By being in Google's ordinary search index and being easy to quote. AI Overviews are assembled from the same index as normal results, fetched by ordinary Googlebot, so there is no separate AI crawler to court and no AI-specific markup to add. What decides the citation is whether a page states a clear, checkable answer to the exact question, near the top, in plain words. Ranking gets you considered. Being quotable gets you named.
A great many sites have blocked Google-Extended believing it keeps them out of Google's AI answers. It does not.
Google-Extended governs one thing: whether your content is used to train Gemini. AI Overviews are generated from the regular search index, crawled by regular Googlebot. The two are unrelated. Blocking Google-Extended costs you nothing and protects you from nothing you thought it did.
The practical consequence is better than it sounds. There is no new crawler to allow and no new file to publish. If Google can index you, Google can cite you.
Most writing about AI visibility is about ChatGPT, because that is the name everyone knows. By volume of questions answered, Google AI Overviews is the larger surface, and it sits on top of the search results your customers were already going to see.
It is also the one where existing SEO work is not wasted. Every other engine asks you to start from something close to zero. Here, the index position you already have is the entry ticket.
Overviews draw on pages that rank, but they do not simply summarise the top result. A page at position eight with one clean, direct sentence answering the question gets cited over a page at position two that buries the answer under six paragraphs of preamble.
This is the single most useful thing to understand about the format. The model is not ranking you. It is looking for a sentence it can lift and attribute, and it will take that sentence from whoever wrote one.
Answer the question in the first paragraph, before any context, in a way that still makes sense if it is pulled out and shown on its own. Most pages do the opposite: they build up to the answer, which is good writing for a reader who has committed and useless to a machine deciding what to quote.
Then be specific. Numbers, dates, prices, named places and named limits survive being quoted. Adjectives do not. "Fast" is unquotable. "Under 200 milliseconds at the ninety-fifth percentile" is a fact a model can repeat and attribute to you.
Structure helps for a mechanical reason: headings that are the actual question, short paragraphs, and lists give clean boundaries around an extractable chunk. A wall of text forces the model to guess where the answer starts and stops, and it often guesses somewhere else.
There is no AI-specific schema type. Ordinary structured data still earns its place, because it states your facts in a form that does not have to be inferred from prose, and a model that does not have to infer is a model less likely to get you wrong.
FAQPage markup is the most directly useful here, because it pairs a question with an answer explicitly. Organization markup matters for a different reason: it is how you state what your company is, rather than leaving an engine to work it out from your homepage copy.
Overviews reduce clicks. On a question that can be fully answered in three sentences, the person reads the answer and never visits anyone. Being cited on those queries buys you a mention and very little traffic.
The queries worth pursuing are the ones where somebody is choosing rather than looking something up. "Best X for Y", "X versus Z", "is X worth it". Those still send clicks, because a decision needs more than three sentences, and being one of the named options is worth considerably more than a visit from someone checking a definition.
Take the ten questions your customers ask before buying. Search each one and look at whether an Overview appears and who it cites. Where one appears and you are not in it, open your own page on that topic and check whether the first paragraph answers the question plainly. Usually it does not.
That rewrite is the highest-return work available on this surface, and it is the same work that helps everywhere else. The wider practice has a name, and it is worth confirming the engines can reach you at all before spending a week on copy.
No, and this is the most repeated mistake in the whole subject. Google-Extended controls whether your content is used to train Gemini. AI Overviews are generated from the ordinary Google search index, fetched by ordinary Googlebot. Blocking Google-Extended changes nothing about whether you appear in an Overview.
Only by using nosnippet or max-snippet, which also removes or truncates your normal search snippet. There is no control that removes you from Overviews while leaving ordinary search results intact. Almost nobody who asks for this actually wants the trade.
Being named in the answer is worth something even without the click, because the recommendation itself carries weight. But be honest about the trade: Overviews reduce clicks on informational queries. The queries where being cited pays are the ones where somebody is choosing between options, not looking up a fact.
No special markup exists. Ordinary structured data helps, because it states facts in a form that does not have to be inferred from prose, but there is no AI-specific schema type and anyone selling one is inventing it.
Citeon is in private beta. Tell us your domain and we'll be in touch when there's a place for you.