Optimizing for Google AI Overviews means making your pages liftable: indexed by ordinary Googlebot, each important question answered in one self-contained paragraph, with facts that agree with themselves everywhere else they appear on the web. There is no opt-in, no submission form and no special markup: AI Overviews cites from the same index as normal search. That makes the work more editorial than technical, and it explains why pages that rank well can still be invisible in the answer.
What are AI Overviews, and when do they appear?
AI Overviews are the machine-written answers Google places above the ordinary results, assembled by its Gemini models from pages in the regular search index and linked back to the sources they drew from. They appear most often on question-shaped, informational queries (how much, how to, what is, is it worth) and less on navigational or purely transactional ones. Google rolled them out in the United States in May 2024 and has since extended them to well over a hundred countries. The practical consequence is position: on a query that triggers an overview, the AI answer is read before any blue link, including the one in first place.
Is there an opt-in, or a markup that triggers it?
No. Eligibility is the search index itself: if a page can be indexed and snippeted, it can be cited. The controls that exist work in the opposite direction. Snippet rules (nosnippet, max-snippet, data-nosnippet) limit what Google may quote from a page, in overviews as in ordinary results. And Google-Extended, the robots.txt token often assumed to cover this, governs the use of your content for training Gemini models; it does not affect AI Overviews, which are fed by the standard Googlebot crawl. Two things follow: you can't block AI Overviews without hurting normal search visibility, and anyone selling a dedicated "AI Overviews schema" is selling packaging.
What actually gets a page cited?
The unit being selected is a passage, not a page. A citable passage answers one question in its first sentence, in plain language, and survives being read alone. A paragraph that opens with "as mentioned above" or leans on a table two screens up can't be lifted without distortion, so it isn't. Headings help more than most markup: models chunk pages by heading, so H2s phrased as the searcher's follow-up questions hand the system pre-cut answers.
Structured data (Organization, Article, FAQ) won't buy a citation, but it removes doubt about who is speaking. A claim corroborated on directories, reviews or other indexed pages is a safer bet for the model than one that exists nowhere else. And under all of it, the boring hygiene still decides: static or server-rendered HTML, clean 200 responses, no content that only exists after JavaScript runs. Every page on this site is hand-written HTML with no framework runtime. That is not an aesthetic position; it means nothing stands between the crawler and the words.
Do you still need to rank?
Mostly, but the relationship has loosened. AI Overviews draws heavily from pages that already perform in search, yet the cited links regularly include pages sitting outside the top organic results, because selection happens at passage level. Ranking first no longer guarantees the visible slot, and ranking fifth no longer disqualifies you from it. For a small firm that answers precisely, that is an opening: a clear paragraph can outquote a bigger competitor's vaguer page. Treat classic SEO as the entry fee and passage quality as the tiebreak.
What happens to your traffic?
Be honest about the trade. On queries where the overview answers fully, fewer people click anything: the zero-click pattern is real, and informational posts feel it first. Two things offset it. The clicks that survive are better: a visitor who arrives after reading a summary that named you is further along than one who clicked a blue link cold. And a citation has value with no click at all. For commercial queries, your name inside the answer is the new above-the-fold. If someone asks what a booking system for a Malta car-rental firm costs and the answer cites your pricing page, that is closer to a referral than an impression.
How do you measure it?
Search Console doesn't separate AI Overview impressions or clicks; they're folded into the ordinary web-search totals, so no report will tell you directly. Three usable signals. Watch query-level data for impressions holding steady while click-through falls: that often means an overview has appeared above you. Run your most valuable queries yourself each month, signed out, and record who gets cited. And watch branded search volume as the slow signal that answers are teaching people your name. We run this check on kiwastudio.com monthly, in a plain spreadsheet. The tooling will mature; the habit is what compounds.
Where to start
Pick the one query that brings your best customers, and the page that should answer it. Rewrite its opening paragraph until it answers the question alone, in two or three sentences a stranger could quote without context. Add the follow-up questions as headings. Check the page renders without JavaScript and returns a clean 200. Then check back monthly. It's the same sequence we run for client platforms (Princess Garage, Nova Drive, Dealer Pro), where one citation on a booking or dealer-software query outworks a page of rankings.
The answer box gets written with or without you.
Request a GEO audit