MartechARTICLE

Google explains what makes a page eligible for AI Overviews

Google Search Central updated, in December 2025, the official guide on AI in search. The real news isn't technical: it's confirmation that there's no special markup to get into AI Overviews, only controls to opt out.

Google explains what makes a page eligible for AI Overviews
Image: Sabrina Santos
Article · Martech
Reading
7 min
Code
5

Google explains what makes a page eligible for AI Overviews

Google Search Central updated, in December 2025, the official guide on AI in search. The real news isn't technical: it's confirmation that there's no special markup to get into AI Overviews, only controls to opt out.

Editor's note: the exact update date of the documentation cited below (December 10 or 31, 2025) could not be confirmed with certainty: the English version of the source consulted shows 'Last updated 2025-12-10 UTC,' but the Portuguese version of the same page shows 'Última atualização 2025-12-31 UTC.' We kept the reference to December 10 since that's the date shown in the English version consulted, but readers should factor in this discrepancy between the Google document's own versions when evaluating the claim.

Google Search Central updated, in December 2025, the official guide on AI in search. The real news isn't technical: it's confirmation that there's no special markup to get into AI Overviews, only controls to opt out.

Google Search Central updated, in December 2025, the official guide on AI in search. The real news isn't technical: it's confirmation that there's no special markup to get into AI Overviews, only controls to opt out.

Google Search Central updated, on December 10, 2025, the official "AI features and your website" page, which documents how AI Overviews and AI Mode work from the perspective of whoever publishes content. The date matches the official record in the documentation's footer ("Last updated 2025-12-10 UTC"); an earlier iMasters publication cited December 31 as the update date, but according to the primary source consulted for this piece, the correct record is December 10, 2025.

Almost a year later, it remains Google's most direct reference on the subject, and it's worth revisiting because the text answers a question every Brazilian technical SEO team has already asked: is there any special markup to get into the AI-generated summary?

The short answer, according to the documentation itself, is no. And that changes how it's worth investing engineering effort on this front.

The guide's central point is that there's no additional technical requirement for a page to be cited as a supporting link in an AI Overview or an AI Mode response.

To appear as a supporting link in AI Overviews or AI Mode, a page needs to be indexed and able to be shown in Google Search with a descriptive snippet, meeting Search's technical requirements. There is no additional technical requirement.

Google Search Central, official documentation

In practice, this means investing in a "special for AI" schema is wasted time: what exists today are already well-known SEO best practices, plus one detail that usually slips by unnoticed: parity between structured data and the visible text on the page.

The detail most people ignore: structured data parity

The guide lists recommended practices, and one of them is simple to check and easy to violate without noticing: structured data needs to match the visible text on the page. If the JSON-LD says one thing and the rendered HTML says another, the source becomes inconsistent both for ranking and for citation in generated summaries.

A common example of misalignment, when the marketing team updates the price on the page but forgets the schema:

json
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Plano Pro",
  "offers": {
    "@type": "Offer",
    "price": "79.90",
    "priceCurrency": "BRL"
  }
}

If the page's visible text shows "R$ 99.90/month," Google is reading two different prices for the same offer. The fix is trivial, but it requires process: every deploy that changes price, deadline, or a visible feature needs to update the JSON-LD in the same commit.

json
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Plano Pro",
  "offers": {
    "@type": "Offer",
    "price": "99.90",
    "priceCurrency": "BRL"
  }
}

To validate this automatically, you can write a regression test that compares the schema value with text extracted from the DOM before each deploy, instead of relying on manual review.

Controlling what AI can quote: data-nosnippet

Editor's note: the code in this section was not run in a real environment. Validate before using in production.

The guide mentions, together, four commands to limit what appears from a page in search: nosnippet, data-nosnippet, max-snippet, and noindex. The source doesn't detail the specific scope and effect of each one, only mentioning them as available options; what follows is a practical reading of how these controls are typically used, not a transcription of the documentation.

Generally, noindex, nosnippet, and max-snippet operate at the whole-page level, while data-nosnippet lets you mark a specific HTML element. It's this difference in granularity that makes data-nosnippet the more surgical option: instead of sacrificing the whole page, you mark only the sensitive piece, usually a price, deadline, or frequently changing data point the site doesn't want reproduced out of context in a generated summary.

Before, with no control at all, any sentence on the page becomes a candidate for citation:

html
<article>
  <h1>Comparativo de planos</h1>
  <p>O plano Enterprise custa R$ 4.500/mês e inclui SLA de 99,99%.</p>
  <p>Este preço é válido até 31/12/2026 e pode mudar sem aviso.</p>
</article>

After, with the attribute applied only to the paragraph containing the sensitive data:

html
<article>
  <h1>Comparativo de planos</h1>
  <p data-nosnippet>O plano Enterprise custa R$ 4.500/mês e inclui SLA de 99,99%.</p>
  <p>Este preço é válido até 31/12/2026 e pode mudar sem aviso.</p>
</article>

The rest of the page remains eligible for snippet and for citation in AI Overviews; only that paragraph is left out. It's the right tool for anyone who wants to keep being found but doesn't want a number that changes every week to be reproduced outdated in an AI summary.

Google-Extended solves a different problem, not this one

It's worth separating two controls that the guide treats as distinct things. nosnippet and data-nosnippet decide what appears in Google's search results, including AI Overviews and AI Mode. Training of Google's AI models outside the context of search, however, is controlled by the Google-Extended user-agent, in robots.txt:

User-agent: Google-Extended
Disallow: /

The guide doesn't detail whether there's a cross-effect between the two controls; in the author's reading, since they're presented as distinct mechanisms for distinct purposes, it's reasonable to assume they act independently: one over what appears in search, the other over use for training. They're different keys for different doors, according to this interpretation.

Measuring the effect: what can and can't be isolated in Search Console

The guide confirms that traffic coming from these features is lumped together in Search Console's performance report, under the "Web" search type, with no separate search type for AI Overviews or AI Mode. The document doesn't detail whether any other segmentation feature is available within that report.

The document offers a qualitative observation relevant to anyone cross-referencing data with Analytics: clicks originating from results pages with AI Overviews tend to be higher quality, with users spending more time on the site after clicking. This suggests a possible measurement proxy, even without direct segmentation.

In short: to approximate the effect of AI Overviews in practice, the way to go is to cross-reference Search Console queries with conversational or multi-part patterns (typical of branched search distribution) against time-on-site in Analytics, tracking the variation before and after adjustments such as schema parity or use of data-nosnippet.

When blocking is worth it, and when it costs you

Blocking excerpts or entire pages has a cost: the same control that prevents improper citation in an AI summary also reduces the chance of appearing with a snippet in traditional search. For content that lives on clicks, like affiliate-monetized comparisons or articles with display ads, over-extracting with data-nosnippet can protect the sensitive data but impoverish the snippet that convinces someone to click.

The decision, then, isn't binary between "leave everything open" or "block everything": it's choosing, element by element, what's worth more as click bait and what's worth more as information protected from automatic reproduction. Google's guide doesn't resolve this choice, it only provides the tools; who decides where to apply each one is the product and content team, with data from Search Console itself tracking the result.

Translated from the Brazilian Portuguese original · Read the original

Read also
Martech

Google Ads updates Conversion Lift setup guide with stricter tracking requirements

The Conversion Lift documentation, updated on November 13, 2026, did not change the incrementality test's methodology, but it tightened the technical tracking requirements that feed into the result. The update itself now explicitly lists unsupported configurations, and Google's own recommendation is to audit tracking setup before running a study, especially for offline conversions, GA4 imports, and app conversions.

Sabrina Santos··1 min