How to rank in Google AI Overviews

AI Overviews are built from Google’s ordinary results. Google says its generative AI features rely on its core Search ranking systems to retrieve relevant, up-to-date pages from the Search index, then show prominent links to the pages that support the information in the response^[1]. They may also issue several related searches across subtopics, a technique Google calls query fan-out, and while the response is generated its models identify more supporting pages, so an overview can show a wider and more varied set of links than classic results^[2].

Read those two statements together and “ranking in an AI Overview” means two things, both in your control to some degree:

1. 1Being among the best results for the question, or for one of the searches it fans out into. That is ordinary ranking, on more queries than the one the person typed.
2. 2Holding the passage that supports one statement in the answer. A link appears beside the information it supports, so a page wins its place by stating that piece of information plainly.

Two limits frame the work. An overview appears only when Google’s systems judge it additive to classic results, so on many queries it never appears^[2]. And AI Overviews and AI Mode may use different models and techniques, so the answers and links they show differ^[2]. Optimise for the overview on queries where one appears, and do not expect a link in one to carry over to the other.

To be shown as a supporting link, a page must be indexed and eligible to appear in Google Search with a snippet; there are no extra technical requirements^[2]. The site must also be included in Search generative AI features in Search Console^[1]. The checks that prove both, from robots.txt to snippet rules, are step 1 of this hub’s practical guide, and they come first: a page that fails them cannot be linked however good it is.

Check one more thing before you choose queries: where the feature exists. Google lists the countries, territories and languages in which AI Overviews are shown to all users^[3]. Work on queries in a market and language the list covers.

Start from your own data, not from a list of keywords. Search Console’s Generative AI performance report shows which of your pages are shown in AI Overviews and AI Mode^[4]; the Performance report lists the queries on which your pages already rank. What it does not give you is the overlap, so build the shortlist by hand:

1. 1Export the queries on which one of your pages ranks on the first page of results, for the market you sell in.
2. 2Keep the questions and comparisons: how, why, which, what is, best, versus. Those are where an overview is likely to appear.
3. 3Search each one in that market, signed out, and note whether an overview appears, which pages it links and whether one of them is yours.
4. 4Mark the queries that matter to the business and show an overview that does not link you. Those are your targets.
5. 5Keep the list to a size you can work through in a quarter, and repeat the check before and after each change.

The queries where you rank well and are not linked are the most useful. They show that the page is relevant to Google, so the gap is usually in the searches behind the overview or in the passage, not in the page’s standing. How to record these checks over time, and what the engines report on their own, is covered in how to measure results prompt by prompt.

For each target query, write down the searches an overview would plausibly run to answer it: the cost, the conditions, the alternatives, the next step. You cannot see Google’s fan-out queries, but you can see their trace. Each linked page in the overview usually supports one part of the answer, so the links tell you which sub-questions the answer covers and who answers each one.

Then search each sub-question yourself and note where your pages rank. The map below is a made-up example of the result, for an installer of heat pumps:

Every row ends in one of three gaps, and each has its own fix. A rank gap is ordinary SEO work on the page that should answer the sub-question. A content gap needs a section, not always a new page: put it on the page that already answers the parent question when the topic is the same. A passage gap is the subject of the next section.

Resist the obvious overreaction. Google says that creating separate content for every variation of a query, fan-out queries included, mainly to manipulate rankings or AI responses violates its scaled content abuse policy^[1]. The map is a list of answers your best pages should contain, not a list of new pages to publish.

An overview links a page beside the information that page supports. So the test of a passage is simple: if a reader asked the sub-question, could the overview lift one or two sentences from your page and state a fact with confidence? Vague passages fail it even when the page ranks.

• State the fact with its unit and its condition in the same sentence, so that the sentence is still true when it is quoted on its own.
• Keep one sub-question per section, under a heading that asks it. A section that answers three questions supports none of them cleanly.
• Keep facts current and dated where they change. Google describes grounding as retrieving relevant, up-to-date pages to improve the accuracy and freshness of responses^[1]; a stale price is a reason to choose another source.
• Say something the other linked pages do not. Google asks for non-commodity content with a point of view of its own, such as a first-hand review, rather than a summary of what is already available^[1].

An overview is not only text. Google says its generative AI features can bring in relevant images and video, and that pages following its image and video guidance are already prepared for them^[1]. For shops and local businesses, responses can include product listings and local business details, and Google names Merchant Center feeds and Business Profiles as the way to make products and services visible there^[1].

Google lists what you can ignore for its AI features: special files and markup for AI, chunking pages into tiny pieces, rewriting for every long-tail variant, inauthentic mentions and structured data beyond what rich results need^[1]. The full list, with the research beside it, is in the pillar’s section on what you can skip. One warning bears repeating here, because ranking promises are the commonest pitch in this field: Google says no third-party tool has access to its internal ranking or AI systems^[1].

Nothing happens until Google reads the page again. Google notes that crawling can take anywhere from several days to several months, depending on how often its systems decide a page needs refreshing, and that you can ask for a recrawl in Search Console^[2]. Request it for the pages you changed, then wait for the recrawl before reading the result.

1. 1Check the target queries again in the same market and the same way, and note whether the overview now links your page and which sub-question it supports.
2. 2Compare the Generative AI performance report for the changed pages with the weeks before the change^[4].
3. 3Read clicks with care. A click on a link in an AI Overview counts as a click in Search Console, and every link in the overview shares the overview’s single position^[5], so a page newly linked there can report a better position and a different click-through rate at once.
4. 4Update the fan-out map. Sub-questions change as the overview changes, and a gap you closed can reopen when a competitor publishes a better passage.

If an overview says something wrong about you, the fix starts on your own page, where the fact should be stated plainly. Google also takes feedback on each overview: a reader can report a problem and choose the kind of issue^[3]. Google itself warns that AI Overviews can and will make mistakes^[3], which is one more reason to keep the facts on your pages exact.