What 500 Test Queries Taught Me About How to Rank in AI Overviews

I spent a month running 500 test queries through Google and logging every source that ended up cited in the AI Overview. Same questions my clients' buyers actually type, run on clean sessions, the cited URLs copied into a spreadsheet by hand. I did this because almost everything written about ranking in AI Overviews is guesswork dressed up as confidence, and I wanted to watch the source selection happen 500 times instead of reading one more listicle about it.

500 test queriesTest Queries

What I found lines up with the bigger public studies, with one uncomfortable wrinkle the listicles skip. Here is the whole thing.

How to rank in AI Overviews, the short answer

To rank in AI Overviews, win a top-10 organic ranking for the query, then make one passage answer that query in its first two sentences. AI Overviews select sources through retrieval-augmented generation, so the system pulls from pages already ranking, leans on brands with heavy off-site mentions, and quotes the chunk that most directly answers each fanned-out sub-query. Rank first, get quoted second. Everything else is detail on top of those two moves.

What 500 test queries revealed about AI Overviews

Three patterns showed up over and over in my log. First, the cited source was usually already ranking on page one for the query, but not always, and the gap was wider than I expected. Second, the same handful of domains kept reappearing across unrelated questions, which told me brand familiarity was doing real work in source selection. Third, the passage Google quoted was almost never the whole page. It was one tight paragraph of content, often lifted from under a question-style heading.

None of that is my opinion. Ahrefs ran the version of this at scale and found 76% of AI Overview citations in the top 10, with a median position of 2. Then they re-ran it months later and the number dropped to 38%. Originality.AI landed in the middle at 52% of citations appearing in the top 10. Surfer, looking at the citation sources directly, found that 67.82% of cited URLs don't rank top 10. Those numbers disagree because they measure slightly different things, but together they kill the comfortable myth that ranking number one hands you the citation. Ranking gets your content into the room. It does not get you quoted in the AI Overviews answer, and that gap is the whole game.

It does not get you quoted in the AI Overviews answer, and that gap is the whole game.

How Google AI Overviews select sources

Here is the mechanism, because you cannot optimize a system you treat as a black box. When a query triggers an AI Overview, Google does not answer the words you typed. It runs a query fan-out, breaking your one query into a spray of related sub-queries, then retrieves passages for each and asks Gemini to assemble the answers with citations. This is documented in Google's own patent, query fan-out patent, which describes generating multiple synthetic queries to retrieve a diverse set of candidate documents. Surfer describes the same pipeline plainly: how AI Overviews retrieve and rank pages.

Two things fall out of that design. The fan-out is why topical depth beats a single optimized page. Surfer's analysis found a page is 161% more likely to be cited on top of the main term. And the passage-level retrieval is why your formatting beats your word count. Google quotes a chunk of content, not a page, which is the same machinery I covered in how content chunking actually works in AI search. If you have never thought about which paragraph of yours is the most retrievable, the AI Overviews system has already decided for you.

Rank your content in the top 10 of search first

The single most reliable predictor in my 500 queries was boring: the source was already visible in traditional search. SE Ranking's research puts hard numbers on it, finding AI-generated answers link to organic top 10 in 92.36% of cases. The AI Overview that Google currently shows for the query "how to rank in AI Overviews" says it out loud, claiming engines pull up to 90% of their citations from pages already on the first page of search. In other words, AI Overviews mostly recite the first page of search.

So the order of operations is not negotiable. If your content sits on page three of Google, no amount of AI-specific tuning rescues it, because retrieval-augmented generation is reaching into the same index that ranks you. This is also why the work splits cleanly. Getting into the candidate pool is a ranking problem. Getting named once you are in the pool is a different problem, the one I broke down in how AI Overviews decide which sources to cite. Treat them as two stages, because Google does.

Target informational, long-tail question queries

AI Overviews are not evenly distributed across search. They cluster on informational intent, and they cluster harder on longer questions. Ahrefs found AI Overviews appear on 57.9% of question queries get AI Overviews, while reason-based "why" questions trigger them most of all. Semrush watched the mix shift over a year, with keywords triggering AI Overviews going from 89% informational down to 57% informational as commercial and transactional queries got pulled in too.

For content strategy this is a gift, because it tells you where to aim. Map the long-tail, question-shaped versions of your topic and write content that answers each one on its own terms. Do not chase one head keyword and hope. The fan-out rewards a site whose content has already published the sub-answers, so build the cluster, not the hero page. This is the spine of the content strategy work I do for clients heading into AI search, and it is the cheapest informational advantage left on the board.

Structure your content so Google can lift the answer

Once you are ranking and targeting the right queries, the citation comes down to one paragraph. Surfer calls it the answer capsule, a short self-contained block placed directly under a question-style heading, and their data shows 72.4% of pages cited by ChatGPT contain one. My 500-query log said the same thing in plainer terms: the quoted content was a tight 40-to-60-word block that resolved the question before expanding on it.

So write content for extraction. State the relevant answer in the first one or two sentences under a heading, then support it. Keep one idea per paragraph so the chunk embeds cleanly. Use real question headings instead of clever ones, because the heading travels with the passage and sharpens the match. Lists help too, since Surfer found 78% of AI Overviews contain a list. None of this is writing worse. It is writing so your most useful sentence is not buried under three paragraphs of throat-clearing where no AI will ever reach it.

Build brand mentions and authority off your site

This is the part that surprised people on my team. Forget links for a second. The strongest correlation Ahrefs found with AI Overview visibility was not links or domain rating. It was branded web mentions at 0.664, well ahead of backlinks at 0.218. YouTube mentions correlated even higher in their brand study, and YouTube is the single most-cited domain in AI Overviews. In my own log, the domains that kept resurfacing across unrelated queries were the ones with heavy third-party coverage, not the ones with the prettiest on-page SEO. Those domains had third-party mentions compounding for years.

The practical move is to earn brand mentions on content other people already trust. Pitch the "best of" roundups in your niche, earn expert quotes, show up on the YouTube channels and forums where your buyers already are. These brand mentions are what the model reads as consensus, and they are the closest thing to a real lever you have for AI Overviews visibility. So treat it as a core channel, not a vanity metric. Surfer's fact study adds another lever, finding cited pages carried 38% more verifiable facts on average than uncited ones, so pack relevant data and named sources into your content. This off-site corroboration is the same trust signal behind the citation graph, and it is slower to build than a meta tag, which is exactly why it works.

Track your AI Overviews visibility instead of guessing

• Track queries that trigger an AI Overview
• Check if content is cited in the answer
• Check if content is only in the source list
• Track content's position among sources
• Monitor visibility changes after updates
• Treat AI Overview as a separate visibility layer
• Measure source selection

You cannot manage AI Overview visibility you never measure, and it behaves nothing like a normal ranking. The cited sources change on refresh, so a single check tells you almost nothing. Track which of your queries trigger an AI Overview, whether your content is cited in the answer or only sitting in the source list, and where you land among the sources. Then watch how that visibility moves as you ship changes.

My 500-query log was a manual version of exactly this, and it is tedious by hand. The point is to treat AI Overview presence as its own visibility layer, separate from organic search traffic, so you can tell whether new content is actually getting pulled into AI Overviews answers or just ranking quietly underneath them. Measure the source selection, do not assume it.

What content does not move the needle

Plenty of standard advice is noise here, so I will spend a paragraph killing it. Word count is not a ranking factor for AI Overviews. Ahrefs measured a near-zero 0.04 correlation between content length and citations, and Dan Petrovic's grounding research found Google's selection plateaus around 540 words, with pages over 2,000 words seeing diminishing returns. Padding content to hit a length target dilutes the very chunk you want quoted. Schema is the other overhyped lever. It helps Google parse and index your content, which feeds your traditional ranking, but Google's own guidance is blunt that there are no special requirements to appear in AI features beyond ranking well in search. Freshness theater, slapping a new date on an unchanged page, does nothing the underlying content does not earn.

The honest version of how to rank in AI Overviews

After 500 queries I do not have a trick, and anyone selling you one is guessing. Rank your content in the top 10 for question-shaped queries, write the relevant answers in a quotable passage near the top, and build enough off-site brand mentions that the model already knows your name. That is the whole job, and it maps almost perfectly onto good SEO done for humans, which is the point Google keeps making and the industry keeps ignoring. If you want the full off-page and structural playbook behind this, it lives in our AI search survival manual. The brands that win AI Overviews are not gaming a new system. They are the ones already worth quoting.

By Michael McDougald