Generative search optimization: what Google's AI surfaces need from you

Generative search optimisation is the work of getting your pages used, cited and named by search surfaces that generate an answer instead of listing links. On Google that means AI Overviews and AI Mode. If you came here looking for SGE optimisation, that is the same work under a retired name: the Search Generative Experience was the 2023 experiment, and the features it became are now AI Overviews and AI Mode. The practical news is good, because Google has documented how eligibility works, and it rests on things you already control.

What Google's AI surfaces actually are

AI Overviews is the generated summary that appears above results on some queries. Google's documentation on AI features states that Overviews are shown only when its systems judge them additive to classic Search, so a large share of queries never trigger one.

AI Mode is the conversational surface for questions that need comparison, reasoning or several steps, where a person would previously have run four or five searches.

Both may use what Google calls a "query fan-out": the system issues multiple related searches across subtopics and data sources, identifies more supporting pages while the response is being generated, and shows a wider and more diverse set of links than a classic result page would. That single mechanism is the most important thing on this page for a publisher, and the next section explains why.

What Google documents about eligibility

• There are no additional requirements. Google says the same SEO fundamentals apply, and that no special optimisation is needed for AI features.
• Indexing and snippet eligibility are the entry ticket. To be shown as a supporting link, a page must be indexed and eligible to appear in Search with a snippet.
• No new files or markup are required. Google states in those words that you do not need to create machine readable files, AI text files or special schema.org markup to appear in these features.
• Reporting sits inside Search Console. Sites appearing in AI features are included in overall Search traffic in the Performance report, under the Web search type.
• The controls are the Search controls. Robots directives for Googlebot govern access, and the preview controls nosnippet, data-nosnippet, max-snippet and noindex govern how much of your page can be shown. Google-Extended is the separate control for AI training and grounding in some of Google's other systems.

The practical consequence of the preview controls is worth stating clearly: data-nosnippet lets you exclude a specific block of a page rather than the whole thing, which is the tool to reach for when one section needs protecting and the rest should stay eligible.

What changes for a publisher

Before	Now	What to do about it
One query, one results page, ten links	One query fans out into several searches and a wider link set	Write pages for the sub questions, not only the head term
Position decided the click	Being quotable decides the mention	Answer in the first two sentences under a heading that is the question
Traffic was the only outcome	A named mention is an outcome with no session attached	Track mentions, calls, branded search and direct traffic too
Informational pages earned visits	Informational pages earn citations	Keep writing them, and add pages where a decision happens

On the click question, the public evidence is specific. A Pew Research Center analysis of 68,879 Google searches from the browsing records of 900 US adults in March 2025 found that users clicked a search result on 8 per cent of visits to pages with an AI summary, against 15 per cent without one. Google, for its part, states that clicks from results pages with AI Overviews tend to be higher quality, with users more likely to spend time on the site. Both of those can be true at once: fewer clicks, better ones.

The work, in order

1. Confirm eligibility. Indexed, crawlable, snippet eligible, no stray noindex or nosnippet on pages you want used. Check in Search Console's URL Inspection tool rather than assuming.

2. Find the sub questions. Because of the fan-out, the page that gets used often answers a narrower question than the one typed. Pull People Also Ask, your site search log and your sales calls, and write one page per sub question.

3. Make each page extractable. Question as the heading, answer immediately underneath, one topic per URL, important content in text rather than images or scripts.

4. Put checkable facts in. Counts, prices with the date checked, named sources. These are what a generated answer can safely repeat.

5. Keep structured data honest. Google asks that your structured data match the visible text on the page, and that your Business Profile and Merchant Center information is current.

6. Earn presence elsewhere. Generated answers draw on more sources than your own site, so directories, review platforms and trade press do real work here.

How to measure it

Three streams, kept separately. Search Console for impressions and clicks on the queries that now trigger a summary. A manual audit where you run your top twenty queries and record what appears and which links are shown, which is set out in how to check whether your business appears in Google AI Overviews. And a prompt set across assistants, because your buyers use more than Google.

The AI surfaces Google runs are only one of the places a generated answer now appears, and they are the one most likely to be gated by classic ranking. The rest are not, which means a business can be invisible in one and named in another on the same question. AI Knows Us reads several of those surfaces for the same question set, so the difference between them is something you can see rather than assume.

This page is about Google's generative surfaces. For the term debate between GEO, AEO and LLMO, see LLMO and how it relates to GEO, AEO and SEO. For which domains Google's AI answers actually cite, see which sources appear in Google AI Overviews.

Asked and answered: generative search optimization

Is SGE still a thing?

The name is not. The Search Generative Experience was the Labs experiment, and the features that came out of it are AI Overviews and AI Mode. If a tool or an agency is still selling SGE optimisation, ask them which surface they mean.

Can I opt out of AI Overviews and stay in normal search?

Google documents the preview controls, nosnippet, data-nosnippet and max-snippet, as the way to limit what is shown from your pages in Search, and those are the same controls that apply to AI features. data-nosnippet on a single section is the precise option, because it leaves the rest of the page eligible.

Does schema markup get me into AI Overviews?

Google states that no special structured data is required. Valid markup that matches your visible text is still worth having, because it helps every surface understand the page.

Why do my impressions look fine while clicks fall?

That is the signature of answers being given on the results page. Pull the query list, find where the two lines separate, and decide for each query whether you want the mention or the visit.

How is this different from optimising for ChatGPT?

Google's surfaces are built on the Search index, so ranking and indexing decide eligibility. ChatGPT, Claude and Perplexity use their own crawlers and indexes, so access for those named agents is a separate check. The content work is the same on both sides.

What to do first

• Run URL Inspection on your ten most important pages and confirm they are indexed with no snippet restrictions.
• Search your top twenty queries and record which ones return an AI Overview and whose links appear.
• Write one page for each sub question that appears in those Overviews and that you do not already cover.
• Export sixteen months of Search Console data and find the queries where impressions held and clicks fell.
• Add a monthly prompt run across two assistants, so Google is not the only surface you can see.