Skip to main content
← All guides

How AI search works

Google's AI Overviews and AI Mode are built into Search. They can fan a query out into related searches, retrieve supporting material, and generate a response with links; Google says the same fundamental SEO requirements still apply.

Search foundations first, generation second

Google describes AI Overviews and AI Mode as Search features that surface relevant links and may use query fan-out across subtopics and data sources. There are no additional technical requirements or special AI markup needed to appear; the ordinary Search foundations still apply.

That keeps the optimization target simple. Make the page crawlable, internally discoverable, useful, and available in textual form; keep structured data consistent with visible content. Google can then decide whether the page is useful as a supporting link for a particular response.

StageWhat decides it
IndexingCrawlable, renderable, judged worth storing
RetrievalRelevance to the fanned-out queries
GenerationWhat the retrieved passages actually say
CitationWhich passages the answer leaned on

Query fan-out

Google says AI Overviews and AI Mode may use query fan-out: multiple related searches across subtopics and data sources. That means the links in an AI response do not have to mirror the classic results for the exact wording the user typed.

This helps explain why a page can perform well for an exact query yet be absent from an AI response without implying a technical failure. The response may draw on related searches and supporting pages selected for different parts of the topic.

The practical consequence is not to manufacture every possible phrase. Cover the subject as completely as the user needs, using natural terminology, so the page can be relevant to the meaningful subtopics rather than optimized for one exact string.

No separate file to publish

Google says directly that you do not need new machine-readable files, AI text files, markup or Markdown versions, and that llms.txt is ignored by Search. The entry requirement is the ordinary one, which is why AI visibility work keeps collapsing back into SEO work.

Proposed tacticStatus
llms.txtIgnored by Google Search
A Markdown copy of each pageNot required
AI-specific markup or meta tagsNone exist for this
Content pre-chunked for machinesNot required by Google Search
A second version written for modelsCloaking; a spam-policy violation

It is worth being clear about why these keep appearing despite being ruled out. Each is cheap to implement, impossible to disprove from your own data, and arrives during a period when nobody wants to be the team that did nothing. That combination produces adoption regardless of evidence.

What actually changes, and what does not

Treating AI search as entirely new leads to wasted work; treating it as entirely unchanged misses two things that genuinely differ.

  • The unit of success changes. A citation is binary and varies between runs, so a single observation is not a measurement.
  • The searches used can change. Fan-out means supporting links may be found through related subtopic searches rather than the exact typed query.
  • The technical foundation stays familiar. Google says the same fundamental SEO practices remain relevant and no special AI markup is required.
  • The writing goal stays people-first. Clear, specific text helps readers and gives Search useful material without inventing a separate machine-facing version.

Where to look when you are not appearing

Because the pipeline has distinct stages, absence has distinct causes, and they are worth separating before changing anything.

  1. Confirm the page is indexed. If it is not, nothing downstream can happen and the rest of this list is moot.
  2. Review Search preview controls such as nosnippet, data-nosnippet and max-snippet if you intentionally limit what Google may show.
  3. Confirm Googlebot is allowed by robots.txt and by your CDN or hosting layer; Googlebot is the crawler control for Google's AI Search features.
  4. Search the subtopics the question implies, not the question itself, and see whether you rank for those.
  5. Read your own opening passage and ask whether it answers anything a model could lift out intact.

Only the last step is about writing, and it is the one people start with. The first four are ordinary technical checks, and they account for most cases of a page that never appears in generated answers at all.

Try it on your own site

Questions and answers

Do I need an llms.txt file?

No. Google says its Search systems ignore it, and that no new machine-readable file, markup or Markdown version is required for its AI features.

Should I rewrite content for AI systems?

Google advises against it. These systems handle synonyms and general meaning, and text chopped up for machines usually reads worse for people without improving retrieval.

Why do I rank well but not appear in an AI response?

Query fan-out is one possible reason. Google's AI features may issue related searches across subtopics, so the supporting links do not have to match the classic results for the exact phrase you typed.

What is the actual technical requirement?

Google says there are no additional requirements or special AI markup for AI Overviews or AI Mode. Follow the ordinary Search fundamentals: allow crawling, make important content discoverable and useful, and use preview controls deliberately.

Does nosnippet affect Google's AI features?

Yes. Google lists nosnippet, data-nosnippet and max-snippet as controls for limiting information shown from your pages in AI features in Search.

Is AI Search a separate SEO stack?

Not for Google. Google says AI features are built into Search and that its fundamental SEO best practices still apply. Other assistants can have different systems, so verify their documentation separately.