Open Google, type almost any commercial question, and the first thing you see now is not a list of links. It's a paragraph of synthesized text with three or four small source chips underneath. The user reads the paragraph, sometimes clicks a chip, more often doesn't. The blue links are still there, but they sit below an answer that already answered the question. That answer is the AI Overview, and the chips are the only piece of real estate most pages have left to fight for.
Two years into AI Overviews, the strategy question has narrowed. You're not trying to "rank above" the Overview — there is no above. You're trying to be the Overview, or at least be cited by it. Getting picked as a source means your page made it through Google's retrieval filter, looked structurally readable to the synthesizer, and offered language clean enough to quote. None of that maps cleanly onto the SEO checklist most teams still run. AI Overview optimization is a different game, and the teams winning it have stopped writing for the SERP and started writing for the synthesizer.
What an AI Overview actually does
To optimize for AI Overviews, it helps to understand what they are mechanically. The Overview is a generated paragraph (sometimes a few paragraphs, sometimes with bullets) produced by a Gemini-family model that has been handed a small set of retrieved pages and asked to synthesize an answer. The retrieval set is not your top ten organic results in order — it's a smaller set, sometimes pulled from Google's broader index, sometimes including pages that don't rank on page one for the same query. The synthesizer then writes an answer in its own voice, citing the sources it leaned on most.
Three things follow from this. First, ranking and being cited are different problems. A page can rank fifteenth and still get cited; a page can rank first and get ignored. Second, the synthesizer prefers content that is easy to lift sentences from — clear claims, defined terms, scannable structure. Third, the Overview's citations are not a popularity vote. They reflect what the model actually used while writing.
Most SEO teams still optimize as if they were chasing the first signal (rank) and have built almost nothing for the second and third (liftability, structural cleanliness). That's the gap.
What gets cited: a structural pattern
Look at twenty AI Overviews in your category and the citation patterns become legible. The cited pages share a structure, not a length, not a backlink profile.
They open with a direct definition or claim. The first paragraph of a cited page almost always contains a complete, standalone sentence that answers the query. Not a hook. Not a story. A sentence the model can lift verbatim. If your post opens with three paragraphs of context before the answer, the synthesizer often doesn't reach the answer — it cites the page that gave the answer in line one.
They use H2s that look like questions. "What is X." "How does X work." "When should you use X." The synthesizer is doing a retrieval-then-generation step, and the H2s help it both find the right span and decide which span maps to which sub-question. Pages with descriptive but non-question H2s ("The architecture of X") get cited less often for the same content.
They have one clean sentence per claim. AI Overviews lift sentences, not paragraphs. A 60-word sentence packed with caveats and parentheticals is hard to quote. A 22-word sentence with a single concrete claim is easy to quote. Cited pages have a higher density of liftable sentences than uncited pages on the same topics.
They show recency. Overviews systematically prefer recently published or recently updated content for queries that imply currency — anything tied to a tool, a pricing change, a regulation, or a year. A page from 2022 with the right answer often loses to a page from 2026 with a worse answer, because the synthesizer treats date as a soft confidence signal.
They publish into a clear topical context. A standalone article on an unrelated site gets cited less than the same article on a site that has obvious topical depth. Google's retrieval is happier handing the synthesizer a page from a site that clearly "covers this area" than a page from a site that wrote one post about it.
None of these are tricks. They're consequences of how retrieval-augmented generation works. The model is doing what any reader would do under time pressure: it picks the page that's easy to read, easy to quote, and looks like it knows what it's talking about.
The four-part page shape
Pages that consistently earn AI Overview citations tend to fit a four-part shape. You can apply this shape to a new draft or refactor an existing post around it.
Lede with the answer. The first 60 to 80 words should contain the answer to the query, stated as a complete sentence. Not "let's talk about" — the answer itself. Treat the first paragraph as if it has to stand alone in a featured snippet, because for AI Overviews, it does.
Define the terms. Within the first H2, define the key term or terms in the query. Use a sentence pattern the synthesizer can lift: "An AI Overview is a Gemini-generated answer that appears above Google's blue links on most commercial and informational queries." Definitions are disproportionately cited because they're useful in generated paragraphs.
Cover the sub-questions explicitly. Each H2 should map to a sub-question the synthesizer might need to answer. Use H2 phrasing that mirrors how users phrase the question in search — "How do AI Overviews choose sources" beats "Source selection methodology." Under each H2, lead with a clean topic sentence, then expand.
Close with an FAQ block. Two to six question-and-answer pairs, each with the question as bold text and the answer as one or two short paragraphs. The FAQ does two jobs: it captures sub-queries that didn't fit into the main flow, and it gives Google structured data to render directly when you mark it up with FAQPage schema. Cited pages almost always have an FAQ block; uncited pages on the same topics often don't.
This shape works because every part of it is built to be quoted from. The model never quotes a whole page. It quotes sentences, and the page that contains the most quotable sentences on the topic, in the cleanest order, wins the citation slot.
Signals beyond the page
The page is where most teams stop. The signals around the page are where the Overview citation often gets decided.
Topical depth on the same domain. A single post on AI Overviews from a site that also covers AEO, schema markup, citation tracking, and content structure outranks the same post on a site that wrote one piece about Overviews and otherwise covers something else. Google's retrieval treats topical clustering as confidence; the synthesizer inherits that confidence. If you want one post to be cited, the surrounding posts on the same domain matter.
Recency on the cluster. Overviews skew toward recent. That doesn't mean publishing weekly — it means showing a steady update cadence across the cluster. If your AI Overview post is six months old but you've updated three adjacent posts in the cluster in the last month, the cluster reads as alive. If everything is two years stale, the cluster reads as dormant and the citation goes elsewhere.
E-E-A-T signals on the author and publisher. Author bios with real expertise (a link to a profile, demonstrable history, named role) correlate with citation. Publisher signals — about page, contact, named team, real address if applicable — do the same. None of these are decisive on their own, but the absence of all of them is decisive in the wrong direction.
Mentions and citations off-site. Pages that other domains cite when they discuss the same topic accrue what amounts to a soft authority signal for AI Overviews. The mechanism is debated, but the pattern is observable: pages widely referenced elsewhere get cited more often, even when their on-page content is no better than uncited alternatives.
Schema markup. FAQPage and Article schema are the two that consistently correlate with AI Overview citation. HowTo schema helps for procedural queries. Schema isn't a magic switch, but it makes the page legible to Google's retrieval in a way that plain HTML doesn't.
The off-page signals are slower to build than the on-page changes, but they compound. A team that builds topical clusters with cross-linking, updates them on a cadence, and ships clear author and publisher pages will earn citations on posts the on-page work alone couldn't earn.
What to stop doing
Three habits actively hurt AI Overview performance, and most teams still do all three.
Stop burying the answer. The "story lede" pattern — three paragraphs of setup before the post says what it's actually about — was a remnant of the engagement-time optimization era. AI Overviews don't read for engagement; they read for the answer. If the answer is in paragraph four, the synthesizer often quotes a different page.
Stop writing for the long-tail-keyword density. Stuffing the post with variant phrases ("ai overview seo," "ai overview ranking," "ai overview optimization") used to help on Google's classic ranking signals. It doesn't help the synthesizer, and it makes the prose harder to lift. The cited pages are the ones with cleaner, less keyword-padded language.
Stop optimizing for "above the fold" engagement metrics. Hero images, video embeds, and interactive widgets at the top of the post push the answer down the page. The synthesizer reads the text, not the widget. If the first 200 words of readable text don't contain the answer, the page is harder to cite.
These habits aren't fatal individually. But they were optimized for an SEO regime that's been quietly retired. The new regime rewards the inverse: answer first, structure clean, signals quiet.
Measuring whether it's working
The hardest part of AI Overview optimization is measuring it. Google Search Console doesn't report AI Overview impressions as a distinct category. Click-through rate drops on cited pages because the answer often satisfies the user before they click. The signal you're looking for is usually invisible in your dashboards.
Three measurement workarounds make the work tractable.
Manual citation tracking. Pick ten target queries for the cluster. Once a week, run them in Google logged out and screenshot the AI Overview. Track whether your domain appears in the citations and which page got picked. This is slow but it's the only ground-truth signal.
Branded query volume. When AI Overviews cite you, brand familiarity goes up even if the click doesn't. Watch branded query volume in Search Console over the two to three months after you start optimizing. A slow upward drift is the trailing signal that Overviews are working.
Direct traffic on the cluster. Some users who see a citation will type your domain into the URL bar later. Direct traffic to specific URLs on the cluster (not your homepage) is the cleanest leading indicator that a page is being seen even when it's not being clicked.
None of these are perfect. Together they're enough to tell whether the work is moving the needle. Without them, AI Overview optimization is shadowboxing.
FAQ
Do AI Overviews replace SEO? No. They replace the part of SEO that was about chasing engagement-time metrics on shallow answers. The underlying disciplines — clear writing, topical depth, technical health, schema, recency — matter more, not less. The change is in what the work is for: citations and brand visibility, not raw clicks.
How long does it take to get cited in AI Overviews after optimizing a page? On established domains with topical depth, citations can appear within days of a refactor. On newer domains or for queries with deep competition, it can take weeks to months. The recency signal helps, but it doesn't override domain-level confidence.
Does Schema markup actually help AI Overviews? Yes, but indirectly. Schema doesn't force a citation. It makes the page easier for Google's retrieval to parse and characterize, which makes it more likely to be in the retrieval set the synthesizer sees. FAQPage and Article schema are the two with the clearest signal.
What's the difference between AI Overviews and featured snippets? Featured snippets quote a single source verbatim. AI Overviews generate a synthesized paragraph from multiple sources and cite the ones used most. Optimizing for snippets — short, clean answer paragraphs — also helps for Overviews, but Overviews reward broader topical coverage that snippets do not.
How do I know if AI Overviews are taking my traffic? Look for a pattern of stable impressions and declining click-through rate on commercial or informational queries. If average position is steady and CTR has dropped 20–40% over the past year on those queries, AI Overviews are likely intercepting the click. The page is still being seen — the click is being absorbed.