Heavily AI-Flagged Pages Still Rank Across Google’s Top 10 via @sejournal, @MattGSouthern

AI-heavy pages turn up at every spot in Google’s top 10. But pages the detector reads as mostly AI-written sit lower in those results than pages with little AI text, and fewer of them were found in the index at all, according to new data.

The scores come from Ahrefs’ own AI detector, which the company sells through Site Audit and Site Explorer. It says the tool isn’t perfect and doesn’t work the same way as anything Google might use. Every number here is an estimate of how much of a page reads as AI-written, not a record of who wrote it.

Since 2023, Google has said that it rewards quality no matter how a page was made, and that automation aimed mainly at gaming rankings counts as spam.

More than half of the pages in Google’s top three search results (54.7%) have less than 20% AI-generated content. So, next time someone says that most top search results are mostly written by machines, you can keep that number in mind.

Where AI-Heavy Pages Rank

Pages scoring under 50% AI text took 82.2% of the top three spots. Pages the detector scored at a full 100% were rare but real, at 5.3% of top-three pages.

Pages scoring 80% or higher accounted for 8.4% of first-place results and 11.7% of tenth-place results. The average score rose from 27.1% at position 1 to 30.9% at position 10, although the increase wasn’t linear, since position 9 scored higher than position 10.

These statistics are based on about 150,000 pages from the top 10 results of 100,000 searches in June 2026, each at least 350 words long. The headline count from Ahrefs lists 331,000 pages, representing several different samples.

Ahrefs Found Fewer AI-Heavy Pages In The Index

Ahrefs found 49.28% of pages with the least AI text in Google’s index. This percentage drops to 43.38% for pages scoring between 20% and 50%, 40.72% for those between 50% and 80%, and 40.35% for pages at 80% or higher.

Higher AI Scores, Fewer Pages Found In The Index

Share of roughly 100,000 pages meeting at least one of Ahrefs’ three index signals, grouped by the estimated share of text flagged as AI-generated.

The report’s authors describe this variation as “meaningful but far from disqualifying.” Since Google doesn’t give direct index access, Ahrefs measured this in three ways: whether a page ranked for at least one keyword, had a Search Console impression since January 2026, or appeared in a site: search.

Any one of these means the page is almost certainly indexed. However, pages may still be indexed even if they don’t meet these criteria, so these figures are minimum estimates. That check used a separate sample of roughly 100,000 pages, one per domain.

What Changed Since Ahrefs’ 2025 Data

A 2025 version ran across 600,000 pages, looking at the top 20 instead of the top 10. It put the correlation between AI score and ranking position at 0.011, effectively zero, and said there was no clear relationship between the two. SEJ covered it when it ran.

The new one says higher AI use lines up with lower positions. It gives no such number, and still calls the difference gentle. The two used different samples, went different depths into the results, sorted pages into different groups and published different measures, so they don’t line up cleanly. The clearest change is the wording.

Why This Matters

Nothing here backs up holding off on AI help for fear that Google will spot the output and drop it. Pages the detector scored at a full 100% held top-three spots in this sample.

Higher scores lined up with lower rankings in one sample and fewer indexed pages in another. Report co-author Ryan Law reads this as quality declining as AI use rises.

These samples didn’t evaluate factors like content depth, internal linking, originality, or factual accuracy.

A high score is a reason to look at a page. It isn’t proof the page is weak, or that Google did anything to it.

Looking Ahead

This measures how much of a page reads as machine-written, just as AI help turns up inside everyday writing tools. The more that spreads, the less a detector score may tell you about how a page was actually made.

The bigger questions, the ones SEJ contributor Dan Taylor has been asking about, are around publishing at scale and whether scaled AI content is worth the crawl budget it uses.

Featured Image: Cast of Thousands/Shutterstock