The correlation between AI share and position was 0.011, which the study treats as no relationship. Position one pages carried slightly less AI content than lower positions, a pattern Ahrefs itself calls very weak.

The authors’ reading is that Google probably does not care how content was made, only whether searchers find it helpful. They also cite a separate Ahrefs survey in which 87% of 879 marketers said they use AI to help create content.

What does Google say about AI-generated content?#

The study lines up with Google’s own position. In a February 2023 Search Central post, Google wrote that “appropriate use of AI or automation is not against our guidelines.” The same post draws the line at intent: using automation, including AI, to generate content with the primary purpose of manipulating rankings violates its spam policies.

That policy now has a name. Google’s spam policies define scaled content abuse as many pages generated for the primary purpose of manipulating rankings and not helping users, and list “using generative AI tools or other similar tools to generate many pages without adding value for users” as an example. The policy applies “no matter how it’s created,” so the test is value and scale, not the tool. This site tracks enforcement in the spam policy enforcement report.

How far can these results be trusted?#

The figures are only as good as the detector. Ahrefs states that no detector is perfect and that false positives are possible, so the shares are estimates. The data also covers only pages that already rank, so it cannot show how many AI-heavy pages failed to rank or were removed for spam. Treat it as evidence against a blanket AI penalty, not proof that AI content performs well.

What should a site owner check?#

Google’s helpful content guidance asks two questions that map directly onto this study. Are you using extensive automation to produce content on many topics? And is the use of automation self-evident to visitors, through disclosures or otherwise? A page that passes both is in the territory the Ahrefs data describes, where AI share and position show no link.

In practice, review AI-assisted pages for what a detector cannot score: first-hand detail, accurate facts and a reason for the page to exist beyond a keyword. Word count is not the lever either; Google says it has no preferred length, and this site’s content length correlation write-up covers what the data shows there.