Ranking top ten used to mean getting cited. Ahrefs says that halved in eight months.

What Ahrefs measured

In March 2026, Ahrefs published a study of Google's AI Overviews covering 863,000 keywords and 4 million cited URLs. It found that 38% of pages cited in an AI Overview also appeared in the top ten organic results for the same query. When Ahrefs ran the same study in July 2025, the figure was 76%.

The overlap between ranking and being cited roughly halved in eight months. That is the entire finding, and it is worth stating carefully before anyone builds a strategy on top of it.

What it does not mean

It does not mean rankings stopped mattering. Thirty-eight per cent is not nothing; a top-ten position is still associated with citation far more often than chance would predict. It also does not mean Google is citing worse sources — the study measures overlap, not quality, and offers no view on whether the newly cited pages are better or worse than the ones they displaced.

It does not tell you the cause, either. A change of this size could come from a shift in retrieval, from a change in which queries trigger an Overview at all, from growth in the number of sources cited per answer, or from some combination. The study measures the outcome, not the mechanism, and any confident explanation you read about why it happened is an interpretation rather than a finding.

What it does mean, practically

It means the inference "we rank well, therefore we are being cited" has stopped being safe. In July 2025 that inference would have been right three times in four. By March 2026 it is right roughly one time in three. An assumption with that error rate is no longer a substitute for looking.

Put next to Yext's October 2025 analysis of 6.8 million citations — which found 86% of citations came from sources brands can manage or influence, split 44% first-party sites and 42% listings and profiles — a fairly clear allocation emerges. If roughly half of what gets cited about you sits on third-party surfaces, and ranking is a weakening proxy for citation, then effort spent exclusively on your own pages is being spent on part of the problem.

How to check your own position rather than inheriting the average

These are aggregate figures across hundreds of thousands of keywords. Your category may look nothing like the average, and the only way to know is to measure it directly.

1. Write down ten to twenty questions a buyer would actually ask an assistant about your category — shortlists, comparisons, specification questions — phrased the way a person types them, not as keywords.
2. Run each one and record two separate things: whether you were named, and whether one of your sources was cited. They are different outcomes and being named without being cited is common.
3. For every citation that appeared, record which surface earned it: your site, a listing, a directory, a third-party article.
4. Separately, note your organic position for the same query.
5. Repeat unchanged on a schedule. Answer engines are non-deterministic, so a single run is a sample and only a trend across runs is evidence.

The comparison in step four is the one that turns an industry statistic into a decision. If you rank well and are cited, the foundation is working. If you rank well and are not cited, the problem is at the entity or passage layer rather than the ranking layer. If you rank poorly and are not cited, start with the classic SEO work, because there is nothing above it to fix yet.

A note on reading studies like this one

The figures above are quoted with the study, the date, the sample size and the earlier comparison point deliberately attached. That is partly honesty and partly self-interest: the 2024 KDD paper on generative engine optimisation found that citing sources, adding quotations and adding statistics produced a 30–40% relative improvement on its visibility metric. Real attribution is the tactic. A number without a source attached fails the reader and the engine at the same time.