One meta tag can remove you from AI answers

Google's guidance on generative AI features contains one hard eligibility rule: the page must be indexed and eligible to appear with a snippet. If a page carries nosnippet, or a max-snippet value that suppresses text, it cannot appear in AI Overviews or AI Mode.

That makes snippet directives the only documented on-off switch you have. It also makes them the most likely reason a page you care about is invisible in AI answers for a reason nobody on the team remembers.

The four controls, and what each actually does

noindex removes the page from the index entirely. No ranking, no snippet, no AI feature. Obvious, and rarely the accidental one.

nosnippet keeps the page indexed but stops Google showing any text preview. No description, no featured snippet, and no eligibility for AI features. This is the dangerous one because the page still ranks, so it looks healthy in every report you run.

max-snippet:[number] limits preview length to that many characters. Setting it to 0 is functionally nosnippet. Small values can be enough to make a page useless as a source.

data-nosnippet is an HTML attribute applied to a section rather than a page. Wrap a div in it and that content is excluded from previews while the rest of the page stays eligible. Legitimate uses exist, such as excluding a pricing table you do not want quoted out of context, but it is regularly applied to whole content blocks by mistake.

That is the complete list. There is no AI-specific meta tag, no opt-in flag, no special directive for AI Overviews.

The one most people get wrong

Blocking Google-Extended in robots.txt does not remove you from AI Overviews or AI Mode.

Google-Extended governs whether your content is used for Gemini model development and certain grounding uses. AI Overviews and AI Mode are built on Google Search, so they run on the search index and are governed by search directives. A company that blocked Google-Extended believing it opted out of AI answers has done neither what it intended nor anything useful for visibility.

The reverse mistake also happens. A publisher blocks Google-Extended for legitimate licensing reasons, then panics when it still appears in AI Overviews, and adds nosnippet as a fix. That works, and it also removes the description from every ordinary search listing the site has. The traffic loss shows up two weeks later and takes a month to trace.

Decide those two questions separately, because they are separate systems.

Where legacy directives come from

In enterprise sites these tags are almost never added on purpose this year. They arrive from:

Old paywall projects. Publishers and subscription businesses added nosnippet across article templates years ago to stop content being read on the results page. The business model changed, the tag did not.

Staging templates that shipped. A blanket nosnippet added to a preproduction environment gets copied into a production template during a migration.

Legal or compliance requests. Someone in 2019 asked for a page not to be previewed for a defensible reason. The reason expired, the request never got reversed, and nobody owns it now.

Third-party components. Some CMS plugins, consent tools and personalisation scripts inject data-nosnippet around content regions to avoid caching issues.

Overzealous duplicate-content management. Teams sometimes use nosnippet on filtered or paginated pages to reduce clutter, not realising it also removes AI eligibility for pages that turn out to be the best answer to a specific sub-question.

The audit, which takes an afternoon

1. 1.Crawl the full site and export the robots meta tag and X-Robots-Tag header for every URL. The HTTP header matters as much as the meta tag and is easier to miss, because it does not appear in page source.
2. 2.Filter for nosnippet, noindex, none, and any max-snippet with a value under about 100. The value none is worth calling out because it means noindex plus nofollow, and it appears in old templates more often than you would expect.
3. 3.Search the codebase for data-nosnippet. Then check each instance against a rendered page to see how much content it actually wraps.
4. 4.Cross-reference the flagged URLs against your commercial page list. Most hits will be login pages, cart steps and internal search results, which is fine. The ones that matter are service pages, product pages and technical articles.
5. 5.Check pages that rank but get almost no clicks. A page holding position 4 with a click-through rate near zero often has no description showing, and nosnippet is a common cause.

What to do with legitimate cases

Some pages genuinely should not be previewed. Detailed pricing that varies by contract, content under a licensing restriction, sensitive material.

For those, use data-nosnippet on the specific block rather than nosnippet on the page. The page stays eligible, the paragraph you want protected stays out of previews, and you keep the rest of the page working as a source. This is precise surgery and hardly anyone uses it, because page-level tags are easier to implement.

Also worth saying plainly: if you decide you do not want your content used in AI answers, that is a legitimate business position and the directives above are how you enforce it. What is not legitimate is making that decision by accident and then paying an agency to explain why your AI visibility is poor.

The wider point

Every month brings another proposed AI optimisation technique, and almost none of them have a documented mechanism. Meanwhile the one control Google has explicitly stated exists sits in a meta tag most enterprise sites have never audited.

Start with the switch that is documented before funding the ones that are guessed.