# How to get content cited in Google AI Overviews: 0 of 81 home pages blocked their own snippet > Google publishes one hard eligibility rule for a supporting link in AI Overviews, which is that the page must be indexed and eligible to be shown in Search with a snippet. Lantad requested the home page of 120 real hostnames on 6 September 2026. Of the 81 that answered HTTP 200, none carried a nosnippet directive, three carried max-snippet and all three set it to -1, and 48 carried no robots meta tag at all. - Canonical page: https://lantad.co/blog/how-to-get-content-cited-in-google-ai-overviews-zero-nosnippet - This file: https://lantad.co/blog/how-to-get-content-cited-in-google-ai-overviews-zero-nosnippet.md - Last substantive update: 2026-09-06 ## Key facts - **Published:** 2026-09-06 - **Category:** Findings - **Author:** Lantad - **Length:** 3442 words - **Takeaway 1:** Google's AI features documentation, carrying Last updated 2025-12-10 UTC, states that 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, and that there are no additional requirements to appear, nor other special optimizations necessary. - **Takeaway 2:** How to get content cited in Google AI Overviews begins with a control that almost nobody trips: across the 81 home pages that answered Lantad with HTTP 200 on 6 September 2026, zero carried a nosnippet directive in a robots or googlebot meta tag, and 48 of the 81 carried no such meta tag at all. - **Takeaway 3:** Three of those 81 pages declared max-snippet and every one of the three set it to -1, which Google's robots meta tag documentation, last updated 2026-03-24 UTC, defines as Google choosing the snippet length it believes is most effective rather than as any limit. - **Takeaway 4:** Two of the 81 pages carried a noindex directive, being europa.eu, which redirected to a language selector marked noindex follow, and sendbird.com, whose home page markup carried noindex, nofollow on 6 September 2026. - **Takeaway 5:** Of the 114 hostnames whose requests were not stopped by our own sandbox on 6 September 2026, 18 answered Lantad's crawler with HTTP 403 and five never completed a connection, which measures what one self-identified crawler received on one day and is not evidence about Googlebot, a client Google publishes a reverse DNS procedure for and which was not the client here. ## Summary Ask how to get content cited in Google AI Overviews and the answers you find are mostly lists of tactics. Google's own documentation is much shorter than any of them, and it names exactly one hard gate: the page has to be indexed, and it has to be eligible to be shown in Search with a snippet. Everything else on that page is a pointer back to ordinary Search guidance. The gate is worth taking seriously precisely because it is so narrow, since it is the one thing on the list that a site can fail silently and by accident, in four lines of markup that nobody looks at after launch. Our own [guide to being cited by Google AI Overviews](https://lantad.co/how-to-get-cited/google-ai-overviews) says the same thing in more detail and does not measure it. So we measured it. On 6 September 2026 we requested the home page of 120 real hostnames, one GET each, sent with this scanner's own user agent, and recorded every directive on the response that bears on snippet eligibility: the X-Robots-Tag header, any robots or googlebot meta tag, and any data-nosnippet attribute in the body. The result is a finding about where the difficulty in [generative engine optimization](https://lantad.co/glossary/geo) is not. On the pages that answered, the documented control was already correct essentially everywhere. What actually stopped us was earlier than that, and it was not a directive at all. ## How to get content cited in Google AI Overviews, in Google's own words [Google's AI features documentation](https://developers.google.com/search/docs/appearance/ai-features) carries the line Last updated 2025-12-10 UTC and makes two statements that are worth reading against each other. The first is the requirement: 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, fulfilling the Search technical requirements. The second is the disclaimer that follows it: there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary. Read literally, that is a page telling you there is no separate lever. Eligibility is a Search property, the same one that decides whether a blue link can carry two lines of description underneath it, and the AI feature inherits it. The same page names the controls that revoke it, in one sentence: use nosnippet, data-nosnippet, max-snippet, or noindex controls. Those four are defined in [Google's robots meta tag documentation](https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag), last updated 2026-03-24 UTC, and the definitions matter because two of them are easy to misread. nosnippet means do not show a text snippet or video preview in the search results for this page. max-snippet takes a character count, and the documentation gives two special values: 0 is equivalent to nosnippet, and -1 means Google will choose the snippet length that it believes is most effective. So a page declaring max-snippet:-1 has not limited anything. It has written down the default. data-nosnippet is a boolean attribute on span, div and section elements, marking textual parts of a page not to be used as a snippet, which makes it the only one of the four that operates on part of a page rather than all of it. And any rule that can be written in a meta tag can also be sent as an X-Robots-Tag response header, which is where a directive can arrive without ever appearing in the HTML. None of this is new advice, and that is the point of measuring it rather than repeating it. The AI Overviews layer keeps producing questions whose answers turn out to live in ordinary Search mechanics, which is also why [the AI Overviews opt out does not live on your site](https://lantad.co/blog/ai-overviews-opt-out-does-not-live-on-your-site) and why [the crawler behind the feature is one this scanner does not model](https://lantad.co/blog/ai-overviews-and-the-crawler-we-do-not-model). A site owner looking for a dedicated switch keeps finding a Search switch with a different label on it. ## What 81 real home pages declared about their own snippets The sample is 120 hostnames, 15 drawn from each of the eight categories in the industry seed frame committed to this repository, taken at evenly spaced positions in each sorted category list so the set is not all A names. The categories are government, education, healthcare, news, SaaS, ecommerce, travel and finance. One GET per hostname, redirects followed, a 20 second timeout, this scanner's user agent, and no JavaScript executed, so what was read is what an ordinary HTTP client receives before anything runs. Of those 120, six requests never left our sandbox, and we will come back to that. That left 114 we actually put to a site, five of which never completed a connection. 109 responses came back, and 81 of them were HTTP 200 with an HTML document. Those 81 are the population every directive figure here is drawn from. Zero of the 81 carried a nosnippet directive. Three carried max-snippet, being rappler.com, browserstack.com and tiendanube.com, and all three set it to -1, which is the value that imposes no limit. Three carried at least one data-nosnippet attribute, being nationalgeographic.com with 32 of them, stockholm.se with one and datadoghq.com with one, and a partial exclusion is not an eligibility failure. Two carried a noindex directive. So on their own markup, 79 of the 81 pages that answered met the precondition Google publishes, and not one of the 81 had told Google to withhold a snippet from the whole page. The absence is more striking than the presence. 48 of the 81 pages carried no robots meta tag and no googlebot meta tag at all, which under Google's documented defaults is the same instruction as index, follow. Of the 33 that did carry one, the commonest content was ceremonial: 20 declared index, 20 declared follow, and 13 declared a max-image-preview value. An X-Robots-Tag header appeared on three of the 109 responses the sites returned, being gov.br with noimageindex, intercom.com with index, follow, and berlin.de with noindex on an error page. Only two of those three landed on a page that returned HTTP 200. That pattern is consistent with the one we found when we counted [85 meta elements across five page heads and found two addressing a crawler](https://lantad.co/blog/eighty-five-meta-elements-two-addressed-a-crawler), where both robots meta tags carried nothing but an image preview size. The head of a modern page is mostly social card markup. The directives that decide search behaviour are usually absent, and absence is permissive. It is the opposite shape from an opt out that has to be typed, which is why [Microsoft's AI opt out being a meta tag](https://lantad.co/blog/microsoft-ai-opt-out-is-a-meta-tag) reaches so few sites. It is also why the interesting questions in this area sit in [structured data](https://lantad.co/glossary/structured-data) and in retrieval rather than in permissions, and why [structured data came third in a 1,100 URL citation audit](https://lantad.co/blog/structured-data-came-third-in-a-citation-audit) rather than first. ## The two pages that carried noindex, and the two that only looked like it Two of the 81 declared noindex, and they failed the gate in different ways. sendbird.com is the simple case. Its home page returned HTTP 200 with 103,746 bytes of HTML and a server rendered meta tag reading noindex, nofollow. Whether that is deliberate is not ours to say, and the tag is what a crawler reads. europa.eu is the more instructive one, because the directive is not on the page anyone means by europa.eu. A request to the bare host followed a redirect to european-union.europa.eu and landed on a language selector at a path ending in select-language, 18,133 bytes, carrying noindex follow. The follow matters: the page tells a crawler not to index this document and to keep walking its links, which is the correct configuration for an interstitial. The document a crawler first meets at that address is still one that cannot itself be cited. Which document a fetch actually lands on is the same problem we found when we [fetched robots.txt across fourteen hostnames belonging to six sites](https://lantad.co/blog/check-for-robots-txt-fourteen-hostnames-eight-files) and got eight different files back, and it is why [what a crawler receives at a given address](https://lantad.co/tools/what-gptbot-sees) is worth checking at the address rather than assuming. Two further pages contained the string noindex where it governed nothing, and separating them from the real cases is why this was checked twice rather than counted with a search. williams-sonoma.com carries the string inside a JSON payload describing the metadata of its 404 page, while its own robots meta reads index, follow. abs.gov.au carries it four times, in a pair of HTML comments wrapping a sidebar and in rel="noindex" attributes on tab links, neither of which is a directive Google's documentation defines. Both would have been counted as blocked by a naive grep, which would have doubled the headline. A page can also be long enough that a directive sits beyond where a crawler reads, which is a real constraint given that [Googlebot reads the first two megabytes](https://lantad.co/blog/googlebot-reads-the-first-two-megabytes), though nothing in this sample came close to that. ## 18 of 114 refused the request before any directive mattered The directives were fine. The requests often were not. Of the 109 responses that came from the sites themselves, 81 were HTTP 200 and 28 were not. Eighteen were HTTP 403. Six were HTTP 503, one was 429, one 405, one 400, and booking.com returned HTTP 202, which is an acknowledgement rather than a page. A further five hostnames never completed a connection at all: boj.or.jp, kenya.go.ke, usgs.gov and mhlw.go.jp failed at the transport layer, and qantas.com was still open at the 20 second timeout. The split by category is sharp. All 15 SaaS hosts returned HTTP 200 and none of them refused. Healthcare returned 13 of 15. Travel and ecommerce returned seven each, with four HTTP 403 responses apiece, and finance returned ten with four refusals. The full refusing set is argos.co.uk, coupang.com, petco.com, screwfix.com, mastercard.com, hdfcbank.com, aviva.co.uk, qnb.com, lufthansa.com, sbb.ch, aa.com, tripadvisor.com, mdpi.com, ias.edu, heart.org, berlin.de, india.gov.in and parliament.uk. On this sample the categories selling to consumers refuse an unrecognised client and the one selling to developers does not, which is 15 hosts against 15 and a direction rather than a rate, and it points the same way as the finding that [who blocks AI crawlers splits by credibility](https://lantad.co/blog/who-blocks-ai-crawlers-splits-by-credibility) rather than by size. Two things about those refusals are worth stating precisely. The first is that a refusal is not a robots.txt decision, and it does not look like one from either side: [a site that bans GPTBot in robots.txt can still serve it an HTTP 200](https://lantad.co/blog/gptbot-bans-served-a-200-anyway), and a site that allows everything can refuse at the edge. The status code carries meaning of its own, since [a 404 and a 503 on robots.txt are opposites](https://lantad.co/blog/robots-txt-404-and-503-are-opposites) in how a well behaved crawler treats them, and six of these responses were 503. The second is that these decisions were stable rather than random, which we checked because it is the obvious objection. We ran the same 120 requests three times on 6 September 2026. Across the 114 site-answered hostnames, only three returned a different status on different passes: qnb.com and parliament.uk answered 403, then 200, then 403, and seoul.go.kr answered 200 twice and then 503. The other 111 gave the same answer every time. The refusal is a policy, not a hiccup, and it is the same wall that stops a browser automation client, which is how [headless Chromium was blocked on 15 percent of sites](https://lantad.co/blog/headless-chromium-blocked-on-15-percent-of-sites) in an earlier measurement. We publish [the user agent and address this scanner sends from](https://lantad.co/bot) so that a site owner who wants to see this in their own logs can find it. ## What this measurement does not establish about Googlebot The largest limit is the client. Every request here identified itself as LantadBot, and Googlebot is a different client with a different reputation. Google publishes [a verification procedure for its own crawlers](https://developers.google.com/search/docs/crawling-indexing/verifying-googlebot), last updated 2026-03-20 UTC, which is a reverse DNS lookup on the accessing address, a check that the domain is googlebot.com, google.com or googleusercontent.com, and a forward lookup back to the same address, with published IP range files as the automatic alternative. A site can therefore allow Googlebot specifically while refusing everything it does not recognise, and 18 sites refusing us establishes nothing whatever about whether they refuse Google. We did not ask as Googlebot, we could not have, and impersonating one would have produced a number that was worse than useless. This is the same distinction as [a user agent being a claim rather than an identity](https://lantad.co/blog/a-user-agent-is-a-claim-not-an-identity), pointed the other way. The second limit is the six requests that never left our own sandbox. The environment this scan ran in answered HTTP 403 with a body reading Blocked by egress policy for taobao.com, arstechnica.com, dw.com, nrc.nl, sueddeutsche.de and usatoday.com. Those are our proxy's refusals, not theirs. They are excluded from every figure above, and the reason they are mentioned at all is that a first pass counted them as site refusals and inflated the 403 total by a third. A measurement that cannot tell its own infrastructure from its subject is not a measurement, and [how this scanner reaches a verdict](https://lantad.co/methodology) is written down so that this sort of correction is checkable rather than trusted. The third is scope. This is 120 home pages on one day. A home page is not a representative page of a site, an eligibility control on a home page says nothing about the article pages that actually get cited, and one request is a moment rather than a policy. Nothing here observed a crawl, an index decision, or a citation. Being eligible for a supporting link is a precondition and it is not a prediction: our [standing measurements](https://lantad.co/research) keep finding that the distance between allowed and cited is where the entire difficulty lives. What the measurement does support is a narrower claim, and it is the useful one. If you are working out how to be cited by an answer engine and you start by auditing your snippet directives, you are auditing something that was already correct on 79 of 81 real sites. The two failures were an interstitial and a page whose owners may well have meant it. Meanwhile 18 sites returned an error to a crawler that asked politely, and that is not a markup problem, not a robots.txt problem, and not visible in any of the four controls Google names. ## Questions and answers **What does Google require for a page to be cited in AI Overviews?** One thing, according to its AI features documentation carrying Last updated 2025-12-10 UTC: the page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. The same page states there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary. **Does max-snippet:-1 limit what an AI Overview can quote?** No. Google's robots meta tag documentation, last updated 2026-03-24 UTC, defines -1 as Google choosing the snippet length that it believes is most effective, so the directive imposes no limit at all. A value of 0 is the one that matters, because the documentation says it is equivalent to nosnippet. All three pages that declared max-snippet in this measurement set it to -1. **How common is it for a real site to block its own snippet?** Rare, on this sample. Across the 81 home pages that returned HTTP 200 to Lantad on 6 September 2026, none carried a nosnippet directive and two carried noindex. 48 of the 81 carried no robots or googlebot meta tag at all, which under Google's documented defaults permits both indexing and a snippet. **Does an HTTP 403 to a scanner mean Google is blocked too?** No, and nothing here tests that. The 18 refusals were answers to a client identifying itself as LantadBot. Google publishes a reverse DNS and IP range procedure for verifying Googlebot, last updated 2026-03-20 UTC, so a site can recognise and allow Googlebot while refusing clients it does not recognise. Separating the two needs a site's own access log. --- Lantad measures whether AI crawlers can actually read a page: it fetches as a non-rendering crawler, renders as a browser, and reports the gap. Free scan, one URL, no signup. Method and weights: https://lantad.co/methodology | All pages as markdown: https://lantad.co/md | Crawler policy: https://lantad.co/bot