AI search is any search experience that answers in generated prose instead of — or above — a list of links: Google's AI Overviews and AI Mode, ChatGPT search, Perplexity, Microsoft Copilot. AEO (answer engine optimisation) and GEO (generative engine optimisation) are industry names for trying to be a source those systems draw on. They are not terms Google uses, and neither is a separate discipline with its own markup.
The practical question is narrower than the vocabulary: when a system composes an answer about something you know, can it find your page, read it, trust it and attribute it? Almost everything that decides that is ordinary search work, done carefully.
What Google says a page needs
Google's documentation is unusually direct. To be eligible as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet. That is the technical requirement. Google states that you do not need new machine-readable files, AI text files or special markup, and that there is no special schema.org structured data to add.
The controls are the ordinary ones. noindex, nosnippet, data-nosnippet and max-snippet apply to AI features as they apply to regular results, and robots.txt rules for Googlebot are the crawl control.
Source: AI features and your website, Google Search Central.
Which crawler governs which system
This is where most confusion lives — including in something I published and have since corrected. Blocking a training crawler is not the same as leaving a search product.
| System | What decides whether your pages are used |
|---|---|
| Google Search, AI Overviews, AI Mode | Googlebot, plus the standard snippet controls |
| Gemini model training | Google-Extended — which Google states does not affect inclusion in Google Search |
| ChatGPT search results | OAI-SearchBot |
| OpenAI model training | GPTBot |
| Perplexity search results | PerplexityBot |
| Microsoft Copilot | Grounded in the Bing index, crawled by Bingbot |
Sources: Google's common crawlers; OpenAI's crawlers.
What makes a page usable as a source
- It can be crawled and indexed. The first filter and the one most often failed quietly — a robots rule, a
noindexleft over from staging, a canonical pointing somewhere else. That is technical SEO. - The content is in the HTML. Text that only appears after a script runs can fail to appear at all.
- It states things plainly. A definition in the first sentence, a number with its sample and date, a claim with its source. Padded prose gives a system nothing to lift.
- It has evidence nobody else has. Original measurement is the clearest reason to cite one page over ten similar ones, which is why this site publishes its own studies.
- The author is identifiable. A named person with a consistent profile, linked from every article, is easier to trust than an anonymous page. Structured data can describe that relationship, but only if it matches what is visible.
What I do not recommend
There is no "GEO schema". FAQ blocks invented to have something to mark up are what Google has already reduced. Hidden instructions addressed to language models are detectable and are a straightforward way to look manipulative. And llms.txt is a proposed convention, not a standard — Google's guidance above lists no file of that kind among anything its AI features need. None of these substitute for being worth citing.
What I have measured
In September 2026 I checked robots.txt on 34 Nepali websites for named AI crawlers. Twenty-seven had no position at all — details in the AI crawler study. Structured data, the part of a page a system can verify rather than infer, was absent from 16 of the same 34: the structured data study.
