How to get cited in AI Overviews when ranking is not enough
The short answer
Ranking first does not get you cited. The relationship is far weaker than most teams assume, and it has weakened fast.
Ahrefs examined 863,000 keyword SERPs and 4 million AI Overview URLs. Only 37.1 percent of cited pages ranked in the top 10 organic results for that query.
The striking part is where the rest sat. 36.7 percent ranked beyond position 100 entirely, almost exactly matching the top-10 share. Another 26.2 percent sat between 11 and 100.
Key takeaways
- Top 10 and beyond-100 are nearly tied as sources of AI Overview citations, at 37.1 percent and 36.7 percent.
- The link to ranking has weakened quickly. In late 2024, roughly three quarters of citations came from top-10 pages. It is now around a third.
- Being indexed is the gate. Being top 10 is not the path. Those are different claims, and conflating them is why good rankings feel like they stopped working.
- Video is a real slice. YouTube URLs made up 18.2 percent of non-ranking citations, and 5.6 percent of all AI Overview citations.
- Citation pays. Cited pages earn 2.07 percent click-through against 0.94 percent uncited. Both trail the 3.35 percent of a page with no Overview.
Where this comes from. We run organic for more than 250 enterprises at Pepper and track over 10 million prompts across every major engine. We hear this exact complaint most weeks. Rankings are stable, citations are not appearing. That work shapes the view here, along with the client reviews we sit in and the conversations we have at the events we run.
What are AI Overview citations?
An AI Overview citation is a link Google attaches to its generated answer, crediting a page it drew on.
Three things separate it from a ranking, and all three matter for how you work.
- It is awarded per passage rather than per page. Google picks a piece of text answering one part of the question. Your best two hundred words compete, rather than your whole article.
- It is not positional. There is no first place. A page either appears among the cited sources or it does not.
- It moves independently of rank. A page can hold position three and never be cited, while a page on page eight is quoted regularly. What a results page contains explains how the surface got this way.
That third point is the one worth internalising, and the data below makes it concrete.
See where you show up. Pepper’s GEO platform tracks Brand Visibility, Domain Prompt Presence and Share of Voice across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. Your team can log in, connect Search Console and GA4, manage the prompt set and run your own agents in the Agent Atlas. A growth team works the same account alongside you. Book a growth audit or see where you show up.
Where cited pages actually rank
This is the finding that should change how you brief work.
The distribution is close to flat, which is not what anybody expects. A page ranking beyond 100 is about as likely to be cited as one in the top 10. Position is simply not doing the work people assume it does.
Read the trend alongside the level. In late 2024, roughly 76 percent of AI Overview citations came from top-10 pages. By early 2026 that had fallen to between 17 and 38 percent, depending on whose dataset you read. The direction is consistent across every measurement, even where the levels disagree.
On the disagreement, stated plainly
On the disagreement, stated plainly. Published top-10 overlap figures for the same period range from about 17 to 38 percent. They use different datasets, different parsing and different definitions of a citation. We use the Ahrefs figure because it publishes its sample, and we treat the range as the honest uncertainty rather than picking the most dramatic end.
And the authors’ own caveat. Ahrefs notes that its parsing improved between studies, so it now sees more citations than before. Earlier analyses may have undercounted, which could affect the distribution as well as the total.
Resolving an apparent contradiction
We have written elsewhere that classic ranking gates AI citation. This study looks like it says the opposite. It does not, and the distinction is worth ten minutes.
Two different claims are involved.
- Being retrievable through search gates citation. Across 1.4 million ChatGPT prompts, 88.46 percent of citations came via the search channel, against 1.93 percent from Reddit. If you are not in the index, you are not in the candidate pool.
- Ranking in the top 10 does not determine citation. Only 37.1 percent of AI Overview citations came from top-10 pages, and a near-identical share came from pages beyond 100.
So a page at position 60 sits fully in the index, fully retrievable and entirely eligible. It simply is not near the top. Being in the index is the gate. Being at the top of it is not the path.
That resolves the thing most teams find maddening, which is stable rankings and no citations. You cleared the gate years ago. The work that remains is passage-level, and rank reporting cannot see it. Brand visibility covers how to report it instead.
What actually earns the citation
Four things, and none of them is a ranking tactic.
Work down in order. Each one is cheaper than the one below it, and the first two cost nothing but editing time.
- Answer in the first two sentences under every heading. Google selects a passage. A buried answer loses to a thinner page that leads with one.
- Write sections that survive being lifted out. A section needing the paragraph above it cannot be quoted cleanly. Remove back-references like “as we saw above”.
- Put something in it a model cannot compose. A price, a sample size, a named method, a dated observation. Generic explanation is the one thing the engine does not need you for, and earned third-party coverage is how the specifics travel beyond your own domain.
- Consider video where it fits. YouTube accounted for 18.2 percent of non-ranking citations. On Google’s surfaces specifically, chaptering matters, and our analysis of 100 million citations covers when video is worth the production cost.
The broader evidence on passage-level craft sits in what AI engines actually cite, including the finding that title similarity to the question is the strongest measured signal.
What citation is worth
Worth stating plainly, because it justifies the effort and also bounds it.
Seer Interactive’s 2025 averages give click-through by what the page contained. No AI Overview: 3.35 percent. AI Overview present and citing you: 2.07 percent. AI Overview present and not citing you: 0.94 percent.
So citation is worth roughly 120 percent more clicks per impression than no citation. That is a large, real gain, and it is the case for doing this work.
It also does not restore what the format took. A cited page still trails a no-Overview page by about 38 percent. Anyone promising citation returns your old traffic is selling something. We set out the wider strategic response in what to do when clicks fall.
Our methodology: how we weighted the levers
| Lever | Weight | Why it carries this much |
|---|---|---|
| Passage-level answer placement | 35% | Google selects a passage, and this is the cheapest thing to change |
| Non-composable specifics | 30% | The one thing an engine cannot produce without you |
| Index presence and crawlability | 20% | The gate. 88.46 percent of citations came via the search channel |
| Ranking position | 15% | Weighted low deliberately, since only 37.1 percent of citations came from top-10 pages |
**What we could not verify and left out.** Any single authoritative figure for top-10 overlap, because published estimates for the same period span roughly 17 to 38 percent. Also excluded: the claim that cited brands earn 35 percent more organic click-through, and the widely repeated 68 percent zero-click figure. Neither traces to a published method we could open.
Citation levers at a glance
| Lever | What it changes | Typical 2026 cost | Evidence behind it | Where it falls short |
|---|---|---|---|---|
| Answer-first structure | Whether a passage can be lifted | A day of editing per page | Google selects passages, not pages | Reads slightly repetitive to a linear reader |
| Self-contained sections | Whether the passage survives removal | Habit change for writers | Same mechanism | Retrofitting an archive is slow |
| Original data or specifics | Whether you are replaceable | The real cost, and the real return | Generic text is what engines replace | You have to actually have something |
| Question-shaped headings | Whether the passage matches a sub-question | Free, one line per heading | Title similarity is the strongest measured signal | Fails if the section then does not answer it |
| Video | Access to a different citation pool | Production, the main line item | 18.2 percent of non-ranking citations | Concentrated on Google and Perplexity |
| Chasing top-10 rank | Position, not citation | Your existing SEO budget | Only 37.1 percent of citations rank top 10 | Necessary for traffic, weak for citation |
| Platform plus a growth team | Tracking, plus the work it points at | Custom | Whole-programme | Still gated by your publishing pace |
How to choose what to fix
Rank reports cannot diagnose this, which is why the problem persists in well-run programmes.
The scorecard
Score a page out of 100 before rewriting it.
| Factor | Weight | How to score it honestly |
|---|---|---|
| Answer in the first two sentences | 30% | Under every heading, and someone has actually checked rather than assumed |
| Sections survive being lifted out | 25% | You cut one out, read it alone, and it still makes sense |
| Contains something non-composable | 25% | A price, a number, a named method or a dated observation appears on the page |
| Headings read as questions | 15% | Read one aloud and ask whether a buyer would type something close to it |
| Page is indexed and renders as HTML | 5% | Confirmed in Search Console and by a raw fetch, rather than assumed |
The live test, over 90 days
Fix 30 prompts in your buyers’ words and hold them steady. Real questions, such as “how do I get cited in AI Overviews”, “best data observability platform for a small team”, “how much does sales enablement software cost” and “what is a SERP”.
Run them on Google. For each, record three things: whether an AI Overview appeared, whether it cited you, and which passage of yours it quoted.
Rerun the identical set monthly for 90 days. Three readings is the shortest honest window. One is a snapshot, two could be noise, three shows direction. Keep the set fixed, because changing it resets the comparison.
That third column is the one worth adding here. It shows which two hundred words are doing the work, and those passages are rarely the ones anyone predicted. The free method is in how to measure AI search visibility without expensive tools.
Weak approach versus strong approach
The weaker approach: treat missing citations as a ranking problem, build links, chase position three, then wait. Position and citation are linked weakly enough that this can succeed entirely and change nothing.
The stronger approach: leave the ranking alone and rewrite the page so each section answers one question completely in its first two sentences.
One buys position, which the data says is worth about 37 percent of the story. The other buys quotability. That is the part rank reporting cannot see, and most competitors are not doing it.
Red flags
- “Rank higher to get cited.” Only 37.1 percent of AI Overview citations came from top-10 pages.
- Guaranteed AI Overview placement. Nobody controls generated output, and Google warns against providers guaranteeing rankings.
- Schema markup offered as the fix. A controlled study measured minus 4.6 percent on AI Overviews.
- A single top-10 overlap figure quoted as fact. Published estimates span roughly 17 to 38 percent.
- Traffic recovery promised through citation. A cited page still trails a no-Overview page by about 38 percent.
- Rank tracking as the only reporting. It cannot see citation at all, which is the metric that now matters here.
Five questions worth asking
- “Which of our passages has Google actually quoted?” Most teams have never looked.
- “Does every section answer its heading in the first two sentences?” The cheapest fix available.
- “What is on this page that a model could not write without us?” If nothing, the page is replaceable.
- “Are we indexed and rendering without JavaScript?” The gate, and worth confirming rather than assuming.
- “What would you tell us to stop doing?” Often, chasing position on queries where citation is the real objective.
Reduced to one principle: win the passage, not the position.
One closing note that costs us something. Most of this is editing work on pages you already own, which is a far smaller engagement than a ranking programme or a content calendar, ours included in the general point that agencies are paid more for the larger one. Do the editing first and see what moves.
What does this cost?
| What you are buying | Typical 2026 cost | What it covers |
|---|---|---|
| Checking which passages get quoted | Free, about two hours a month | Thirty prompts on Google, with the quoted-passage column |
| Answer-first restructuring | Roughly a day of editing per page | The highest-evidence change here |
| Question-shaped headings | Free, one line per heading | Matching a passage to a sub-question |
| Original data or named expertise | The real cost, and the real return | The non-composable part |
| Video production | Varies widely | Access to 18.2 percent of non-ranking citations |
| Multi-engine tracking | Roughly $99 to $400 a month | Citation tracked per engine, with history |
| Platform plus a growth team | Custom | Pepper: tracking, plus the people and agents doing the work |
The first row is free and it tells you whether the rows below are worth buying.
What nobody should promise you
Guaranteed placement in an AI Overview. Nobody controls generated output, and Google warns against providers guaranteeing rankings.
Citation through ranking alone. Only 37.1 percent of cited pages ranked in the top 10, and a near-identical share ranked beyond position 100.
Traffic recovery through citation. A cited page still trails a no-Overview page by roughly 38 percent.
A precise top-10 overlap figure. Published estimates for the same period span roughly 17 to 38 percent, depending on dataset and parsing.
When ranking really is the problem
The answer that costs us the sale. If your pages are not indexed, or you do not appear in the first hundred results for the queries you care about, rewriting for citation is premature and we would tell you not to start there.
The index is the gate, and that part of the old advice still holds. Get retrievable first through ordinary SEO, then make the passage worth quoting. The other order produces beautifully structured pages Google never had the chance to consider. That is an expensive way to learn the sequence.
Frequently asked questions
Do AI Overviews only cite pages that rank in the top 10?
No, and it is not close. Across 863,000 keyword SERPs and 4 million AI Overview URLs, 37.1 percent of cited pages ranked in the top 10. Another 36.7 percent ranked beyond position 100 entirely.
Why do I rank well but never get cited?
Because citation is awarded per passage rather than per page. Google selects a piece of text answering one part of a question. A well-ranked page with a buried answer loses to a lower-ranked page that leads with one.
Has the link between ranking and citation changed?
Sharply. In late 2024 roughly 76 percent of AI Overview citations came from top-10 pages. By early 2026 published estimates put it between 17 and 38 percent, depending on the dataset.
Does classic SEO still matter for AI Overviews?
Yes, as a gate rather than a path. Being in the index makes you eligible, and 88.46 percent of ChatGPT citations came via the search channel. Being at the top of it is a separate question.
How do I make a page more likely to be cited?
Answer the heading in the first two sentences, write sections that survive being lifted out of the page, and include something a model cannot compose, such as a price, a sample size or a named method.
Is being cited in an AI Overview worth it?
Yes, measurably. Cited pages earned 2.07 percent click-through against 0.94 percent for uncited ones, roughly 120 percent more clicks per impression. Both still trail the 3.35 percent of a page with no AI Overview.
Does video get cited in AI Overviews?
More than most teams expect. YouTube URLs made up 18.2 percent of citations from non-ranking sources, which is 5.6 percent of all AI Overview citations. Google’s surfaces cite video far more than ChatGPT or Gemini do, so it is worth checking whether your buyers use them.
Will schema markup help me get cited?
No, on the best available evidence. A controlled study of 1,885 pages found minus 4.6 percent on AI Overviews and no significant effect elsewhere. It remains worth having for rich results in classic search.
Where to go next
Take one page that ranks well and is never cited. Read each heading, then read the two sentences underneath it.
If those two sentences do not answer the heading on their own, you have found the problem, and it is an afternoon of editing rather than a quarter of link building.
Learn what gets cited · Run the technical audit · Book a growth audit
Sources and further reading
Studies and primary sources
- Ahrefs. Analysis of 863,000 keyword SERPs and 4 million AI Overview URLs, drawn from Ahrefs Brand Radar, measuring cases where the same URL appeared in both the AI Overview and the regular SERP for the same query. 37.9 percent of cited URLs appeared within the first 10 blocks counting all SERP features, and 37.10 percent ranked in the top 10 counting organic results only. Positions 11 to 100 accounted for 26.20 percent and beyond position 100 for 36.70 percent, organic only. Among non-ranking citations, 18.2 percent were YouTube URLs, which is 5.6 percent of all AI Overview citations. The authors note their parsing methodology improved since the previous study, so earlier analyses may have undercounted citations. Link
- Ahrefs. Study of 1.4 million ChatGPT prompts. Citation rate by retrieval channel: search 88.46 percent, news 12.01 percent, Reddit 1.93 percent, YouTube 0.51 percent, academia 0.40 percent. Used here to distinguish index presence from ranking position. Link
- Ahrefs. Controlled study of 1,885 pages adding JSON-LD schema, matched against roughly 4,000 controls using difference-in-differences. Google AI Overviews minus 4.6 percent and statistically significant, AI Mode and ChatGPT indistinguishable from zero.
- Seer Interactive. Organic click-through by AI Overview state, 2025 averages: 3.35 percent with no AI Overview, 2.07 percent when present and citing, 0.94 percent when present and not citing.
- Google Search Central. AI search optimisation guidance, 15 May 2026. Confirms AI Overviews and AI Mode run on core Search ranking and quality systems with no separate AI index, and warns against providers guaranteeing rankings.
Further reading
- Pepper. What AI engines actually cite for the passage-level craft, the YouTube citation data for the video slice, and the visibility strategy that still works for the wider response.
- Deliberately excluded: any single authoritative top-10 overlap figure, since published estimates for the same period span roughly 17 to 38 percent across datasets. Also excluded, the claim that cited brands earn 35 percent more organic click-through, and the widely repeated figure that around 68 percent of US searches end without a click.
Latest Blogs
Most advice on this is inference dressed as fact. OpenAI documents exactly one thing clearly, which is access. Here is what is documented, what is inferred, and how to verify the rest for yourself.
Being named in an AI answer and having your page used as its source are different outcomes with different causes and different fixes. The gap between them is the most useful diagnostic in AI search, and most reporting collapses it into one number.
Every platform reports share of voice and almost none of them define it the same way. Here is the definition worth using, the three choices that change the number, and how to measure it yourself without buying anything.
Get your hands on the latest news!
