Why Your AI Content Isn’t Ranking
If you have been publishing AI-assisted content and wondering why it hasn’t moved, there’s a widely repeated statistic that has probably reassured you, and a newer one that complicates it considerably.
The Number Everyone Quotes
Ahrefs studied 600,000 pages and found that 86.5% of top-ranking pages contained some AI-generated content, with a correlation between AI content and ranking position of 0.011 — effectively zero.
That figure appears in nearly every article written on this subject. The conclusion drawn from it is always the same: Google does not penalise AI content, so produce away.
The first half of that is true. Google has said since early 2023 that it evaluates the quality of content rather than how it was produced, and it has not wavered.
The second half does not follow, and a study published two weeks ago shows why.
Tomorrow’s forecast, in your inbox.
One email. Five minutes. Written for owners, not engineers.
The Number Almost Nobody Quotes
Ahrefs ran it again in June, larger. A million pages pulled from the top ten positions across 100,000 search results, with 150,000 containing enough text to score reliably.
| Among top-ranking pages | Share |
|---|---|
| Less than 20% AI content | 54.7% |
| Under 50% AI content, in the top three | 82.2% |
| Correlation between higher AI use and position | Negative |
Their own summary is direct: higher AI use is correlated with lower ranking positions. Fully AI-generated pages can still reach the top three, and they are a small minority when they do.
Other work points the same way. A separate analysis found the top result was human-written with roughly 80% probability, against 10% for purely AI-generated pages. Detection company Pangram found that the number one position does not correlate with much AI usage at all.
So both statements are accurate and only one of them circulates. There is no penalty for AI authorship. The top of the results is still overwhelmingly human-written, and the more AI in a page, the further down it tends to sit.
AI can produce the page. It can’t produce the reason the page deserves to rank.
Notice Who Publishes the Reassuring Version
Search this question and read the bylines. An AI humanising tool. An AI content generator. Two AI detection companies. Several agencies selling AI content services.
The finding that AI content is fine sells software. The finding that the top positions remain human-written does not.
One study deserves a specific note: it scores pages for AI probability using a detector sold by the company that published the study. The result may well be sound. The structure is circular enough to mention.
The cleanest source here is Ahrefs, whose newer finding cuts against the narrative that would sell more of what they sell.
What Google Actually Penalises
Two things, and neither is authorship.
The first is scaled content abuse — large volumes of low-value pages produced primarily to influence rankings, by any method. The March 2026 core update reinforced this. A person producing a thousand thin pages by hand is in the same category as a machine producing them.
The second sits in Google’s quality rater guidelines, which state that content can receive the lowest quality rating when it lacks effort, originality and added value — and specifically flag pages where all or almost all of the main content is AI-generated without meaningful human input.
Google has also confirmed it runs systems for detecting AI-generated content, with a named team responsible for it. Detection triggers evaluation rather than automatic demotion.
Put together, the picture is coherent. Nothing is being punished for having been written by a model. Pages are being sorted by whether they add anything, and a page with no human contribution has a hard time adding anything.
What Separates the AI Content That Ranks
The pattern in the pages that do well is consistent, and it is unglamorous.
They contain something the model could not have produced. A number from somebody’s own account. An example from a real engagement. A correction of a claim everyone else repeated. A price checked against the actual page rather than a summary of it. Firsthand operation of the thing being described.
None of that is about writing style, which is where most of the effort goes. It is about information that did not exist in the training data because somebody went and got it.
Where This Might Be Wrong
Detection is imperfect and everyone in this field admits it, including Google. Heavily edited AI content becomes difficult to distinguish, which means some pages scored as human-written probably started as drafts. That cuts both ways: it weakens the finding, and it also describes the workflow that appears to succeed.
These are correlations rather than experiments. Human-written pages may rank better because the people producing them also do more research, hold more expertise, and run older domains. Authorship may be a proxy for effort rather than a cause of ranking.
Which, if true, changes nothing about what to do.
The Forecast
The volume advantage from AI content keeps shrinking, and it shrinks for a reason that has nothing to do with Google’s policy. When production cost falls to near zero for everybody at once, publishing more stops being a differentiator, because your competitors got the same discount on the same day.
What remains scarce is anything that required somebody to go and find out. Original data, firsthand operation, a verified correction. Those cost the same as they always did, which is precisely why they hold value while everything around them gets cheaper.
High confidence, because this is arithmetic rather than prediction. An abundant input cannot be a competitive advantage.
Your Next Move
Good — 20 minutes. Take your three best-performing pages and your three worst. Read them side by side and look for what the good ones contain that the weak ones don’t. In most cases it will be specifics — a number, an example, a case, a name. The weak ones will be accurate and general.
Better — per piece. Before publishing anything, find one thing in it that could only have come from you. A figure from your own work. A client situation. Something you tested. If nothing in the piece qualifies, that is the piece explaining its own performance.
Best — ongoing. Shift the effort from producing more to finding out more. Run a small test and publish what happened. Check a claim everyone repeats and report what the primary source actually says. Ask ten customers the same question and publish the answers. That work does not get cheaper, which is the entire point of doing it.
Today’s Instrument
None today.
The tools that fit this topic are AI detectors and humanisers, sold by several of the companies whose research this issue questions. Recommending one would be an odd way to end. The move here is editorial and it costs nothing. We earn nothing from this issue.
Sources: Ahrefs, analysis of one million pages across 100,000 search results, June 2026 · Ahrefs, earlier study of 600,000 pages on AI content and ranking correlation · Pangram Labs on AI usage at top ranking positions · Google Search Quality Rater Guidelines · Google guidance on scaled content abuse, March 2026 core update