How to Appear in Google AI Overviews

AI Overviews and AI Mode are Search features, not chatbots

An AI Overview is a generated summary that appears above or among the blue links for some queries, with links to supporting pages. AI Mode is a separate, more conversational surface inside Search where a single question can trigger a longer, multi-part answer. Google introduced both as extensions of Search rather than replacements, and described AI Mode as using a query fan-out technique that breaks a question into subtopics and runs multiple related searches at once (Google).

The distinction from assistants matters operationally. A model like Claude or ChatGPT may answer from training data, a browsing tool, or a partner index, and each has its own crawler policy. Google's AI features draw on the same crawl and index that powers ordinary results, which means the technical work you already did for Search — clean crawlability, server-rendered content, correct canonicals — is the same work that makes you eligible here. Nothing about google ai mode seo starts from zero.

The practical consequence: if a page is blocked in robots.txt, requires JavaScript that Googlebot cannot execute, or sits behind a login, it is not a candidate for citation. Eligibility is inherited, not granted.

• AI Overviews: generated summary with supporting links, shown on selected queries in normal Search results.
• AI Mode: conversational Search surface that decomposes a question into multiple sub-searches before answering.
• Chat assistants: separate systems with their own indexes and crawler tokens — different levers entirely.

What changes when an overview is on the page

The reason businesses care is behavioural. Pew Research Center analysed real browsing data from US adults and found that when a Google results page included an AI summary, users clicked a traditional search result link in 8% of visits, compared with 15% of visits on pages without one. Users clicked a link inside the AI summary itself in about 1% of visits (Pew Research Center).

Read that carefully before drawing the wrong conclusion. It does not say organic search is over; it says the value of a position shifts when a summary is present. On those queries, being named and quoted inside the overview is worth more than holding a blue link below it, because the summary is what most people read. On queries where no overview appears — and many commercial queries still show none — nothing has changed.

So the goal is not "rank first". The goal is to be the source the summary uses for the sentence that matters to your buyer: the pricing comparison, the eligibility rule, the definition, the shortlist.

What Google's documentation actually says about eligibility

Google's guidance is unusually direct here, and it rules out several things vendors sell. There is no structured data type, no meta tag, no feed and no submission form that makes a page eligible for AI Overviews. Google states that success in its AI experiences follows the same fundamentals as Search: unique, satisfying, people-first content, good technical health, and compliance with Google Search Essentials.

Two documented specifics are worth internalising. First, preview controls apply: if you set nosnippet, use data-nosnippet around a block, or cap max-snippet aggressively, that content will not be used in AI features. This is the one deliberate opt-out most sites can trigger by accident. Second, Google-Extended is a separate control governing Gemini app grounding and model training, and Google says it does not affect inclusion in Search or AI Overviews — blocking it will not remove you from the summaries in results, and allowing it will not add you.

Structured data still earns its place, but for the right reason: it helps Google understand entities, products, prices, availability and authorship. It is a comprehension aid, documented at schema.org and in Google's structured data guidelines, not a ticket into a generated answer.

The controls you actually hold

Strip away everything speculative and a short list of genuinely controllable factors remains. Each one is either documented by Google or a direct consequence of how extraction works: a model can only quote text it can fetch, parse and attribute.

The highest-leverage change for most sites is structural rather than editorial. Long, meandering pages that bury the answer in paragraph nine are hard to extract. Pages that state the answer in one clean sentence, then substantiate it with specifics — numbers, conditions, dates, named comparisons — give a generator something quotable and verifiable.

Write for fan-out, not for a head keyword

Because AI Mode decomposes a question into sub-searches, the unit of competition shifts from the keyword to the sub-question. A query like "best payroll software for a 12-person agency in France" is not one lookup; it fans out into pricing tiers, French payroll compliance, integrations, support in French, and minimum seat counts. Different pages — possibly from different sites — can win each fragment.

That argues for coverage and precision over volume. A page that answers one narrow question completely, with the constraint stated in the text (the country, the company size, the price band, the exclusion), is easier to match to a sub-query than a 4,000-word guide that gestures at everything. It also argues for keeping the answerable facts on the page rather than in an image, a PDF or a gated calculator.

This is the same discipline that drives AI visibility across assistants generally, and it compounds: the sub-question pages that get cited in AI Mode are usually the ones that get cited by ChatGPT and Perplexity too, because they solve the same extraction problem.

How to measure whether you appear

Ranking tools were built for a ten-blue-links page and do not tell you whether a summary named you. Google reports AI Mode activity within Search Console performance data under the Web search type, so impressions and clicks from those surfaces are folded into your existing reports rather than broken out as a separate channel. That gives you aggregate movement but not the answer text.

To see the answer text you need prompt-level sampling: run the questions your buyers actually ask, repeatedly, and record whether the generated response mentions you, which URL it cites, and which competitors appear alongside. Volatility is high — the same question can produce different sources on different days — so a single check proves nothing and a trend over weeks proves a lot.

That is the job Growli does: we run your buyer questions across Google's AI surfaces and the major assistants, track who gets named and cited, and turn the gaps into a prioritised list of pages to fix. How it works covers the methodology.

Whatever tool you use, log three things per question: were you mentioned, were you linked, and what was the sentence that mentioned you. The third is where the actionable diagnosis lives — a wrong price, an outdated positioning line or a competitor's framing repeated as fact.

What does not work

Three tactics keep resurfacing and none of them are supported by Google's documentation. Adding a bespoke schema type invented for AI does nothing. Stuffing an FAQ block with twenty questions you never answer properly does nothing except dilute the page. And publishing volume for its own sake works against you, since Google's guidance on people-first content explicitly targets scaled, low-value output.

There is also no paid route in. AI Overviews cite organic sources; ads are a separate placement. If a vendor offers guaranteed inclusion in AI Overviews, they are describing something that does not exist in Google's product.

The honest version of ai overviews optimization is unglamorous: be indexable, be snippet-eligible, answer one specific question per page in a quotable way, keep facts current and consistent across the pages Google can see, and measure whether the answers changed.

FAQ