For two years the standard answer to “how do we get into AI answers?” was comfortable: rank well in Google and the AI will cite you. Comfortable because nothing had to change — the SEO budget was approved, the rank tracker installed, and AI visibility came free with work you were already doing.
That answer now has a measured expiry date. Ahrefs re-ran its study of how often AI-Overview-cited pages also rank in Google, and the overlap has roughly halved in about seven months. This page is about what that means operationally. It is not an argument for abandoning SEO, and we make the case against that below.
The short answer: Increasingly, no. In Ahrefs’ March 2026 analysis of 863K keyword SERPs and 4M AI Overview URLs, 37.9% of URLs cited in AI Overviews also appeared in the first 10 blocks of the same query’s SERP — down from about 76% in its July 2025 study. Ranking still helps, but it is no longer a reliable proxy for AI visibility. If you want to know whether engines cite you, you now have to measure citations directly.
What the 2026 data actually says
The source is Ahrefs’ study Update: 38% of AI Overview Citations Pull From The Top 10, published 2 March 2026 by Louise Linehan, last updated 31 May 2026. The sample: 863K keyword SERPs and 4M AI Overview URLs, from Ahrefs Brand Radar.
The population measured matters. Ahrefs looked at cases where “the same URL appeared in both the AI Overview and the regular SERP, for the same query” — that is, of the pages Google cited in its AI Overview, how many was Google already ranking for that exact query? They ran it twice.
- All SERP blocks counted (ads, featured snippets, People Also Ask, video packs and organic listings each tracked as a separate block): 37.9% of AI-Overview-cited URLs appeared within the first 10 blocks. The remainder split almost evenly between positions 11–100 (31.2%) and beyond the top 100 blocks (31.0%).
- Standard blue links only, ignoring ads and SERP features: 37.10% ranked in the top 10, 26.20% ranked 11–100, and 36.70% did not rank in the top 100 at all.
Both cuts land in the same place. However you define “ranking”, only around 37–38% of AI Overview citations came from pages on page one for that query, and roughly a third came from pages Google was not ranking anywhere in its top 100.
Ahrefs’ own reading of the shift: “Google is selecting far fewer pages straight from the original SERP”, which they attribute to query fan-out — the system splitting one search into multiple sub-queries and citing pages that surface consistently across those sub-query SERPs, rather than pages that win the original one.
Read the methodology before you quote the drop
You will see “76% to 38%” repeated everywhere this year with no caveat. It deserves one, and that is the difference between a claim you can defend in a board meeting and one you cannot. The July 2025 baseline came from a different study: 1.9M citations from 1M AI Overviews, analysing the top three most visible citations in each response, which found 76.10% of AI-Overview-cited pages ranked in the top 10. The 2026 study is larger, is not restricted to the top three citations per response, and Ahrefs states it improved its parsing methodology in between so that it can see more of the citations that appear in AI Overviews. AI Overviews also switched to Gemini 3 in January 2026.
So the two figures are not the same instrument pointed at two dates. Some of the gap is almost certainly better detection of citations that were always there, further down the answer, from lower-ranking pages. Treat it as a substantial weakening, not a precise 38-point collapse.
It is wider than Google
AI Overviews are the surface most tightly coupled to Google’s own index, so they are the friendliest case for the rank-equals-citation theory. Off Google, the overlap is thinner still.
In a separate Ahrefs study from August 2025 of 15,000 long-tail queries run through four assistants (ChatGPT, Gemini, Copilot and Perplexity) and both major search engines, an average of about 12% of AI-assistant citations also ranked in Google’s top 10 for the same prompt. Perplexity was the outlier at 28.6%; the rest sat between 6% and 9%. Ahrefs reports 80% of those citations did not rank anywhere in Google for the original query.
There is a concrete example of what fills the gap. Among AI-Overview-cited pages that did not rank in Google’s top 100 for the same keyword, 18.2% were YouTube URLs, and those accounted for 5.6% of all AI Overview citations in the dataset — a surface with no organic ranking for your query supplying a meaningful share of the answer.
What each measurement can and cannot tell you
| Instrument | What it measures | Still genuinely useful for | Where it misleads in 2026 |
|---|---|---|---|
| Google rank tracking | Position of your URL for a keyword | Organic click acquisition, competitive share on commercial terms, catching technical regressions fast | Treated as an AI visibility proxy. Around 37–38% overlap with AIO citations means it is wrong most of the time in that role |
| AI Overview citation tracking | Whether your URL appears as a source inside Google’s generated answer | The specific Google surface, and spotting fan-out topics you are absent from | Single checks are noisy; AIO composition changes often, so one screenshot is not a position |
| Multi-engine citation polling | Whether you are named or linked in answers across ChatGPT, Gemini and others, on a fixed prompt set | The only view covering assistants outside Google, and whether a win holds | Costs real setup effort; blending engines into one number hides per-engine reality |
| Referral analytics | Sessions arriving from AI surfaces | Proving downstream value of citations | Undercounts badly — most AI answers resolve the question without a click |
Three things that change operationally
1. Rank tracking stops being your AI visibility report
It remains a good report about rankings and is now a poor proxy for citation. If your monthly deck reports positions and then draws conclusions about AI visibility, those two halves are no longer connected by the evidence. Keep the rank tracker; stop letting it answer a question it does not measure. Our guide to share of answer as a metric sets out the replacement KPI and how to define the prompt set.
2. Citation-correlated attributes now earn their budget independently
When citation tracked ranking, anything that helped citation was justified by the ranking it also produced. Now the case has to stand alone — which is fine, because these attributes are cheap relative to a link-building programme.
A direct answer capsule near the top. Genuine comparison structure, including the cases where you are the wrong choice. Entity consistency, so an engine can resolve that scattered mentions are one business — the groundwork we cover in entity consistency for AI search. Third-party corroboration, because engines lean on sources others agree with. And freshness, because stale pages drop out of volatile answer sets first. Our synthesis of the largest 2026 citation datasets ranks these by leverage.
None of those require you to outrank anybody. That is the point.
3. You have to measure citations directly, and repeatedly
Most teams skip this, because it means building an instrument rather than buying a dashboard. We run one on ourselves: LeadsNow polls a fixed registry of real buyer prompts, in Australian and US variants, across ChatGPT, Gemini and DuckDuckGo on a standing cadence, logging every hit and miss. The published slice of that dataset covers 5,051 polls over 77 days across 68 prompts, and two findings from it bear directly on this page.
First, 74% of the prompts we were ever cited on were cited by only one of the three engines — there is no single AI ranking to win, so there is no single position to track. Second, citations are not stable the way rankings are: once first cited on a prompt, ChatGPT kept citing us in only about 44% of that prompt’s polls from the first citation onward (median), against roughly 70% for Gemini.
That is one domain in one niche, and we present it as directional rather than universal — but it is the same shape the Ahrefs data shows at scale, arrived at independently.
The honest counterpoint: do not abandon SEO
The wrong lesson from a 38% figure is that rankings no longer matter. Three reasons.
It is a weakening correlation, not an inversion. Top-10 pages remain hugely over-represented among AI Overview citations relative to their share of the index: they are a tiny fraction of the pages eligible to be cited, and collect around 37–38% of citations.
Ranking also produces clicks in its own right. AI Overviews compress organic click-through but have not eliminated it, and the traffic that still arrives is traffic you own. We cover that trade-off in what to do about AI Overviews on one in four searches.
And the two disciplines share most of their inputs: crawlability, coherent site structure, pages that actually answer the query, external sources referencing you. What changes is the marginal budget. The next dollar is better spent on answer structure, corroboration and measurement than on chasing position 4 to position 2.
How we treat this in a commercial pipeline
We are a pay-per-result lead generation and appointment-setting business — 50,769+ AI-booked sales appointments since 2017 and 1M+ leads generated. AI search visibility matters to us for the reason it matters to you: it feeds a pipeline we get paid on outcomes from. So we treat citation polling as instrumentation, not reporting — the way a good performance team treats a holdout rather than a platform-attributed conversion count. For the general version of that discipline, our companion pages on running an incrementality test on lead gen spend and designing a pilot that can fail make the same argument in the paid-acquisition context: a proxy you cannot falsify is not a measurement.
To talk through how this applies to your pipeline, book a call.
Frequently asked questions
Do AI Overviews cite pages that rank in Google?
Sometimes, and less often than they used to. Ahrefs’ March 2026 analysis of 863K keyword SERPs and 4M AI Overview URLs found 37.9% of URLs cited in AI Overviews also appeared within the first 10 blocks of the same query’s SERP, with roughly a third coming from pages that did not rank in the top 100 at all. Ranking improves your odds, but it neither guarantees a citation nor is required for one.
Has the overlap between rankings and AI citations really halved?
Directionally yes, with a caveat worth stating. Ahrefs compares about 76% in July 2025 to about 38% in its 2026 update, but the two studies used different samples — the earlier one analysed the top three citations in each of 1M AI Overviews, the later one analysed all cited URLs across 863K SERPs after Ahrefs improved its citation parsing. Read it as a substantial weakening of the relationship rather than a precise 38-point drop.
Why would an AI Overview cite a page that does not rank?
Because the answer is not built from your query alone. Google performs a query fan-out, splitting one search into related sub-queries, and pages appearing consistently across those sub-query results can be cited even when they lose the original SERP. That is why answering adjacent questions thoroughly can matter more than moving up two positions on the head term.
Is this only a Google problem?
No, and Google is the mild case. A separate August 2025 Ahrefs study of 15,000 long-tail queries found that an average of about 12% of citations in AI assistants also ranked in Google’s top 10 for the same prompt, with Perplexity the outlier at 28.6% and the others between 6% and 9%. Assistants outside Google are further from the search index, not closer to it.
Should we stop investing in SEO?
No. This is a weakening correlation, not an inversion, and top-10 pages remain massively over-represented among cited sources relative to how few pages rank there. Rankings also still produce clicks that AI answers do not. The change is where the marginal dollar goes: toward answer structure, entity consistency, third-party corroboration and direct citation measurement rather than toward incremental position gains.
How often should we check whether AI engines cite us?
Weekly at minimum, per prompt and per engine. In our own 5,051-poll dataset, once ChatGPT first cited us on a prompt it kept citing us in only about 44% of that prompt’s polls from the first citation onward (median), versus roughly 70% for Gemini. With that much volatility, a single spot-check tells you almost nothing about whether you hold the position.
What is the fastest signal that our AI visibility is decoupled from our rankings?
Pick ten prompts where you rank in the top three and ask each engine the question directly, twice a week for a month. If your citation rate on those prompts moves independently of your positions, your rank tracker is no longer reporting on AI visibility and you need a second instrument.
See if we’re a fit
A few quick questions. If it’s a fit, our live calendar loads on the next screen. If it isn’t, we’ll point you to free resources instead — you won’t have to sit through a sales call to find out.
We get paid a performance fee equivalent to 10–20% of the sales we help you generate.
Are you OK with that?
If you’re not willing to pay 10–20% as a performance fee, are you happy to pay a $4,000+ per month retainer?
Check If You Qualify 👇
How many leads per month do you currently get?
What’s your current advertising spend or marketing budget (Meta, Google, SEO, etc.)?
What’s the average sale worth to you over that customer’s lifetime?
Given your business currently gets less than 10 leads per month, we’d need to do much more groundwork to set up end-to-end sales systems. Are you OK with a $2,000/mo retainer to do so? (no lock-in)
What’s your work email?
Hey! We might be able to add $100k+ / mo... Enter your email to choose a time!
We’re probably not the right fit — yet
Our model is pay-on-performance — we only win when you’re making sales, and it works best alongside an active marketing engine with advertising budget to get seen. Booking a call now would waste your time, and we’d rather be straight with you.
Grab the free stuff instead — it’s the same playbook we use:
Read the growth blog · Lead-gen FAQ
When the timing’s right, come back — the calendar will be waiting.