How Do You Show Up in Google AI Overviews?

"Showing up" in an AI Overview is two problems wearing one name: eligibility, which is mechanical, and selection, which is competitive. Google documents the first completely and the second only in principles. We have done both on a production site — a new glossary page was cited in the Overview for its term within days of shipping — so this page lays out the checklist we actually run.

What does Google require before a page can appear?

The documented bar, per Google's AI-features documentation: the page must be indexed and eligible to appear as a snippet, with no snippet controls blocking extraction. That is the whole eligibility spec — there is no AI Overview registration, no special file, no markup.

Most sites pass this table already, which is exactly why passing it earns nothing by itself. What AI Overviews are and how they behave is covered in AI Overviews explained; the rest of this page is about selection.

What actually gets a page selected as a source?

Nothing guarantees selection — and the honest evidence stack has three tiers. Platform-documented: Google's May 2025 guidance says AI features reward the same things as Search — unique, helpful, people-first content — with no special optimization. Study-supported: Seer Interactive's CTR data (2025) shows selection matters, with cited sites earning roughly 35% higher CTR than uncited peers while AIO queries overall lost 61–70% of clicks.

Then there is the tier we can speak to first-hand: structure. Overviews quote passages, not pages, so we build every candidate page around self-contained blocks — a 40–75-word direct answer under the H1 and question-shaped H2 sections that survive being lifted alone. Across our three production builds this is the pattern behind every extraction we can document, including the AI Overview win [our data]. It is proven in our fleet, not proven as an industry law; we mark the difference deliberately.

The test we apply to every candidate passage is simple: read it with the rest of the page covered. It passes when it names its subject (no dangling "it" or "this"), answers the H2's question in its first sentence, and carries at least 1 sourced number. It fails when it needs the paragraph above to make sense — the extraction process does not respect your transitions, and a dependent passage gets truncated mid-thought or skipped for a competitor's cleaner one.

What did the page that won a citation look like?

It was not a restructure — it was a new page, built citation-shaped from its first commit. Three days into our greenfield insurance build, we shipped a glossary page for the build's brand-generic core term: a quotable 1-sentence definition as the opening line, DefinedTerm schema, and primary-source-cited figures in the quotable sentences. Within days, manual checks of the live SERP showed it as a cited source in the AI Overview for that term — including an inline, hyperlinked brand mention in the answer body [our data].

We record that as an observation, not a controlled experiment. A new page on a new domain ships everything at once — structure, schema, sourcing — so nothing isolates a single cause, and Search Console has no AI Overview dimension to consult. The full narrative — what we observed and what we refuse to conclude from it — is written up in our first AI Overview citation win.

What doesn't help?

Schema, specifically, is not the lever the industry sells it as. Ahrefs tested 1,885 pages and found no causal lift in AI citations from structured data — which matches Google's own statement that no special markup drives AI features. Schema still earns its keep elsewhere; the honest accounting is in does schema help AI search.

The same discipline applies to every "AIO optimization" product promising placement: no vendor operates Google's selection systems, so no vendor can sell you a citation. If a tactic's evidence cannot be traced to a platform document, a named study, or logs someone actually operates, treat it as decoration.

How do you measure whether it worked?

Imperfectly — Search Console does not report AI Overview presence as a separate dimension, so you approximate. Our method: mine GSC for the question-shaped queries a page targets, watch for the impression-up-click-down signature on those queries, and pair it with manual checks of the live SERP. Our GSC miner automates the query side across builds [our data].

The mechanics, including what GSC genuinely cannot tell you, are documented in AI Overview impressions in Search Console. Whatever method you use, set it up before restructuring — a win you cannot measure is indistinguishable from no win. And expect noise: AI Overviews are non-deterministic surfaces that come and go on the same query across observations, so log trends rather than reacting to any single look.

When is this work not worth doing?

When your queries are transactional or your economics need the click. AI Overviews cluster on informational questions; a page that converts on "buy" queries gains little from citation work, and a publisher paid per pageview cannot bank a relative CTR advantage inside a pool that shrank 61–70%. We route that effort into conversion paths instead, and we would rather say so than sell restructures indiscriminately.

For informational pages that anchor your authority, the work compounds: the same extractable structure serves featured snippets, ChatGPT, and Perplexity simultaneously. That larger system — structure, crawler policy, measurement — is the subject of our generative engine optimization guide.