How to appear in Google AI Overviews

What they are

An AI Overview is the generated summary Google places above the ordinary results for some queries. It answers the question in prose and cites a handful of sources as links.

Two consequences follow immediately.

The click economics change. A user whose question is answered in the overview often does not scroll. Informational queries that used to send traffic now frequently resolve on the results page. This is the "zero-click" problem, and it is the strongest argument for treating a citation — not a click — as the unit of value.

Being cited is a new kind of position. The sources named in an overview get prominence above the traditional first result. That is worth having even when the click does not follow, because it is an implicit endorsement in front of the whole audience for that query.

How Google assembles them

The important structural fact: AI Overviews draw from Google's existing index and ranking systems. They are not a separate corpus with a separate submission process. Google generates the summary from pages it already crawls, ranks and considers reliable for that query.

That means there is no AI Overviews API, no submission form and no opt-in. The route in is the route you already know, applied more strictly.

It also means classical SEO remains the foundation. If you do not rank on page one for a query, you are unlikely to be cited in its overview.

What appears to influence citation

Google has not published a ranking recipe, so treat the following as informed observation rather than doctrine:

1. Ranking for the query already. Cited sources overwhelmingly come from pages that rank well. This is the entry ticket.
2. A direct answer near the top of the page. Pages that state the answer plainly, early, in extractable prose get lifted more than pages that build to it slowly.
3. Question-shaped headings. Sections matching how people actually phrase the query.
4. Genuine expertise signals. Named authors with real credentials, cited sources, evidence of first-hand knowledge. Google's guidance on this is consistent and it applies here.
5. Technical readability. Server-rendered content, fast pages, valid structured data. If rendering is required to see your answer, assume it is not seen.
6. Freshness where the query implies it. Anything with a year in the query, or about a moving subject, is judged partly on recency.

What does not work

• There is no submission. Anyone offering to submit you to AI Overviews is selling something that does not exist.
• Schema alone does not do it. Structured data helps machines read you; it does not make Google choose you.
• Blocking Google-Extended does not remove you. That control concerns training use, not whether your pages appear in AI Overviews as a source. If your goal is to be cited, blocking it achieves nothing useful; if your goal is to stay out of training, understand the two are separate.

The measurement problem

Here is the awkward part, and it is worth stating rather than glossing.

AI Overviews are personalised, volatile and query-dependent. The same query produces an overview for one user and not for another, with different sources on different days. Search Console does not currently break out AI Overview impressions as a separate dimension, so you cannot simply look up how often you were cited.

This is why AI visibility is measured by sampling rather than by reading a dashboard: ask the question many times, record what comes back, and treat the result as a rate with a confidence interval rather than a fact. It is also why single anecdotes about AI Overviews — yours or a competitor's — carry so little information.

You can check the readable, technical half of this on your own site with a free scan, and our methodology sets out how the sampled half is measured.