Optimising for AI Overviews

To be quoted in an AI Overview you must make the page indexable, put the answer in the first sentences and give every claim a source you can check. No tag, schema or file guarantees the citation: Google builds the summary at request time and picks the sources itself.

What an AI Overview is

An AI Overview is a model-generated summary that Google shows above the organic results when it judges that a synthesis answers better than a list. The block cites the pages behind each sentence, with inline links and a set of sources. Google announced the launch in the United States in May 2024 and extended it to other markets and languages; AI Mode is a separate product, with a results page generated end to end.

The four things that matter

First: indexable HTML. If the text appears only after JavaScript runs, the crawler may never see it. Second: the answer within the first sentences, in a paragraph that makes sense on its own. Third: facts in a table when the facts are numbers, dates and comparisons. Fourth: a visible author and a date, because a summary prefers a page that states who wrote it and when.

The local check

curl -sL https://ui.eneawork.it/blog/indicizzare-un-sito-su-google > page.html
node -e "const fs=require('node:fs');const h=fs.readFileSync('page.html','utf8');const t=h.replace(//g,' ').replace(//g,' ').replace(/]+>/g,' ').replace(/\s+/g,' ').trim();console.log('words:',t.split(' ').length);console.log('json-ld:',(h.match(/application\/ld\+json/g)||[]).length);console.log(t.slice(0,280));"

The command downloads the page the way a crawler sees it without running JavaScript, strips scripts and tags, and prints the word count and the JSON-LD blocks. If the words are few and the text does not carry the answer, the content lives in the client: an AI Overview has nothing to cite. The same check applies to every page you want quoted.

What you do not need

You do not need a dedicated markup: Google documents no schema.org type for entering AI Overviews. You do not need keyword density: the model reads the text, not the repetition. You do not need FAQPage: FAQ rich results have been restricted since 2023 and the markup does not decide the summary. Google does not read an llms.txt; the file remains useful to other tools, not to this one.

The EneUI catalogue

Pages that work as a source usually share a shape: a FAQ, a table, a documentation block. The catalogue offers live examples: the Kokonut UI FAQ, MIT licence, the FAQ category with the blocks that show visible questions and answers, and the full catalogue of 1,556 blocks. Checked on 1 November 2026.

The limitation

Inclusion cannot be measured. Search Console does not separate AI Overview impressions from web impressions, and the summary changes with language, query and user. The most you can verify is that the page is indexed and that the text sits in the initial document; the rest is a bet on the quality of the page.

Where to continue

What to write in an llms.txt file follows its own rules; the markup that really validates is in schema markup; the difference between server and client rendering is in Next.js or Vite for a marketing site. The sequence for getting indexed is in getting a site indexed by Google.