How to Appear in Google AI Overviews

An AI Overview is generated at query time from pages Google has already crawled, indexed and ranked. There is no separate submission, no separate index and no distinct set of eligibility criteria you can apply for. In practice, the pages cited in an Overview overwhelmingly come from the conventional results for that query or for closely related queries the system also considered.

This has one blunt consequence: if you do not rank for a query, you almost certainly will not be cited in its Overview. AEO work applied to a page with no organic presence produces no Overview visibility, because the page is never in the candidate set. Classical SEO is not obsolete here — it is the gate.

The second consequence is happier. Improvements that help you rank also feed the Overview pipeline, so the work compounds rather than splitting your effort into two competing programmes.

Google-Extended is a robots.txt control that governs whether your content can be used to ground and improve Google's generative AI products. It is not a crawler in the usual sense and it does not affect whether Googlebot indexes you or how you rank in classical results. Disallowing it withdraws your content from AI experiences while leaving your search presence intact.

For most sites the calculation is straightforward. If being named as a source in an Overview is worth more than the clicks you might lose to the Overview answering the question outright, allow it. Tools, calculators, local businesses and anything with a task at the end of the journey generally come out ahead, since the Overview cannot complete the task for the user.

Publishers whose revenue depends on the user reading the full article on their page have a real argument the other way, and it is a legitimate business decision rather than a technical mistake. Whichever way you go, write the rule explicitly and review it annually rather than inheriting a default nobody chose.

Patterns that recur across observed Overviews: the page answers the exact question rather than a broader topic; the answer appears as a compact, self-contained unit that can be lifted without editing; the page uses headings that mirror how the query is phrased; and lists, steps or tables carry the substance rather than long narrative paragraphs.

Freshness matters on anything with a time dimension, and stated expertise matters on anything Google treats as consequential — money, health, employment, law. On those queries an anonymous page competes badly against one with a named, credentialed author and a review date, regardless of how good the writing is.

There is also a negative pattern worth naming. Pages built primarily to capture a query, with padding around a thin answer, are consistently passed over. The extractor finds the answer, notices there is nothing else of substance to anchor, and prefers a source with more corroborating detail around the same claim.

Expect informational queries to lose clicks. If the entire value of the page is a definition or a single figure, an Overview that states it removes most reasons to click through. That is unpleasant but not mysterious, and no amount of markup reverses it.

Expect task queries to hold up. If the user needs to run a calculation with their own inputs, compare options against their situation, download something or buy something, the Overview functions as a referral rather than a replacement. This is the strategic argument for building tools rather than only articles: a tool cannot be summarised out of existence.

The practical response is portfolio management. Audit which of your pages exist purely to state a fact, and either accept their decline, merge them into a stronger page, or add the interactive or evaluative element that gives a user a reason to arrive.

Start with the fundamentals, because they gate everything: crawlable, server-rendered HTML; one canonical URL per topic with no www or protocol duplicates; an accurate sitemap; and pages that rank at all for their target query. If a page is not in the top twenty, the highest-value work is classical relevance, not markup.

Then apply the extraction layer. One question per page, stated in the title and the H1. A direct answer in the first hundred words. Question-form H2s. Numbers in tables. FAQ markup on genuine questions. Author, publisher and review date in both visible text and schema.

Then measure the right thing. Position tracking alone will mislead you, because a page can hold position three and lose half its clicks to an Overview it is not cited in. Track clicks and impressions together per query, and note which queries have grown an Overview, so you can tell a ranking problem apart from a format change.