Make the page reachable, make the passage quotable, and make the business verifiable somewhere other than your own website. That is the whole strategy, and everything below is detail underneath it.
What has changed is that ranking well no longer gets you there by itself. It helps. It is not the entry requirement it was a year ago.
The Short Answer
An AI Overview cites about four sources. Digital Applied analysed 1,000 of them across ten query intents and recorded an average of 4.2 citations per overview.
Four slots, not ten. Either you are in the citation list or, for that query, you are effectively invisible — because the person reading the answer often never scrolls to the blue links at all.
The good news is that the four slots are not reserved for whoever ranks first. The bad news is that the same study found the top 1% of cited domains capture 47% of all citations. Wikipedia, Reddit, YouTube, the big publishers and the government and education domains take a large share of what is available before anyone else competes.
Source: Digital Applied, "1,000 AI Overviews Analyzed: Citation Pattern Study", April 2026 — 1,000 AI Overviews, 10 query intents, 4,200 unique cited URLs. Retrieved August 2026.
Ranking Is No Longer The Lever
This is the finding that reorders everything, and most advice on this topic has not caught up with it.
Ahrefs studied 863,000 keywords and 4 million URLs and found that the overlap between ranking in Google's top ten and being cited in the AI Overview fell from 76% to 38% within a year.
Read that again. A year ago, three quarters of AI Overview citations came from top-ten pages. Now it is a bit over a third. AI Overviews are doing their own retrieval rather than reading down the results page.
The Digital Bloom's 2026 position report puts numbers on what ranking is still worth: position one in Google carries a 33.07% citation probability in AI Overviews, dropping to 13.04% at position ten.
Ranking first roughly triples your odds versus ranking tenth. That is a real advantage and worth having.
But a 33% probability at position one also means two thirds of the time the page ranking first is not cited. Something other than position is deciding, and that something is what the rest of this article is about.
Sources: Ahrefs AI Overview citation study, March 2026 (863,000 keywords, 4 million URLs); The Digital Bloom 2026 AI Citation Position report. Retrieved August 2026.
What Actually Correlates With Getting Cited
Ahrefs analysed 75,000 brands and measured how strongly various signals correlate with AI visibility. The result reorders a priority list the industry spent a decade building around links.
| Signal | Strength | r |
|---|---|---|
| YouTube mentions | 0.737 | |
| Branded web mentions | 0.664 | |
| Branded anchor text | 0.527 | |
| Brand search volume | 0.392 | |
| Backlinks | 0.218 | |
| Content volume | 0.194 |
Source: Ahrefs analysis of 75,000 brands, Spearman correlations with AI visibility across ChatGPT, AI Mode and AI Overviews. Retrieved August 2026.
The three strongest signals are all off-site and brand-shaped rather than link-shaped. Backlinks — the metric most SEO dashboards still lead with — correlate at roughly a third the strength of brand mentions.
AirOps reported a related finding: around 85% of brand mentions originate from third-party pages rather than a brand's own domain. You do not build this on your own website. You build it by being talked about elsewhere.
These are correlations, not proven causes. Brands that get mentioned a lot on YouTube are also usually bigger, better known and more established — and any of those could be doing the actual work.
The honest reading is that these numbers tell you where to look, not what to do. Anyone presenting a correlation coefficient as an instruction is overselling what the data supports.
The One Study That Tested Cause
Almost everything published about AI citations is observational. One notable exception is a peer-reviewed generative engine optimisation study from Princeton, Georgia Tech and IIT Delhi, presented at KDD 2024, which actually changed content and measured what happened.
Its findings are the closest thing to evidence rather than pattern-matching:
- Up to 40% visibility lift from optimising content for generative engines overall
- 41% lift from adding statistics — the single strongest change tested
- Quoting credible sources and citing authorities added further gains
- Keyword stuffing reduced visibility. The old tricks do not merely fail here; they cost you
That last point deserves attention. A tactic that was neutral-to-mildly-positive in classic SEO is actively negative for AI citation. If your pages were written to hit a keyword density target, that is now working against you.
Source: "GEO: Generative Engine Optimization" — Princeton University, Georgia Institute of Technology and IIT Delhi, presented at KDD 2024. Retrieved August 2026.
Is Your Writing Actually Quotable?
The practical version of all of the above is a question you can test right now. Paste a paragraph from one of your own pages — ideally one that answers a real customer question — and see how it scores against the factors the research supports.
The default text in that box is deliberately terrible, and it is the kind of writing that fills most service pages. Run it, then load the good example and compare. The difference is not style. It is whether anything in the passage can survive being lifted out of context.
Page-Level Signals That Predict Citation
From the same 1,000-overview study, the page characteristics that tracked with being cited:
- Named sources in the body: 2.1× more citations. Pages that cite someone else get cited more. This is the finding most people find counterintuitive and it is consistent with the KDD result.
- Schema markup: 2.3× more citations. More on the tension in that below.
- Length over 2,500 words: 1.6× more than pages under 800. Not a reason to pad — a reason not to publish thin.
- Median cited page age: 14 months. Freshness is not the lever most people assume. Established pages are doing well.
Separately, Cyrus Shepard's Zyppy meta-analysis of 54 studies (May 2026) ranked factors by strength of evidence rather than by correlation size:
Source: Cyrus Shepard / Zyppy meta-analysis of 54 AI citation studies, May 2026, scored 0–10 by evidence strength. Retrieved August 2026.
URL accessibility ranks first. Before anything clever, the crawler has to be able to reach and render the page. That is a robots.txt check and a rendering check, and it is the least glamorous item on any list you will read this year.
The Schema Argument, Honestly
Here the evidence points two ways, and most articles pick whichever direction suits them.
Google says: no special schema.org markup is required for AI Overviews or AI Mode, and structured data should match the visible text on the page. That is Google's own published guidance for site owners.
Independent analysis says: schema-marked pages were cited 2.3 times more often across 1,000 AI Overviews.
Both can be true. The most reasonable reading is that schema does not unlock anything — but pages carrying accurate schema tend to be better organised in general, and that organisation is what gets quoted. The markup is a symptom of a well-built page rather than the cause of a citation.
Which is why we still ship it on everything, and why we would not sell it to you as a lever. There is more on where those claims get oversold on our AI search optimisation page.
What To Actually Do, In Order
- Check your robots.txt. Are GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended allowed? A surprising number of sites block them through a plugin default and never find out. Highest-evidence factor, five-minute check.
- Check the page renders without JavaScript. If your content only appears after a script runs, some crawlers see an empty page.
- Phrase headings as the questions people ask. Then answer directly underneath — the query-answer match factor, scored 9.2 by the Zyppy meta-analysis.
- Write self-contained passages. Roughly 40 to 120 words that make sense pulled out of the page entirely. Use the grader above.
- Put checkable facts in. Years, licences, response times, prices, named cities. The KDD study found statistics gave the largest single lift at 41%.
- Cite someone else. Named sources in the body correlated with 2.1× more citations. Linking out is not leaking authority.
- Earn mentions off your own site. The strongest correlations in the Ahrefs data are all off-site. Local press, industry associations, supplier pages, sponsorships, podcasts, YouTube.
- Fix your traditional SEO too. Position one still triples your odds against position ten. These are not competing projects.
No secret markup. No proprietary framework. No "AI SEO" product distinct from doing the work well.
Seven of those eight items are things a competent SEO programme should already be doing. The eighth is checking a text file. That is the honest shape of this discipline in 2026.
How To Measure Any Of It
You cannot rank-track AI Overviews. Answers are generated per session and vary by phrasing, prior conversation and location. There is no fixed list to read and no stable position to record.
What you can do is test prompts on a schedule. Build a fixed set of the questions your buyers actually ask, run them monthly, and log four things: whether you were named, how you were described, which competitors appeared alongside you, and whether the facts stated about you were correct.
That last one catches problems nothing else does. Assistants routinely quote hours, service areas and phone numbers a business changed years ago, sourced from a directory nobody remembered existed.
One more measurement caution from BuzzStream, who analysed 595 prompts: 76.1% of citations were unique to a single platform, and only 0.8% overlapped across ChatGPT, Gemini, AI Overviews and AI Mode. Visibility in one engine tells you almost nothing about visibility in another. Check them separately or do not bother.
Questions People Ask
Make the page reachable, make the passage quotable, and make the business verifiable elsewhere on the web. Practically that means allowing AI crawlers in robots.txt, writing self-contained answers of roughly 40 to 120 words directly under question-phrased headings, including checkable specifics and at least one named source, and earning mentions on sites you do not own. Ranking still helps but is no longer sufficient: Ahrefs found only 38% of AI Overview citations come from pages in Google's top 10.
No, and this is the biggest change of the last year. Ahrefs analysed 863,000 keywords and 4 million URLs and found the overlap between Google top-10 ranking and AI Overview citation fell from 76% to 38% inside a year. The Digital Bloom found position one carries a 33.07% citation probability while position ten drops to 13.04%. Ranking helps considerably. It is no longer the entry requirement it used to be, and pages ranking well outside the top ten do get cited.
The evidence points in two directions and it is worth knowing both. Google's own AI features guidance states that no special schema.org markup is required for AI Overviews or AI Mode, and that structured data should match the visible text on the page. Independent analysis by Digital Applied of 1,000 AI Overviews found that schema-marked pages were cited 2.3 times more often. The most reasonable reading is that schema does not unlock anything, but pages that carry accurate schema tend to be better structured in general, and that structure is what gets quoted.
Content that can be lifted out and still make sense. Digital Applied's analysis of 1,000 AI Overviews found pages with at least one named source in the body were cited 2.1 times more often, and pages over 2,500 words were cited 1.6 times more than pages under 800. A peer-reviewed study from Princeton, Georgia Tech and IIT Delhi presented at KDD 2024 found that adding statistics produced a 41% visibility lift, the single strongest change tested, while keyword stuffing actively reduced visibility.
Around four. Digital Applied's study of 1,000 AI Overviews recorded an average of 4.2 citations per overview. That is the whole shelf, which is why the competition for inclusion is sharper than for a page of ten blue links. The same study found the top 1% of cited domains capture 47% of all citations, so a small number of very well-known sites take a large share of what is available.
Not the way you track blue links. AI answers are generated per session and vary by phrasing, prior conversation and location, so there is no fixed list to read and no stable position to record. What you can do is build a fixed set of buyer questions, run them on a schedule, and log whether you were named, how you were described, and who appeared alongside you. Anyone reporting a precise AI position number is presenting a sampled estimate as a measurement.
The correlation data says yes, by a wide margin. Ahrefs analysed 75,000 brands and found YouTube mentions correlated 0.737 with AI visibility and branded web mentions 0.664, while backlinks reached only 0.218 and content volume 0.194. The three strongest signals were all off-site and brand-shaped rather than link-shaped. These are correlations rather than proven causes, but the gap is large enough that ignoring off-site mentions while chasing links is hard to justify.
No. AI Overviews are generated per query, vary between sessions and users, and depend on retrieval systems no agency controls or can inspect. What can honestly be done is remove every reason a system would leave you out: blocked crawlers, unclear identity, contradictory business details, content that cannot be quoted without context, and claims nothing else on the web supports. Any guarantee of AI placement is a promise about something the person making it does not control.
Want To Know What AI Says About You Now?
We'll run your buyer's questions across several assistants and send you exactly what came back — including when the facts are wrong or a competitor gets named instead. Then a robots.txt and structured data check on your own site.
Written, free, no call required. No guarantees offered, because nobody can honestly make them.
Get A Free Audit 📞 1-800-481-8638Further reading: our Omaha SEO overview covers how AI visibility sits on top of everything else, AI search optimisation takes apart the industry claims that do not hold up, and how long SEO takes explains why the two tracks run on different clocks. Every figure in this article is attributed to a named study with its sample size. No client data or case study results appear anywhere.