How to Appear in Google AI Overviews: What Google Documents, and What It Says It Ignores

Appearing in Google AI Overviews requires no special optimization, according to Google's own documentation: a page must be indexed and eligible to be shown in Search with a snippet, and nothing beyond that. The levers sold hardest for this, llms.txt files, AI-specific markup and content chunking, are the ones Google states it ignores.

What Google documents as the requirements

Google names two requirements and no others. A page must be indexed, and it must be eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. Google's AI features documentation states there are no additional technical requirements, as of its December 2025 update. Everything else written about qualifying is inference.

Google's wording is flat: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." The optimization guide, updated July 2026, puts it the other way around: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

Two mechanics matter more than the requirements. Both AI Overviews and AI Mode may use what Google calls a query fan-out, issuing several related searches across subtopics, so the query you optimized for is often not the one that retrieves you. Overviews also, in Google's words, "often don't trigger".

The levers Google states it ignores

Four of the most heavily marketed AI-visibility tactics are ones Google's own guide says do nothing for Google Search: llms.txt and other machine-readable files, breaking content into small chunks, rewriting pages specifically for AI, and adding special structured data. On the files, the guide says they neither help nor harm, as of July 2026.

On the tactic sold most often, the guide is quotable: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them."

• llms.txt and other special files. Ahrefs measured 137,210 domains in May 2026: 97 percent of published llms.txt files got zero requests, and no AI bot requested one from a site that lacked it.
• Chunking. The guide says there is no requirement to break content into tiny pieces for AI.
• Writing specifically for AI. According to Google's guide, pages need no separate register for generative features.
• Structured data as an AI lever. Not required, and no special schema to add, though still worth keeping for rich results in ordinary Search.
• Any tool claiming inside knowledge. The guide's sentence: "No third-party tool has access to our internal ranking or AI systems."

One popular tactic is worse than useless. The guide states that spinning out a page for every variation of how people might phrase a question, fan-out queries included, violates Google's scaled content abuse policy when done primarily to influence rankings.

Ranking in the top 10 is not the gate, and no one can give you the number

A top-10 organic ranking is neither required nor sufficient for a citation, and the published share of AI Overview citations coming from top-10 pages ranges from 38 to 76 percent depending on who measured it and when. The direction is settled and the magnitude is not quotable. Treat any single figure with suspicion.

The most useful row is the pair at the top. Ahrefs published 76.10 percent in July 2025 and re-measured its own question at 38 percent in March 2026: same publisher, the answer cut in half. Either number is a snapshot, not a property of AI Overviews.

What survives all four studies is the part worth acting on: a large minority, and on some measurements a majority, of citations go to pages that do not rank top 10 for the query. A page that cannot crack position eight can still be the passage an Overview lifts.

Where on a page the citation is actually taken from

Position inside the page is measurable and large. Surfer's analysis of 100,000 AI Overview citation placements across more than 10,000 responses, data collected 2 June 2026, found 38 percent of all citations pulled from a page's first 100 words, up from 20 percent in its June 2025 sample.

Surfer also split out why. Pages whose first paragraph neither confirms nor answers the question asked were cited 23 percent of the time in its sample; pages that did both, 45 percent. Surfer sells a product built on that finding and says the data is correlational.

Much of what gets a local business named is not on its own site

Review and trust profiles carry measurable weight. Seer Interactive's study for Trustpilot analysed 804,491 AI responses across 1,926 brands and four platforms in a March 2026 window, over eight verticals including Home Services. Brands with no Trustpilot profile had a median AI citation rate of 1 percent; those with 1 to 13 reviews, 53.5 percent.

Two caveats come with it, both from the study. Trustpilot commissioned the work and the measurement is Trustpilot-specific, so it is not evidence about Google reviews. And the authors report 99.5 percent of Trustpilot citations arriving through ordinary organic ranking rather than any engine seeking Trustpilot out.

How to tell whether any of this is working

There is no separate AI Overviews report. Google states that sites appearing in AI features are counted in overall Search traffic in Search Console, in the Performance report under the Web search type. Clicks from AI Overview pages are not broken out, so an AI-visibility figure from Search Console is an inference.

The controls are equally plain. Because AI is built into Search, robots.txt directives for Googlebot are the access control, and nosnippet, data-nosnippet, max-snippet and noindex limit what a page can show. There is no AI-specific opt-out that keeps classic Search intact, as of December 2025.

So measure the inputs you can see: whether your definitional paragraphs are genuinely first on the page, and whether your Business Profile and directory entries say the same true things. Expect noise on any one check, because AI answers vary run to run.

Being the passage that gets lifted: what this looks like when it runs

Getting cited is a publishing discipline rather than a technical install, which is why it survives Google's mythbusting list. In a build it is one answer per page in the first hundred words, a named primary source behind every number, directory entries that agree with each other, and reporting that shows which pages earn visits.

Mirastart builds that layer for Charlotte businesses. Content built answer-first with its sources on the page, Business Profile and directory consistency, the booking and missed-call systems that catch the traffic when it lands, and reporting that puts pages next to booked work all run in production today. Retainers start at $3,000 a month and every account stays in the client's name.