Roughly a year ago, Ahrefs analysed 1.9 million citations across 1 million AI Overviews and found 76.1% of cited pages ranked in the top 10, with a median rank of 2 for the first citation. A later Ahrefs analysis across 863,000 keywords found only 38% of citations came from top-10 results: 31.2% came from positions 11–100, and 31.0% came from beyond position 100. In about twelve months, AI Overview citation substantially decoupled from classic ranking. Today around two-thirds of citations come from pages that do not rank in the top 10. Chasing position 1 is no longer the reliable route to being cited.
Key Takeaways
- Ahrefs' earlier study (1.9M citations, 1M AI Overviews): 76.1% of cited pages ranked top 10; median rank of the first citation was 2.
- Ahrefs' later study (863,000 keywords): only 38% of citations came from top-10 results, 31.2% from positions 11–100, and 31.0% from beyond position 100.
- That is a shift from roughly three-quarters to roughly two-fifths of citations coming from the top 10, a substantial decoupling.
- Ahrefs' own follow-up framing is blunt: ranking #1 is "a coin flip at best" for getting cited.
- Practically: a page that ranks poorly can still be cited if it answers a specific sub-question well.
- This changes what content is worth producing, passage-level usefulness starts to matter more than page-level authority.
- These are two studies by one vendor using different keyword sets; the direction is clear but the exact magnitudes should be held loosely.
- As of late August 2026 the trend is still developing and no equivalent long-run independent replication has been published.
Citations used to cluster near the top of page one. In the later dataset they spread across the full ranking range.
The two datasets, side by side
The clearest way to see the change is to put the numbers next to each other.
| Metric | Earlier Ahrefs analysis (1.9M citations, 1M AI Overviews) | Later Ahrefs analysis (863,000 keywords) |
|---|---|---|
| Citations from top-10 results | 76.1% | 38% |
| Citations from positions 11–100 | Remainder of ~23.9% | 31.2% |
| Citations from beyond position 100 | Small minority within remainder | 31.0% |
| Median rank of first citation | 2 | Not comparable, distribution flattened |
| Practical read | Ranking top 10 was close to a prerequisite | Ranking top 10 is roughly a two-in-five affair |
| Strategic implication | Win rankings, citations follow | Win the sub-question, ranking may not follow or matter |
How to read the table honestly
The two studies use different sample constructions, one is citation-indexed, one is keyword-indexed, so this is not a controlled longitudinal comparison. What it is, is two large snapshots from the same analyst using the same broad method, roughly twelve months apart, pointing the same direction. That is strong enough to act on and not strong enough to quote as a precise rate of change.
What "decoupled" actually means
Before: rank was the gate
In the earlier picture, the AI Overview behaved like a summarisation layer over the existing top of page one. If you weren't in the top 10, you were largely invisible to it. SEO strategy and AEO strategy were effectively the same strategy.
Now: rank is one signal among several
In the later picture, roughly 62% of citations come from outside the top 10, and nearly a third come from pages beyond position 100 entirely. Whatever selects a passage for citation is clearly not primarily reading the classic ranking order.
Ahrefs' own framing
Their follow-up put it about as plainly as a vendor can: ranking #1 is "a coin flip at best" when it comes to getting cited. That is a striking thing for a company whose product is substantially about rankings to say, and it is worth weighting accordingly.
Why this decoupling is happening
I want to be careful here: the studies establish the what, not the why. These are plausible mechanisms, not confirmed causes.
Answers are assembled from passages, not pages
If a system is looking for a paragraph that resolves a specific sub-question, the best paragraph may sit on a page that ranks poorly for the head term. Page-level authority is a weak predictor of passage-level usefulness.
Queries are being decomposed
A single AI Overview may answer a compound question by fanning out into several narrower retrievals. Each retrieval finds its own best source. The union of those sources will naturally include pages that never ranked for the original query.
Retrieval indexes may differ from the ranked index
The candidate pool a generation system draws from need not be the classic ranked result set. If it isn't, ranking correlation would be expected to weaken over time, which is what the data shows.
Diversity constraints
Systems that deliberately vary their sources to avoid citing the same handful of domains repeatedly would also flatten the rank distribution.
The likeliest mechanism: passages get retrieved, not pages. A weak page with one excellent paragraph can win the citation.
What this changes about content strategy
Stop treating rank as the goal and start treating it as a byproduct
If two-thirds of citations come from outside the top 10, then a content plan organised entirely around ranking targets is optimising for a proxy that has weakened. Rankings still drive classic organic clicks, that has not changed, but they are no longer a reliable predictor of citation.
Write to sub-questions explicitly
The unit that gets cited looks like a passage that cleanly answers one narrow question. That argues for content with clear question-shaped headings, self-contained answers directly under them, and no requirement to read the surrounding page for context.
Value specificity over comprehensiveness
A 4,000-word guide that covers a topic broadly may rank well and be cited rarely. A 900-word page that definitively answers one awkward sub-question may never rank and be cited often. Both have value; they are not the same asset.
Don't kill pages for poor rankings alone
Content pruning based purely on ranking or traffic is now riskier. A page in position 60 that is cited regularly is producing brand exposure that your rank-based reporting cannot see. Check citation and recommendation behaviour before you delete.
What it does not change
Rankings still matter for clicks
Classic organic traffic still flows through the ranked results. Nothing in these studies suggests you should stop caring about rankings, only that you should stop assuming they buy you citations.
Quality and trust still matter
Nothing here implies low-quality pages get cited. It implies that the rank of a good page is a poorer predictor of citation than it used to be.
Fundamentals still apply
Crawlability, clear structure, accurate claims, and genuine subject expertise remain the base layer. The decoupling changes targeting, not craft.
How to test this on your own site
Step 1: Pull your citations
Collect the pages of yours that appear as sources in AI answers, using whatever tooling you have or manual prompt testing if you have none.
Step 2: Join them to rank data
For each cited page, record its ranking position for the query that produced the citation.
Step 3: Build your own version of the table above
You will almost certainly find pages cited from outside the top 10. The proportion is what tells you how your site specifically behaves, which matters more than any industry average.
Step 4: Look at what the cited passages have in common
In my experience the answer is usually structural: a direct question heading, a short definitive answer immediately beneath it, and specific numbers or steps. Structure is easier to replicate than authority.
What this means for smaller and newer sites
This is the genuinely encouraging part. In the old picture, a site that could not crack the top 10 was locked out of AI Overview visibility entirely. In the current picture, roughly a third of citations come from beyond position 100, a zone that includes a great deal of newer and smaller-domain content.
The opportunity
If you cannot outrank an incumbent on a head term, you can still win the passage that answers the sub-question the incumbent covered in one sentence. That is a realistic path for a startup or a niche brand in a way that head-term ranking is not.
The caution
"Cited" is not "recommended," and neither is automatically "revenue." Track what actually converts before restructuring a content programme around citation counts.
Caveats worth stating plainly
Both datasets come from one vendor. The samples are constructed differently, so this is a directional comparison rather than a clean before-and-after. AI Overview behaviour varies by query type, country, and vertical, and none of these aggregates tell you how your category behaves. As of late August 2026, no equivalent long-run independent replication has been published. The direction of travel, away from rank dependence, is well supported. The precise numbers are one analyst's view.
The practical upside of decoupling: passage quality can beat domain size on specific sub-questions.
Frequently Asked Questions
What did Ahrefs' earlier AI Overview study find?
Analysing 1.9 million citations across 1 million AI Overviews, Ahrefs found 76.1% of cited pages ranked in the top 10, and the median rank of the first citation was 2.
What did the later Ahrefs analysis find?
Across 863,000 keywords, only 38% of citations came from top-10 results. 31.2% came from positions 11–100 and 31.0% came from beyond position 100.
Does ranking #1 still get you cited?
Ahrefs' own follow-up framing describes it as "a coin flip at best." Ranking first improves your odds but no longer reliably delivers citation.
Can a page that doesn't rank still be cited?
Yes, roughly a third of citations in the later dataset came from pages beyond position 100. A page that answers a narrow sub-question well can be cited without ranking for the broader query.
Should I stop doing traditional SEO?
No. Rankings still drive classic organic clicks. What has changed is that rankings are no longer a dependable proxy for AI citation, so you need to plan and measure the two separately.
What kind of content gets cited from low rankings?
In practice, pages with question-shaped headings and short, self-contained, specific answers directly beneath them. Passage clarity appears to matter more than page length or domain strength.
Why did the decoupling happen?
The studies do not establish causation. Plausible mechanisms include passage-level retrieval, query decomposition into sub-questions, a retrieval index that differs from the ranked index, and deliberate source diversification.
How reliable are these numbers?
They come from large samples but from a single vendor, using differently constructed datasets, so treat the direction as well supported and the exact magnitudes as approximate. No equivalent independent replication has been published as of late August 2026.
Should I delete pages that rank poorly?
Check their citation behaviour first. A page ranking in position 60 that is regularly cited is generating exposure your rank-based reporting will not show you.
What is the single most useful change to make?
Restructure your existing pages so each section poses a real question and answers it completely in the following two or three sentences. It is cheap, it helps human readers, and it is the pattern most consistent with what gets cited.
Work with me
If you want help working out which of your pages are being cited despite ranking badly, and what to build more of, you can see my work and get in touch through the contact form at younusfardeen.com. I have four-plus years of marketing experience helping edtech and startup brands grow organically, and this kind of audit is usually where the quickest wins are hiding.