Answer · checked 31 August 2026
How do I make my site work with Google's AI Mode?
Google's AI Mode does not visit your site to answer a question. It breaks the question into five to eleven sub-queries, runs each against Google's existing search index, and assembles an answer from the passages it gets back — so a page that is not already indexed and crawlable by ordinary Googlebot cannot be cited at all, no matter what else it publishes. The work is therefore in two parts: be in the index, and be made of passages that answer one question each in plain server-rendered text.
Colour International LLC · 31 August 2026 · colourintl.com/learn/google-ai-mode
What AI Mode actually does
AI Mode uses a technique usually called query fan-out. One typed question is decomposed into several narrower ones, each is run as its own retrieval against Google's index, and the model composes a single answer from the strongest passages returned across all of them, citing the sources those passages came from.
Two consequences follow, and both are the opposite of what the folklore says. First, retrieval is passage-level, not page-level — a long page that covers eight topics can be cited for one of them and ignored for the rest, and a page that buries its answer under four paragraphs of preamble competes badly against one that leads with it. Second, because the fan-out is against the index rather than the live web, there is no fetch of your server to answer the question. Files you publish for agents matter only insofar as they were crawled and indexed earlier.
The claim you will see everywhere that is wrong
A common version of this advice says AI Mode "fetches a handful of files and decides who to cite," and that inclusion requires allowing Google-Extended in robots.txt. Both halves are false and it is worth being specific about why, because acting on them wastes the effort.
AI Mode does not fetch files per question; it retrieves from the index, as above. And Google-Extended is not a crawler and does not gate AI Mode or AI Overviews. It is a token controlling whether your content is used for Gemini model training and grounding on some Gemini surfaces. Eligibility for AI Mode follows ordinary Googlebot access. A site can disallow Google-Extended entirely and still be cited in AI Mode; a site that disallows Googlebot is invisible to it no matter how many manifests it publishes.
This site disallows AI training in its own content signals and is still indexed and cited normally. Those are different levers and confusing them is the most expensive mistake in this area, because it points the work at the wrong file.
What actually moves the needle
Serve the answer without JavaScript. Retrieval works from what the crawler received. If the substance of a page arrives after hydration, the indexed passage is the shell. This is the single most common failure we measure and it is invisible in a browser.
Put one answer at the top of one page. A passage is chosen because it stands alone. Lead with the direct answer in a paragraph that makes sense lifted out of context, with the subject named rather than pronouned, then show the working underneath.
Give it something checkable. Models cite what they can attribute — a number, a date, a named sample, a method. A page asserting that AI readability matters competes with ten thousand identical pages. A page reporting that ten named platforms scored a median of 54 out of 100 on a stated date competes with nothing, because nobody else has the number.
Then do the machine-readable layer — sitemap, structured data, a plain-text guide, a capability manifest. It is real work and it is what we sell, but it is downstream of being crawlable and quotable, and any advice that leads with it has the order wrong.
How to check where you stand
Scan your domain at colourintl.com. Sixteen weighted checks, free, no account, and the transcript of all eight requests is on the result page, so you can see exactly what a machine receives from your site rather than what your browser renders. The rendering check and the crawlability checks are the ones that matter for this page's question.
Related
llms.txt vs robots.txt — what do AI crawlers actually respect?
robots.txt is respected: it is a thirty-year-old convention that every major crawler including GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended reads and obeys, and it is the only file on this list that changes what a machine is permitted to do.
Agent readiness vs SEO, AEO and GEO — what is actually different?
SEO, AEO and GEO are all about winning a contest for attention: ranking a page, owning a featured answer, or getting mentioned by a model.
Check your own site
Scan any domain at colourintl.com — sixteen weighted checks, free, no account, and the transcript of all eight requests is printed on the result page. Everything it finds is yours to implement at no cost.
This page is also available as Markdown: curl -H "Accept: text/markdown" https://colourintl.com/learn/google-ai-mode