To rank in Google AI Overviews, you rank in Google Search: the overviews pick their sources from the regular index using the core ranking systems that already decide the blue links, so there’s no separate algorithm to work on. What changes is what the model does around the ranking: it runs a set of related searches behind the scenes, pulls pages for each one, and quotes from the ones that answer best. So the work is four things: be eligible and indexed, rank in the regular results, cover those related searches, and write pages a model can quote.
The evidence is Google’s own documentation, plus the citation datasets, mostly Ahrefs’. We run AI visibility optimisation every day, so the numbers below are the ones we’d actually act on. The measurement lives in Search Console, and it’s free.
How AI Overviews pick their sources
Google’s AI Overviews choose their sources from the regular Search index, using the core ranking and quality systems that already decide the blue links. That’s Google’s own description of the mechanism, from its AI optimisation guide: generative features are “rooted in our core Search ranking and quality systems”. The retrieval step has a name, retrieval-augmented generation, which Google also calls grounding: the model runs searches against the index, reads what comes back, and assembles the answer from those pages, with citations attached.
Eligibility is short enough to list. The page has to be indexed, it has to be eligible to show with a snippet, and the site has to be included in Search generative AI features in Search Console. If you’re in the regular results and your snippets show, you’re in the running for the overview too.
You can also switch yourself out, which is worth knowing even if you never do it. The familiar preview controls carry across unchanged: nosnippet, max-snippet, data-nosnippet, noindex. Google’s guidance on its AI experiences says it plainly: “More restrictive permissions will limit how your content is featured in our AI experiences.”
That’s the whole mechanism, and it’s why there’s no separate AI Overviews optimisation to invent. One body of work serves both the blue links and the overview, which is pretty good news if you’ve been doing this a while.
Which queries get an AI Overview
About 21% of keywords trigger AI Overviews, and the keywords that get one are almost all informational: 99.9% of them, in Ahrefs’ study of 146 million SERPs. Question queries get an overview 57.9% of the time, “why” queries 59.8%, and queries with seven or more words 46.4%. The keywords that trigger an AI Overview, in other words, are informational questions.
The practical step is boring and worth doing anyway: search your target query and look at the page. If there’s an overview there today, the work below is aimed at something real. If there isn’t one, the same work still serves regular Search and the chat assistants, so it’s rarely wasted, but an overview won’t appear because you optimised for one.
Rank in the regular results first
AI Overviews cite pages that already rank in Google: 76% of the 1.9 million citations in Ahrefs’ research were pages that also sat in the traditional top 10, and the median position of the top-cited URLs was 2. So the entry ticket is ordinary ranking. A page stuck on page three isn’t on the shortlist, and no amount of quotable formatting changes that.
Two honest caveats. It’s a correlation, so a top-two ranking doesn’t earn you a citation by itself; the overview still has to quote something, and a page that ranks but doesn’t answer the question gets passed over. And the top 10 is the bulk, not the rule: a quarter of citations went to pages outside it.
The contrast worth knowing: AI Overviews run on Google’s own ranking, so here your ranking work pays twice. The chat assistants are close to the opposite, in our measurements: in the 10x AI Visibility Refresh study, the correlation between Google rank and AI citation was basically zero (Spearman rho 0.000, p = 0.914), and 41 cited pages had no Google impressions at all.
Build coverage across the fan-out queries
AI Overview fan-out queries are the related searches the model generates alongside your keyword, and covering them separates the pages that get cited from the pages that merely rank. Query fan-out is Google’s own term for the set of concurrent, related queries the model generates to fetch more results. An AI Overview is assembled from what those searches bring back, so a page that only answers the main keyword misses most of the pool.
Google’s example is a searcher asking how to fix a lawn that’s full of weeds. The model fans out to “best herbicides for lawns”, “remove weeds without chemicals” and “how to prevent weeds in lawn”, and the overview is built from what those searches return.
The evidence that coverage matters comes from Joshua Hardwick’s research with SurferSEO, reported in Ahrefs’ post: pages that ranked across the fan-out queries were 161% more likely to be cited than pages that ranked only for the primary term, and the Spearman correlation between fan-out rankings and being cited was 0.77.
So the work is to answer the branches on the page that already ranks. If the overviews in your space pull from comparisons, prices and how-to steps, those become sections on your page, clustered and linked from it. The page then earns credit for the whole cluster instead of one query.
Building a thin page per fan-out query is the trap, and Google has named it. The scaled content abuse policy covers pages created “primarily to manipulate rankings or generative AI responses”, and the AI optimisation guide repeats the warning for fan-out variations specifically. One good page that answers the branches beats ten thin ones.
On the assistant side, this is what Omnipresence records: the searches each assistant ran before answering, and the pages those searches pulled back. The fan-outs are rarely the keywords you’d guess, so reading them beats guessing them.
Write pages that are easy to quote
You make pages easy for AI to quote by answering first: the first sentence under each heading states the answer, the support follows, and any passage lifted out on its own still makes sense. A retrieval system reads in chunks, and a chunk that doesn’t carry its own meaning doesn’t get quoted, no matter how good the page is. Easy-to-quote pages are, mostly, pages that answer the question at the top.
That’s testable, which is the useful part. Our 3 Numbers That Decide If AI Cites Your Page study frames it as chunk frequency, chunk score and extractable units: AI scores one chunk of a page at a time, so you benchmark your passages against the pages already being cited, and you make the facts extractable rather than woven into a narrative.
The title tag is the one element with a clean measured lift. Across 1,000 searches and five AI engines in our title tag study, adding the current year to the title raised citations by as much as 192%, while title length, brand suffixes and question titles did nothing. Two honesty notes: those were assistant citations, not AI Overviews, so treat the number as evidence of what a retriever rewards; and the lift came from genuine freshness, so it’s a reason to keep pages current, not to stamp a year on a stale one.
Google’s mythbusting is the part that saves you money. The AI optimisation guide says it directly: llms.txt and similar files are not used by Google Search, “chunking” content into small pieces is not required, rewriting content just for AI systems is not needed, and structured data is not required for generative AI search. So skip the llms.txt file, skip the AI-only rewrite, and keep schema for the rich results it actually earns.
Freshness is the part most sites under-work. In the 10x AI Visibility Refresh study, 98 posts were refreshed and nothing new was published, and citations went up 10x in 91 days. The pages were already good; they had gone stale, and the refresh brought them back.
Earn the mentions around your brand
Whether brand mentions help AI Overviews pick you is measurable, and Ahrefs has measured it: brand mentions correlated 0.664 with AI Overview visibility, YouTube mentions 0.740 and branded anchors 0.527. A study of the top 50 websites mentioned across 76.7 million AI Overviews put mentions on highly-linked pages at 0.70, all reported in Ahrefs’ research. Those are correlations: they say the pages around a brand move with its AI Overview visibility, and they don’t say which lever to pull first.
The practical read is that the ecosystem around your brand is part of your visibility, and the highest of those correlations is YouTube mentions at 0.740. So the videos, the roundups and the comparison pages in your space are part of where the overview reads from. Getting into them legitimately is the work, and it’s slow.
You can watch this happen in our Credibility Layer study, which recorded 200 ChatGPT reasoning sessions on one local query: 5,933 searches, 1,982 citations, and every recommendation traced back to a 30-search credibility audit of the pages around each business. Where the business ranked in Google barely moved the outcome, and Reddit, for all the noise it gets, was cited zero times. That’s the assistant side, not AI Overviews, but it shows what a retrieval system reads when it decides who to recommend.
So the step is old-fashioned: search your query, open the overview, and list the pages and channels it actually quotes. Those are the places to earn your way into, by being genuinely worth citing. Google’s guide says seeking inauthentic mentions isn’t helpful, which matches what the data shows anyway.
How to tell it’s working
You measure AI Overview visibility in Search Console, in the generative AI performance report Google added for AI Overviews and AI Mode, and it costs nothing. Google rolled the reports out to all websites worldwide as of August 31, 2026, and they measure your AI Overview visibility as impressions from generative AI features in Search, plus the pages that appeared, countries, devices and dates, from hourly to monthly granularity. What they don’t give you is clicks or citation positions, so you know you appeared, and you don’t know whether you were the quoted answer or a supporting link.
Since the overview is rooted in the same ranking as the regular results, the generative report tends to move with the regular one, and a page gaining regular rankings is usually worth watching there too.
The AI Overviews number is free, and Search Console is the right tool for it. What Google doesn’t show you is the four assistants, and they don’t follow Google’s ranking, so don’t read this report across to them.
Omnipresence records whether ChatGPT, Claude, Gemini and Perplexity recommend you or a competitor on a buyer question, which pages beat yours, and whether the lift came from your work or a model update. It doesn’t track AI Overviews or AI Mode; Search Console owns that. Your agent gets the playbook for the pages that lost, so the work list closes instead of sitting in a report.
FAQ
Can you opt out of AI Overviews?
Yes, with the same preview controls that govern your snippets: nosnippet, max-snippet, data-nosnippet and noindex. Google has said more restrictive permissions limit how content is featured in its AI experiences, so the controls you already know are the ones that work.
noindex takes you out of Search entirely, overview and blue links together, which is a big hammer for one feature. Worth knowing before you decide: Google’s own data says the clicks that do come from AI experiences tend to be higher quality, with more time on site.
Can AI-generated content rank in AI Overviews?
Nothing in the eligibility rules asks how the page was made. Sources come from the same index, so an AI-assisted page competes on the same terms as any other: indexed, snippet-eligible, useful enough to rank. Google draws its line at scale and purpose instead: mass-producing pages to manipulate rankings or generative AI responses is spam under the scaled content abuse policy.
Is ranking in AI Overviews the same as ranking in ChatGPT?
No, and the difference decides where your effort goes. AI Overviews mostly quote pages already in Google’s top 10, so ranking work pays twice there. The assistants barely follow Google’s ranking at all, so there the quotable page and the pages around your brand do the work, and you measure them separately from Search Console.
That’s the job. Be eligible, rank in the regular results, cover the branches of the query, and make the page easy to quote, then watch Search Console for the result. It’s the same work it always was, with one more surface reading it.