Google AI Overviews optimization

The short version

AI Overviews are the generated summaries at the top of a Google results page. They cite their sources, which makes them the most measurable part of AI search and the most familiar to work on — the levers are largely the ones SEO already uses.

Being cited in an Overview is not the same as being named by an assistant asked directly. The two read different things and move independently, and reporting them as one number is wrong about most of the picture.

What is an AI Overview, mechanically?

When Google decides a query warrants one, it does not simply summarise the top ten results. It runs its own follow-up queries behind your query, retrieves pages for each, and composes an answer with links to a few of them.

Three consequences follow, and all three change what you do.

You are competing for a citation, not a rank. Appearing on page one is necessary but not sufficient. Pages ranked fourth are cited while pages ranked first are not, routinely.

The queries behind the Overview are not your query. Google decomposes it into sub-questions. Coverage of the adjacent questions matters more than exact-phrase matching on the head term.

It is unstable. Overviews appear, disappear and rewrite themselves, including vanishing entirely for queries where Google decides an answer carries risk — health, finance and legal topics most often.

Which queries get an Overview?

There is no published list, and the set changes. What is observable:

• informational and comparison queries get them most often;
• transactional queries get them less — Google still prefers to show products and ads where money is close;
• regulated topics get them inconsistently, and often not at all;
• navigational queries almost never.

The practical implication for a local business: your “best X in city” queries are more likely to produce an Overview than your “book an appointment” queries. That makes Overviews an upper-funnel surface for most service businesses, which is worth knowing before you budget for it.

What is worth doing

Cover the sub-questions, not just the head term

If a buying question has five obvious follow-ups, a page that answers all five is a better retrieval candidate than five thin pages each answering one. Google is retrieving for the decomposed questions, so breadth within a page beats breadth across pages.

Practically: write the page as the buyer’s whole decision, not as one keyword.

Put the answer first

Overviews pull self-contained passages. A paragraph that only makes sense after three paragraphs of preamble is a poor candidate however well written.

The rule that works: the first two sentences under a heading answer that heading completely.

Use the question as the heading

Not Pricing but How much does an emergency appointment cost? This is how buyers phrase things and how the decomposed sub-questions are shaped.

Stay eligible

Overviews draw on the same index as ordinary results. A page that is blocked, slow, thin, or renders only in JavaScript that the crawler does not execute is not a candidate at all. This is ordinary technical SEO and it remains the floor.

Mark up what is really on the page

Schema helps a machine parse structure; it does not conjure authority. Mark up what a human can see and nothing else — the honest version is in schema markup for AI search.

Expect the click-through hit

An Overview that answers the question fully reduces clicks even when it cites you. That is not a reason to opt out — being absent from the answer is worse than being cited without a click — but it does mean traffic alone stops being a usable measure of visibility.

This is the reporting change most teams have not made yet. A page whose impressions rise while clicks fall may be winning, not losing.

What AI Overviews are not

Being in an Overview does not mean ChatGPT will name you, and being named by ChatGPT says nothing about Overviews.

When four assistants were asked the same questions in one market, they named 63 businesses between them and only four were named by all four. Treating “AI search” as one surface produces reporting that is wrong about most of it, which is why we measure and report each separately.

How do you measure Overview visibility?

This is the awkward part, and it is worth being precise about what is and is not available.

What Search Console gives you. Impressions and clicks for your site, including impressions that came from a results page carrying an Overview. What it does not give you: whether an Overview appeared for a query, whether you were cited in it, or who else was.

What you can infer. A pattern of impressions rising while click-through falls on informational queries is consistent with Overviews appearing. It is evidence, not proof, and it should be labelled as such in any report.

What you have to do by hand or by tool. Ask the queries, look at the result, and record whether the Overview appeared and which sources it cited. That is the only place the citation data exists.

What to record each time:

1. Did an Overview appear for the query at all?
2. Were you cited?
3. Who else was cited, and from which domains?
4. Did the Overview answer the question fully, or leave a reason to click?

The fourth matters commercially and is almost never tracked. An Overview that fully answers a question is a lost click even when it names you; an Overview that summarises and leaves the detail behind a link is a qualified visitor.

How Overview work differs from assistant work

The levers overlap, but the emphasis differs in a way worth planning around.

If your own site is strong and your off-site presence is thin, you will typically do better in Overviews than in assistant answers — and a measurement that reports both separately makes that visible immediately. That asymmetry is usually the most useful thing in a first report.

A worked example

From a measurement on one local market, business anonymised.

Search: top three for the obvious commercial keyword, stable for a year.

Overviews: cited on two of the six informational queries checked, both where the business had a page answering the sub-question directly.

Assistants: named in 38% of general recommendation answers, 0% of price answers.

The diagnosis. Own pages were doing their job; off-site description was not. The four sources cited for price questions listed the business once, without prices.

What was done. Listings corrected on all four sources; one page rewritten so the price question is answered in the first two sentences.

What moved. Price answers went from zero to roughly a third. General recommendation moved four points and was reported as no change, because four points is drift.

Frequently asked

Can I opt out of AI Overviews? Not selectively. Blocking Google’s crawler removes you from ordinary results too, which is a much larger cost than the clicks an Overview takes.

Does being cited in an Overview help my ranking? There is no evidence either way, and anyone claiming a direct effect is guessing.

Do Overviews use the same sources as ChatGPT? Sometimes, and not reliably. They are different retrieval systems reading different things.

Should I write pages specifically for Overviews? Write pages that answer buying questions directly. That serves Overviews, assistants and readers at once, which is why it is the only advice in this article we would give unconditionally.

A checklist for Overview work

Before writing anything

1. Check whether Overviews appear at all for your buying questions. Twenty minutes with your own queries.
2. For each one that does, record which sources were cited and whether you were among them.
3. Note whether the Overview fully answered the question or left a reason to click. This determines what a good outcome even looks like.

On your own pages

4. Use the buyer’s question as the heading.
5. Answer it in the first two sentences under that heading.
6. Cover the obvious follow-up questions on the same page rather than across several thin ones.
7. Check the page renders without JavaScript and loads quickly.
8. Mark up what is visibly on the page, and nothing else.

In reporting

9. Separate impressions from clicks and expect the two to diverge.
10. Never report Overview performance and assistant performance as one number.

A short glossary

AI Overview — Google’s generated summary above ordinary results, with citations.

Query fan-out — Google decomposing your query into follow-up questions and retrieving for each. Why breadth within a page matters more than exact-phrase matching.

Citation — a source linked from the Overview.

Zero-click — an answer complete enough that the reader does not visit any source. Common with Overviews, and the reason traffic stops being a usable measure.

Questions people actually ask

Can I stop Google using my content in Overviews? Not without removing yourself from ordinary results, which costs far more than the clicks an Overview takes. There is no partial opt-out worth having.

My impressions rose and my clicks fell. Is that bad? Possibly the opposite. An Overview citing you produces impressions without clicks. Check whether the queries where it happened are ones where an Overview now appears.

Do Overviews use my structured data? Google’s documented use of structured data applies to its surfaces generally. Whether it decides a citation is not published, and anyone stating it confidently is guessing.

Why did my Overview citation disappear? They rewrite constantly, and Overviews vanish entirely for some queries as Google adjusts which topics get them. This is normal and not usually about you.

Is being cited worth anything if nobody clicks? Yes, in the same way a recommendation is worth something. Somebody read your name in an answer to a buying question. It is simply not measurable as a session.

Should I build pages specifically for Overviews? Build pages that answer buying questions directly. That serves Overviews, assistants and readers at once, and it is the only advice here we would give unconditionally.

Where to start

Check whether Overviews even appear for your buying questions — twenty minutes with your own queries will tell you more than any article, including this one.

Then measure the assistant side separately, because it will not match. The free check does that for one business at no charge, the method explains how the questions are written, and what AI visibility is covers the distinction between these surfaces in full.

A note on tooling

Several tools now sell “AI Overview tracking”. What they can genuinely do is check, on a schedule, whether an Overview appears for a query and which domains it cites — which is useful and is the same thing you would do by hand.

What none of them can do is tell you why you were or were not cited. Treat a tool as a way to collect the observation cheaply, not as an explanation, and be sceptical of any product that reports a score rather than the citations themselves.

One thing to do tomorrow

Search your five most valuable buying questions and note which produce an Overview and who is cited. If none produce one, this surface is not where your budget belongs this year — and knowing that is worth as much as knowing the opposite. Twenty minutes, no tools.