AI Overviews Optimisation: How to Appear in Google's AI Summary

The Google results page has changed its face. Where a user once saw ten blue links, they increasingly find, right at the top, a paragraph written by AI that answers their question directly, with a few links to the pages it drew on. Since AI Overviews rolled out, launched in the United States in May 2024 and then extended to many countries, the challenge for a business is no longer simply to rank first, but to be one of the few sources Google names inside that summary. Here is how that choice is made, and how to influence it. This page is the English companion to our French pillar on optimisation Google AI Overviews.

What a Google AI Overview is

An AI Overview is an AI-written summary that Google shows at the top of certain results pages, above the classic links. Google composes a synthetic answer to the query from several pages in its index, then displays links to the sources it used. Appearing as one of those sources is the whole point of AI Overviews optimisation.

The feature has a short but fast history. Google first presented its integration of generative AI into Search in 2023, under the name Search Generative Experience, in a testing phase. In May 2024 the company announced the rollout of AI Overviews to the general public in the United States, promising to extend it to other countries. Since then, the AI summary has spread across a growing share of queries, and in 2025 Google added a dedicated mode, AI Mode, which pushes Gemini-assisted conversational search even further.

It is worth drawing a clear distinction. Not every query triggers an AI Overview: Google reserves the summary for questions where it judges that AI adds real value, typically comprehension questions or those that ask you to weigh several options. On a purely navigational or very simple query, the summary often does not appear. When it does appear, however, it occupies the top of the screen and captures the first glance. That is precisely where the new visibility battle is fought.

For a brand, being cited in an AI Overview is not the same as ranking well below it. It means appearing as a reference Google judged reliable enough to fold into its own answer. That cited-source status is what this page sets out to help you win.

How Google builds its AI Overviews

To build an AI Overview, Google leans on its existing index: it identifies the pages most relevant to the query, extracts their most useful passages, then a Gemini model writes a synthesis answer while linking the sources whose information it reused. Citation follows extraction: Google surfaces the pages it actually used to compose the summary.

The decisive point is the grounding in Google's index. Unlike an assistant that would query a third-party engine, the AI Overview draws directly from the corpus of pages Google has already crawled and indexed. The immediate consequence: all the classic technical SEO work remains an absolute prerequisite. If your page is not indexed, poorly crawled, blocked by the robots file, or if its content is only rendered through unexecuted JavaScript, it does not even enter the pool of candidates the model can reuse. Our analysis of AI crawlers and websites invisible to engines details the most common technical traps.

Once the candidate pages are gathered, the model does not copy an entire page. It lifts fragments that answer the question directly, then assembles them into a coherent answer. Google stresses this behaviour in its official documentation for site owners: there is no tag or special action to be eligible for AI features; the same practices of helpful, reliable content, designed first for people, make a page fit to be reused. In other words, the AI Overview rewards good content, not a technical trick.

This mechanism matches what academic research established independently. The founding study on the subject, GEO: Generative Engine Optimization, presented at the ACM SIGKDD conference in 2024 by a team from Princeton and the Allen Institute for AI, showed that effective optimisation for generative engines plays out at the passage level, not the page level, because these engines extract and reuse small blocks of text. The same study measured, on a benchmark called GEO-bench, that targeted editorial adjustments could raise a piece of content's visibility in generative answers by up to 40 percent, the most effective levers being the addition of cited statistics and the citation of named sources, all in clear writing.

What makes a page get reused: the 3 filters

A page reused in an AI Overview passes three successive filters: findability (being indexed and accessible to Google's crawlers), extractability (offering self-contained passages that answer a question directly), and credibility (named sources, cited figures, authority signals, up-to-date content). Failing a single filter is enough to stay out of the summary.

1. Findability filter
   Your page must be indexed and correctly crawled by Google. In concrete terms: no blocking of useful pages in the robots.txt file, content present in the rendered HTML and not hidden behind unexecuted JavaScript, a decent response time, a page well understood semantically. It is the prerequisite nothing makes up for, and it overlaps entirely with classic SEO fundamentals. That is also why working on AI Overviews always starts from a sound technical foundation.
2. Extractability filter
   The model must be able to lift a passage that stands on its own. A paragraph that opens with the direct answer to a question, understandable without reading what comes before, is reused far more than fluent prose where the information is diluted across several screens. It is the most actionable and fastest lever to put in place: it does not require more content, just better arrangement.
3. Credibility filter
   At equal relevance, Google favours content backed by named sources and verifiable data, with an identified author and a freshness suited to the topic. "According to a recent study" does not pass this filter. "According to the GEO study presented at ACM SIGKDD in 2024" does. These signals stem directly from E-E-A-T, the same principles that govern getting cited by ChatGPT and the other AI engines.

These three filters are cumulative and ordered. A beautifully structured page blocked from crawlers will never be reused. An indexed page drowned in an indigestible wall of text will not be extracted. An extractable page with no source whatsoever stays less credible than its sourced rival. Being reused in an AI Overview means passing all three, not excelling at a single one.

AI Overview vs classic ranking: what changes

Classic SEO rewards you with a position in the list of links; the AI Overview rewards you with a citation inside the written summary, above that list. Both share the same foundations (indexing, relevance, authority), but the AI Overview puts more weight on the extractable passage and the sourced figure. An excellent ranking helps you become a candidate; it does not guarantee being reused in the summary.

The most costly confusion is to believe that "being first on Google" is the same as "being in the AI Overview". Both goals rest on the same base, but diverge on what makes the final decision.

The most operational line is the last one: measurement. Search Console gives you your positions and clicks, but it does not tell you, query by query, whether you appear in the AI summary or who else is cited there. Tracking AI Overviews therefore requires a manual survey, or specialised tools that are still young. It is a constraint, but also an opportunity, because most of your competitors are not doing it yet. To place the discipline as a whole, read our deep dive on GEO vs SEO and our practical method to appear in both ChatGPT and Google AI Overviews.

The method to appear in AI Overviews

To appear in an AI Overview, you proceed in order: consolidate the technical SEO foundation so you are indexed and understood, restructure the content into direct answers by question, anchor each claim in a named source, build a coherent body of content that establishes topic authority, then measure every few weeks. This is the method Cicero Studio applies: GEO audit, editorial production, automated semantic meshing.

1. Consolidate the technical foundation and indexing

Before any editorial work, you check the foundation: indexed pages, a robots.txt that does not forbid access to useful content, content present in the HTML, a reasonable load time, clean markup. Because the AI Overview draws from Google's index, a page Google has not correctly crawled and understood can never be reused, whatever its content. It is the least glamorous and most indispensable step.

2. Restructure into direct answers

For each real customer question, you open the relevant section with a short, self-contained answer of two or three precise sentences. This page itself illustrates it: each section starts with a boxed block that answers the heading's question directly. That is exactly the format a model lifts to compose its summary.

3. Anchor each claim in a source

You replace vague phrasing with named, verifiable sources. This effort, tedious but rewarding, is precisely the one the founding academic study identified as among the most effective for gaining visibility in generative answers. An unsupported claim is reused less than a sourced one, because it is less reliable in the eyes of the model and the reader alike.

4. Build a coherent body, not an isolated page

A reused page is good. A network of contents that reinforce each other and signal your authority on a topic is what durably settles your brand inside AI summaries. You organise pages into topic clusters: a pillar that frames the subject and satellite articles that dig into it, linked by contextual internal links. That is the role of automated semantic meshing.

5. Measure, then iterate

You retest the target queries at regular intervals, from the region you are aiming at, note whether the AI Overview triggers and who it cites, and prioritise the content that stays absent. Measuring visibility in AI Overviews still requires manual work in 2026; that is an operational reality, not a lack of method.

This is exactly how Cicero Studio works: a GEO audit that measures your current presence in AI summaries, a human, AI-assisted editorial production that creates the missing content, and automated semantic meshing that makes it all work together. Agency-quality work, software-grade productivity. For context on adoption in our home market, see how AI Overviews now trigger on 86 percent of business queries in France.

The mistakes that leave you out of AI Overviews

The most frequent mistakes are: believing a good ranking is enough, unintentionally blocking crawler access or leaving content poorly indexed, drowning the answer in fluent text with no extractable passage, leaving claims unsourced, waiting for a miracle tag that does not exist, and targeting queries that do not trigger an AI summary at all.

Mistake 1: counting on ranking alone

Ranking well helps you become a candidate, but does not decide reuse. If your page offers no extractable, sourced passage, a competitor ranked slightly lower but better structured can take the citation in the summary.

Mistake 2: neglecting indexing and access

An overly restrictive robots.txt, content rendered only browser-side, a page never correctly crawled: each is a barrier that excludes the page from the candidate pool. It is the quietest mistake, because it does not show in the content itself.

Mistake 3: diluting the answer

A fine literary text where the information arrives in the third paragraph is bad for extraction. The model needs a block that answers, right away, an identifiable question. Form counts as much as substance.

Mistake 4: asserting without sourcing

Unsupported claims are treated as less reliable. Every figure, every fact should be able to lean on a named, verifiable source, ideally as a link. That is the heart of the credibility filter.

Mistake 5: waiting for a miracle tag or targeting the wrong queries

No special markup makes you eligible for AI Overviews; Google has put it in writing. And not every query triggers a summary: on purely transactional or navigational searches, the AI Overview is rare. Optimising a deep content piece for an immediate-purchase query is misplaced effort.

Measuring and checking your appearances

You measure your appearances by manually typing your customers' real questions into Google, from the target region, noting query by query whether an AI Overview triggers and whether your brand appears among the linked sources. You repeat the test regularly, because the trigger and the content of the summary change from one query to another, and also depend on language and region. This survey is the starting point of a GEO audit.

In practice, you draw up a list of ten to twenty questions your customers genuinely ask, submit them to Google, and note for each: does an AI Overview appear, is your brand cited as a source, who else is, and in what form. This is exactly the protocol I have verified by hand on dozens of sites, and we tested its stability by replaying the same queries several days apart. That survey gives an honest snapshot of your starting point, far more useful than an abstract score. You repeat it at regular intervals to track progress and identify the content to reinforce first. For the part specific to conversational engines, our pillar on getting cited by ChatGPT completes the approach, and our GEO audit method frames the full exercise.

Alexis Dollé

CEO & Founder of Cicero Studio

SEO and GEO specialist and founder of Cicero Studio, I have tracked visibility in AI engines since the first rollouts of AI-assisted search, testing dozens of sites by hand on their business queries, in Google as in the assistants. I confirm every observation with my own surveys before writing it here. This page is the synthesis of what I see day to day at Cicero Studio. Our conviction: appearing in an AI Overview cannot be decreed. It is built page after page, with method and sources.

LinkedIn →

What this page does not cover

In fairness, and because it is exactly the kind of transparency AI engines reward, here are the limits to know before building a strategy around AI Overviews. Stating what a method does not do is often worth more than overselling what it does: a forewarned reader, like a model assessing your reliability, trusts a claim that sets its own boundaries more.

Going further

We document our approach publicly, which is our best proof. Each resource below digs into a specific angle of visibility in AI engines: the mechanics of citation, the step-by-step method, the authority signals, or how to measure your results. Here are the most useful pieces to go deeper, depending on what matters most to you: