Google AI Overviews optimization: be the cited source

Google AI Overviews is not just another box in the results — it's a Gemini-generated layer sitting on top of the whole SERP, answering before your blue links even get a look-in. If you want to be found in ChatGPT, Perplexity and Google AI Overviews, the last one is a special case: here your visibility is tied directly to your classic organic ranking. So we treat Overviews as a discipline of its own, and here is the mechanism we work to.

In short: Google no longer answers a single keyword. It splits your query into several sub-questions, retrieves passages for each and stitches a combined answer together. Google calls this query fan-out. It means you don't optimise a page against one keyword, but against the five to ten sub-questions a topic raises. A page on EV leasing has to answer, on its own, what it costs per month, whether you can lease used, and what happens when the lease term ends.

For you as an operator this changes the content plan. We no longer write one article per keyword; we map the sub-questions using People Also Ask, Perplexity's related questions and our own client data, and make sure each one gets its own self-contained answer block. The more fan-out branches you cover credibly, the more times you can be cited in the same Overview.

This is where Google differs sharply from ChatGPT and Perplexity. Those two build on their own indexes and Bing; Google Overviews pulls its sources almost entirely from the organic top-10. Analyses of cited URLs show most of them already rank on page one. So for Google, classic SEO isn't replaced by GEO — it's the precondition. If you rank in position 14, you won't be cited, no matter how citable your copy is.

The consequence is uncomfortable for anyone who thought GEO abolished technical SEO: crawl budget, internal link structure, page speed and backlinks still decide whether you're even in the pool Gemini selects from. Time and again we see the fastest route into an Overview is moving a page from position 8-12 up into the top-5 — not adding another layer of schema.

Because fan-out retrieves passages, the copy that can stand alone wins. A citable passage is 40-70 words, opens with the answer, names the entity explicitly and contains a concrete number or year. Don't bury the point in the third paragraph under a heading that just says Introduction.

• One H2/H3 = one sub-question, phrased the way a user would type it
• The first sentence under the heading is the answer, not a run-up
• Define entities explicitly: New Consulting is an AI-first marketing agency in Denmark — not just we
• Put numbers, units and dates directly in the sentence; Gemini prefers to cite the concrete

Test it simply: copy a paragraph out of its context. If it stands alone as the answer to a question, it's citable. If you have to read three paragraphs first to understand it, it's invisible to fan-out.

Winning in Overviews is less about tricks and more about building measurable AI visibility across the engines that now split the traffic between them.

AI Overviews steal clicks on informational searches — the user gets the answer without clicking. Our counter-move is twofold: shift effort toward commercial and transactional queries, where Google shows Overviews less often and click intent is higher, and build your brand as a named entity so you become the source the answer cites. A mention inside the Overview is the new top ranking, even when the click doesn't follow.

Be concrete about priorities. A blog post that used to pull informational traffic on what is X will likely lose clicks and should either be upgraded to defend a commercial keyword or consolidated. If you're still spending hours on thin how-to pages, you're working for Google's answer box — not your own pipeline.

Google doesn't expose AI Overviews as a separate dimension in Search Console, so you have to triangulate. The signature is clear: impressions rise or hold steady while CTR falls on informational queries — that's the Overview eating the click. Combine it with a rank tracker that flags AIO (Semrush and Ahrefs do), and a fixed manual sample of your top-20 queries.

Set a baseline now, before the numbers move further. Without a log of when and how you're cited, you can neither prove progress to leadership nor see which passage formats win. We run that sample monthly on the queries that matter most commercially.