Two robots directives written for search-result previews now decide whether Google's AI features may use your page at all. Here is what Google's documentation actually says, and how to check your own site.
Two robots directives that predate Google's AI features now decide something they were never chosen for: whether those features are allowed to use your page as input at all.
This is not an inference from observed behaviour. It is written in Google's own robots meta tag documentation, in the entry for nosnippet:
This applies to all forms of search results (at Google, web search, Google Images, Discover, AI Overviews, AI Mode) and will also prevent the content from being used as a direct input for AI Overviews and AI Mode.
And in the entry for max-snippet:
This applies to all forms of search results (such as Google web search, Google Images, Discover, Assistant, AI Overviews, AI Mode) and will also limit how much of the content may be used as a direct input for AI Overviews and AI Mode.
Read those two sentences together and the shape of the problem appears. The control that used to govern a grey line of preview text under a blue link now governs whether a generative answer may quote you. The directive did not change. Its blast radius did.
The eligibility rule, stated plainly
Google's documentation on its AI features states the entry condition in one sentence:
To be eligible to be shown 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.
Two conditions, joined by an "and". The first — indexation — is the one every standard reporting surface already tracks. The second, snippet eligibility, has no equivalent dashboard, and failing it used to cost only a preview line.
The second condition is also the easier of the two to fail without noticing. Indexation failures are loud: the page disappears from the index report. A snippet restriction is silent. The page stays indexed, keeps its ranking, keeps its clicks, and quietly stops being available as input to the feature that is increasingly answering the query before anyone clicks anything.
Where these directives come from when nobody set them
In practice the directives are rarely a decision. They arrive from somewhere else and stay.
A `max-snippet` cap set for a different purpose. The directive was designed to control how much preview text an aggregator or search result could show, and sitewide caps are frequently inherited from that era of decision-making. A cap chosen to limit a preview is now, per the documentation quoted above, a limit on how much of the page Google's AI features may use.
`max-snippet:0`, chosen because zero looks like a safe default. Google's documentation is explicit about what that value means: "Equivalent to nosnippet." Setting it to zero is not a conservative middle setting. It is the strongest setting, spelled differently. If you want Google to pick, the documented value is max-snippet:-1, which Google describes as letting it choose the snippet length it believes is most effective.
An `X-Robots-Tag` response header nobody reads. These directives can be delivered as an HTTP header as well as a meta tag, and the header is where CDN rules, edge middleware and security templates like to put things. A page can carry a perfectly clean <meta name="robots" content="index, follow"> in the head and still be sending X-Robots-Tag: nosnippet from the edge. Auditing the HTML alone will not find it.
`data-nosnippet` around the wrong element. Google documents data-nosnippet as a way to designate textual parts of an HTML page not to be used as a snippet, on div, span and section elements. It is a good tool used precisely — for a paywall teaser, a legal disclaimer, a boilerplate footer. It is a bad tool applied to a template wrapper, which is how a site ends up excluding the one direct-answer paragraph it wrote specifically to be quoted.
What this does not mean
Three corrections, because the opposite error is just as expensive.
It does not mean you should strip every snippet control. nosnippet exists for real reasons. Paywalled reporting, licensed content, member-only material and pages under a contractual restriction are all legitimate places to keep it. The point is not that the directive is wrong. The point is that it now buys and costs something different from what it bought and cost when it was set, and it deserves a fresh decision rather than inheritance.
It does not mean you need a new file or a new markup vocabulary. Google's AI features documentation is unusually direct on this:
You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add.
That sentence is worth keeping to hand, because advice sold as AI-search preparation frequently contradicts it. Structured data still earns rich results in ordinary Search and still helps a machine parse your page. It is not an entry ticket to the AI features, and Google says so in its own words.
It does not mean one page is competing for one slot. Google describes the retrieval behaviour as a fan-out:
Both AI Overviews and AI Mode may use a "query fan-out" technique — issuing multiple related searches across subtopics and data sources — to develop a response.
A single answer can therefore be assembled from several sub-queries you never targeted directly. The practical consequence is that eligibility is not a per-keyword property. A snippet restriction applied sitewide removes you from every sub-query at once, including the ones you would have won.
How the two directives differ in practice
It is tempting to treat nosnippet and max-snippet as a switch and a dial for the same thing, and for the blue-link preview that is roughly true. For the AI features Google describes them slightly differently, and the difference is worth holding on to.
nosnippet is described as preventing the content from being used as a direct input. It is categorical. There is no partial state and no negotiation: the text of that page is not available to the feature.
max-snippet is described as limiting how much of the content may be used as a direct input. It is a quantity. A generous cap leaves most of a page available; a tight cap leaves a fragment; zero, per Google's own note, is the categorical version again.
That distinction matters when you are deciding what to keep. A publisher who genuinely needs to restrict preview text has a graduated option and does not have to choose between full availability and none. A team that set a tight cap for a reason nobody now remembers is, in effect, choosing the fragment without knowing they chose it.
The asymmetry to keep in mind is that the cap is expressed in characters of snippet, not in importance of passage. Google decides which characters those are. You cannot use max-snippet to say "quote the definition but not the pricing" — that is what data-nosnippet is for, applied deliberately to the block you want withheld.
What to do about a restriction you decide to keep
Keeping a restriction is a legitimate outcome, and it does not have to mean giving up the query. If a page must stay restricted — licensed material, a paywalled report, something under contract — the answer is usually a second, unrestricted page that does a different job: a summary, a methodology note, a definitions page, a public index of what the restricted material covers. That page can be fully available as input while the restricted one stays closed.
This is ordinary editorial planning rather than a trick. The restricted asset keeps its commercial value; the open asset carries the entity, the terminology and the link. What you must not do is leave the restriction in place, publish nothing else on the subject, and then conclude from the absence of citations that AI search does not work for your category.
The check, in the order worth doing it
- Read the response header, not just the HTML. Request a representative
page the way a crawler would and look at X-Robots-Tag before you look at anything else. This is where the surprises live.
- Read the meta robots tag in the head. Look specifically for
nosnippet
and any max-snippet value. Treat max-snippet:0 as nosnippet, because Google does.
- Grep the templates for `data-nosnippet`. You are looking for it on a
wrapper rather than on a specific block. If it encloses body copy, the passage you most wanted quoted is the passage you excluded.
- Check the pages that matter, not the homepage. Directives are often
applied per template. The documentation page, the pricing page and the comparison page can each be governed by a different rule than the marketing pages someone audited last year.
- Write down the decision. For every restriction that stays, record why, in
the repository, next to the rule. The failure mode here is not a bad decision; it is a decision nobody remembers making.
Why this is the cheapest thing on the list
Much of the work aimed at AI search is speculative and slow: rewrite the page, wait, measure, argue about attribution. This one is neither. The rule is documented by the operator, the check takes an afternoon, and the failure it catches is binary — the content is available as input, or it is not.
It is also the kind of problem that survives a redesign. A snippet directive lives in a header rule or a template partial, not in the copy, so it outlasts every content refresh that steps around it. Sites carry these settings for years without anyone re-reading them, which is precisely why re-reading them is worth an afternoon.
Everything above is quoted from two pages of Google's own documentation, both linked in the claims on this post and both re-fetched on 4 September 2026. If Google changes the wording, the claim block is where to check first — and this post is on a quarterly review cadence for exactly that reason.