How Google AI Overviews work

What AI Overviews are and when they appear

An AI Overview is a generated answer placed above the ordinary results, with supporting links inside it. Google does not publish a trigger list, and it does publish a criterion.

A judgement call the system makes per query.

Google search results, question queryThe order a buyer meets, top to bottom

The order, and where the credit sits. Google is quoting a sentence, so a sentence is the unit you are competing for.

Four ads, then Google’s own answer, and the first ordinary result is the sixth thing a buyer meets.

The fourth question that box asks, what triggers one on a given query, is the one Google does not answer. Google gives a criterion and not a trigger list, and the difference matters: a criterion cannot be checked against your page before you publish it.

For the newer AI Mode surface, Google has described the mechanism a little further. It uses a query fan-out technique, issuing multiple related searches at once across subtopics and data sources. Where it does not have high confidence in helpfulness and quality, the response will be a set of web search results instead. So the fallback is the old page.

On scale, Google gives reach and not frequency. AI Overviews are now used by more than a billion people. That is a different claim from how often they appear, which Google does not publish.

What the independent studies measured about being cited in these answers, and where they disagree, is collected in AI search engine optimization.

What makes a page eligible for an AI Overview

Google does not publish how often they appear. It does publish who can be in one, and that part is settled. This is the part with a published rule, and the rule is shorter than the industry built around it. Eligibility here is ordinary Search eligibility, stated in one sentence and followed by a second that removes a whole category of advice.

Indexed, and allowed to show a snippet.

The eligibility rule, on the page that sets it

1. 1The sentence is in bold in Google’s own callout. The line is not buried, and it is not a paraphrase of anything.
2. 2The scope is named twice on this page: AI Overviews and AI Mode. Nothing here is a claim about ChatGPT or Perplexity.
3. 3What it rules out is a separate technical checklist. What it does not rule out is everything ordinary indexing already asks of you.

Here is Google’s wording. To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. Then, on the same page, comes the sentence that removes an entire category of advice. There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.

Four blunt controls

Google names four: nosnippet, data-nosnippet, max-snippet and noindex. None of them is specific to AI Overviews. The first three restrict what can be quoted from the page while it stays indexed and can still rank; the fourth removes the page from Search altogether. So the trade is real but narrower than it is usually described: you keep your position and lose your snippet, in the overview and in ordinary results alike.

A separate control, Google-Extended, governs AI training and grounding in some of Google’s other systems, not the appearance of supporting links here. Other platforms draw the same line in their own documentation. OpenAI separates OAI-SearchBot, which surfaces sites in ChatGPT search, from GPTBot, which serves the foundation models. Perplexity separates PerplexityBot from its user-initiated fetcher and states the first is not used for training. Confusing the two is the most common way a store blocks the wrong thing, and AI search engine optimization sets out the difference.

Three mistakes from treating this as a channel

An AI Overview is a surface, not a channel. It has no account, no bid, no report of its own, and each of these mistakes follows from acting as if it did.

How to buy separate work for it

Nothing separate is required, and the eligibility rule is the ordinary one. Work sold specifically as AI Overview optimisation is, on Google’s own account, either technical search work under another name or something the platform has not asked for. That does not make the work worthless. It makes the label misleading, and the label is what gets priced.

How to try turning it off

Google states plainly that features cannot be turned off. What exists are snippet controls, and none of them applies to the overview alone. A brand that dislikes being summarised can stop the quoting with nosnippet and keep its ranking, and it loses its snippet in ordinary results at the same time. The other two are narrower: max-snippet caps the length everywhere, data-nosnippet excludes only the part you mark. That is a real remedy with a real price, and none of it is an opt-out.

Reading a click drop as an overview problem

Compare January 2025 with January 2026 in one vendor’s panel. Classic organic click share fell in all four categories measured, by between 11 and 23 points. Paid share rose in each. Search Engine Land, covering the same analysis, reported AI Overview presence ranging from 12 to 33 percent by vertical. Two single months are not a time series, and at least two things moved at once, so attributing the fall to the overview is a guess. If the paid share is what moved, that is a different page, SEO vs PPC, and the wider practice in AI search engine optimization.

What to do about them in order

Start with the things Google names as requirements. The rest are only worth doing once those hold. None of it is specific to this surface, and Google says so itself. For a store the underlying work is the eight ecommerce topics Google publishes, six of which are structure and data.

1. Confirm the page is indexed and can show a snippet

   That is the whole eligibility rule. A page carrying nosnippet is not eligible, and a page that is not indexed is not eligible for anything. On a storefront some of this is set by the platform before you arrive, which is what Shopify decides for you. The wider technical baseline those pages start from is measured every year by the Web Almanac, and it differs by platform, not by effort.

2. Check you have not blocked the wrong crawler

   Google-Extended governs training and grounding in other systems, not appearance here. Blocking it does not remove you from AI Overviews. Blocking Googlebot in robots.txt is not removal either. Google states that a blocked page can still appear in results if other pages link to it. That is why noindex and a block are different instruments.

3. Write the answer to one question completely, in one place

   A paragraph that answers a specific buying question on its own is quotable on its own. That is our reading and not something Google says. Nobody puts it in those terms, and it is ordinary editorial work in any case: what Google rules out is a separate requirement, not the business of answering clearly.

4. Fix the technical requirements instead of adding markup

   Google states there is no special schema to add. That covers this surface only. Product structured data is still worth adding for what Google says it does buy: a richer appearance in Search, shown at each experience’s discretion. Crawlability, indexation and snippet eligibility are the requirements, and an ecommerce SEO audit is how you find out which of them your store fails. Speed is not among them. Core Web Vitals, the one measure here with a published pass mark, is graded separately, and Google never names it as a condition here.

5. Measure impressions and clicks separately

   An overview can cite you without sending a click. And a visit that starts with a citation and ends days later gets credited elsewhere entirely. Shopify’s marketing reports default to last click with a 30-day lookback. Impressions holding while clicks fall is the shape to look for, and it is visible in Search Console without any AI-specific tool.

Why none of it is specific to this surface

None of those steps is specific to this surface. Google makes that point repeatedly. The work that makes a page eligible for an overview is the work that makes it eligible for a result, and ecommerce SEO is where that sits for a catalogue.

The line-by-line version of that check, with what has a published pass mark and what does not, is an ecommerce SEO audit.

What AI Overview work costs

None of that work has a line on an invoice, because there is nothing here to buy. There is no media cost. AI Overviews are not bought, and no product exists to buy your way into one. What has a price is the work sold under the name, and the useful question is what that work would be called otherwise.

A new label on an existing line item.

Understand the label before you read a quote that uses it: answer engine optimization is an industry term, not a platform one, and none of the companies building these answers uses it.

One further label to be careful with: the work is usually sold under answer engine optimization, which is an industry term, not a platform one. The definition it travels under, structuring content so AI platforms select it as a cited source, comes from vendors. It rests on third-party studies, not on platform documentation. What we would ask a supplier: which of these lines would still exist if AI Overviews were switched off tomorrow. The ones that would are the ones worth paying for, and the ones that would not are being priced on a label.

Under whichever label, the work on your pages is our AI visibility service.

How to tell whether it is affecting you

There is no AI Overview report in Search Console. There are three numbers in the ordinary data worth watching, and none of them is a diagnosis on its own.

1. Impressions against clicks on your best informational queries. Impressions holding while clicks fall is consistent with something above you answering, and it is consistent with more ads above you too.
2. Position against click-through rate for the same queries. Holding position while the rate falls says the page above you changed, not your ranking.
3. The split between informational and transactional queries. Google describes AI Overviews as appearing where its systems judge generative AI especially helpful, and does not publish which query types those are, so this split is a reasonable thing to check, not a rule.

Google publishes no rule about which query types trigger an overview, so your own report is the only evidence about your store. Three shapes of demand, three different amounts of concern:

A Search Console performance viewImpressions climbing while clicks stay flat

A flat click line under rising impressions is consistent with an answer above you and with more ads above you. The report cannot tell the two apart.

• Mostly product and category queries. Check it, but expect less. There is no published rule either way.
• Mostly buying-guide and comparison content. Check this first. Buying guides and comparisons are the content most like an answer.
• A mix, which is most catalogues. Split the report before deciding, because an average hides both halves.

Whether this is your problem

The third is where a lot of ecommerce panic is probably misplaced, and probably is the right word, because Google publishes no rule about which query types trigger an overview. A store whose traffic is product and category queries sits on a different part of the surface from a publisher answering questions. AEO vs SEO is where that budget question gets settled.

Search three of your top informational queries before you buy anything to fix this, and note whether an AI Overview appears at all. Three queries on one day is a spot check, not a survey, and Google publishes no rule about which queries trigger the feature. So a clean result buys you an order of work, not an all-clear. The check is still worth ten minutes: if none of your three shows an overview, this is not the first thing to spend on.