Google AI Overviews Cut Top-10 Citation Share From 76% to 38% in 7 Months
Ahrefs analyzed 4M AI Overview URLs. Top-10 ranking pages now earn 38% of citations, down from 76% in July 2025. The fan-out era has arrived.
Google AI Overviews cut top-10 citation share from 76% to 38% in 7 months
Seven months ago, ranking in Google’s top 10 was the most reliable way to land in an AI Overview. Today, top-10 pages earn just 38% of those citations.
That number comes from a March 2, 2026 study by Ahrefs, which analyzed 863,000 keyword SERPs and 4 million AI Overview URLs. The same methodology, applied in July 2025, found that 76% of cited pages also ranked in Google’s top 10. Seven months later, the rate had nearly halved. Ahrefs published the update under the title “38% of AI Overview Citations Pull From The Top 10,” and it’s the clearest signal yet that Google’s AI is no longer just a fancy front-end on Google’s blue links.
If your AI Overviews strategy is “rank well and you’ll be cited,” that strategy was reasonably accurate last summer. It isn’t now.
Where the citations went
The Ahrefs data splits citation sources into three roughly equal buckets:
| Ranking position | Share of AI Overview citations |
|---|---|
| Top 10 (page 1) | 37.9% |
| Positions 11–100 | 31.2% |
| Beyond position 100 | 31.0% |
Almost a third of every citation in a Google AI Overview now comes from a page that doesn’t rank in Google’s top 100 for the query the user actually typed. That’s not a rounding error. It’s a redistribution.
A separate analysis from BrightEdge, published February 12, 2026, measured top-10 overlap at roughly 17% using a different methodology. The two numbers bracket the truth: somewhere between one in six and three in eight AI Overview citations come from page-one rankings.
YouTube’s role in the shift deserves its own line. Among AI Overview citations from pages outside the top 100, 18.2% are YouTube URLs. Across the full dataset, YouTube alone accounts for 5.6% of all AI Overview citations and grew 34% in six months as a citation source. Ahrefs’ Louise Linehan flagged this as one of the clearest signs that AI Overviews are pulling from sources Google’s organic ranking system doesn’t even score.
Why the gap opened
The mechanism is query fan-out. When a user types a question, Google doesn’t just answer that one query — it splits the question into multiple sub-queries and gathers candidate sources from each one. The page selected for the AI Overview is whatever shows up most reliably across the fan-out, not whatever ranks highest for the original phrasing.
We’ve covered the mechanics of fan-out before in our grounding queries piece. What’s new is the magnitude. In July 2025, fan-out was a finishing touch on top-10 results. Today, it’s the primary selection mechanism, and the top 10 is one input among many.
A few practical consequences fall out of this.
Pages with broad topical coverage win. A single article that answers ten related sub-questions can show up in ten fan-out paths. A narrowly optimized page wins one and loses nine.
Video and community sources are first-class citizens. YouTube’s growth rate isn’t an artifact. AI Overviews route around Google’s organic results to pull from where people are actually answering questions, and that’s increasingly on video.
Long-tail content can punch above its weight. A page ranking #47 for the user’s query might rank #3 for one of the sub-queries Google generates internally. That sub-query rank is what gets it cited.
What ranking #1 still buys you
This isn’t an obituary for SEO. Ranking matters — it’s just no longer sufficient.
Position #1 still has the highest single-position citation probability of any organic slot. The Ahrefs dataset puts it at 33.07%, meaning roughly one in three queries where you rank #1 will produce an AI Overview citation that includes your page. That’s a real number. It just isn’t the 70%+ implied by last year’s data.
Across the entire top 10, there’s a 76.1% probability that at least one page from that tier will be cited. So if your goal is “be one of the pages Google considers,” ranking still gets you in the room. The new question is whether you’re the page that wins the fan-out, and that’s a different optimization problem than ranking #1.
The strategic reframe
Most teams still measure AI Overview readiness by checking their top-10 rankings. That measurement was built when 76% of citations came from page one. At 38%, it produces a confident-looking dashboard that’s mostly noise.
The replacement isn’t complicated, but it’s different. Three shifts matter most.
1. Track AI Overview citation share separately from organic rank. They’re correlated but no longer interchangeable. A page ranking #4 for the user’s query might be cited 60% of the time; another ranking #2 might be cited 10%. Without separate tracking, you can’t tell which is which.
2. Optimize for sub-queries, not the head term. Build content that answers the natural fan-out — adjacent questions, comparisons, edge cases, “how do I…” follow-ups. The AI is reading your page through the lens of multiple sub-queries at once.
3. Add video to the citation mix. A 34% growth rate in six months for YouTube as a citation source is not a side note. If your competitors have a YouTube presence and you don’t, the AI Overview has a reason to cite them and skip you, even when your written content ranks better.
The contrarian read
There’s a more uncomfortable interpretation of the Ahrefs data, and it’s worth saying out loud.
Google AI Overviews used to look like Google’s top 10 with extra text on top. That’s why so many SEOs treated AI search as an extension of their existing program. The 76% overlap was the empirical foundation for that view. The 38% number says the foundation is gone.
What replaces it isn’t another ranking system. It’s a retrieval-and-synthesis system that draws from a much wider pool, weighs sources by how well they answer the underlying intent, and produces an answer that wasn’t really “ranked” anywhere. That makes AI Overviews behave more like ChatGPT than like Google. Which is, ironically, exactly what Google was trying to prevent by building its own AI search in the first place.
For brand teams, the practical implication is simple. The era of “we’ll get into AI Overviews by improving our SEO” is closing. The brands that get cited at 60%+ rates from now on will have done three things: built broad topical content that survives fan-out, earned visibility on the off-SERP sources AI Overviews pull from (YouTube, Reddit, Wikipedia, structured data, industry directories), and tracked AI citations as a metric independent of rankings. Not as a sub-section of an SEO report.
That last point is where most teams still lose. We’ve written before about the 15 domains that control 68% of AI visibility and the ghost citation gap, and the through-line in both is the same: the data you need lives outside your traditional SEO stack. AI Overviews data is not in Search Console. It needs its own monitoring.
The 76%-to-38% drop happened in seven months. If the trajectory continues — and the BrightEdge 17% number suggests it might already have — top-10 rankings will explain less than a third of AI Overview citations by the end of 2026. At that point, “rank and pray” is not a strategy. It’s a story you tell yourself about a measurement system that no longer matches how the answer gets built.
Stop guessing about your AI search presence. Start your free RivalHound trial and get real data.