I Checked 35 Publishers Fighting AI Overviews. 15 Blocked Google-Extended. Zero Used the Control Google Says Applies.

TL;DR: Google has never published a clean yes or no, so the honest answer is narrower than the one publishers act on: Google's documentation does not describe Google-Extended as a control over AI Overviews. It scopes the token to training and grounding in Gemini Apps and Vertex AI, and routes control over what Search displays to nosnippet, data-nosnippet and max-snippet. On August 15, 2026 I fetched robots.txt and one live article for 35 publishers who have publicly fought AI Overviews. 30 robots.txt files and 22 articles resolved. 15 of 30 disallow Google-Extended. Zero of 22 articles carry any of the three snippet controls Google does document for Search, and 19 of 22 carry max-image-preview:large, which expands previews rather than restricting them.

Penske Media sued Google in September 2025 over AI Overviews, a complaint filed in D.C. federal court alleging that Google's near-90% search dominance lets it avoid paying for content, and blaming the feature for a one-third drop in its affiliate revenue by late 2024. Penske owns Rolling Stone, Variety, Billboard, The Hollywood Reporter, Deadline, IndieWire and WWD. So I went and read what those seven sites, and 28 other publishers with public grievances about AI Overviews, actually deployed on their own servers.

What does Google say Google-Extended does?

Google's crawler documentation is specific about the scope. Google-Extended is described as "a standalone product token that web publishers can use to manage whether content Google crawls from their sites may be used for training future generations of Gemini models that power Gemini Apps and Vertex AI API for Gemini and for grounding (providing content from the Google Search index to the model at prompt time to improve factuality and relevancy) in Gemini Apps and Grounding with Google Search on Vertex AI." The named products are Gemini Apps and Vertex AI. AI Overviews is not among them.

The AI features page sends you somewhere else entirely: "AI is built into Search and integral to how Search functions, which is why robots.txt directives for Googlebot is the control for site owners to manage access to how their sites are crawled for Search. To limit the information shown from your pages in Search, use nosnippet, data-nosnippet, max-snippet, or noindex controls. To limit AI training and grounding in some of Google's other systems, read more about Google-Extended."

Read those two sentences in order. Snippet controls limit what Search shows. Google-Extended limits "some of Google's other systems." Google's own blog is blunter still, calling Google-Extended "a new control that lets websites manage how their content is used to train our Gemini models," and separately noting that the controls it added "for things like Featured Snippets and image previews" are the ones that "also apply to AI Overviews."

What did the 35 publishers deploy?

The method is repeatable in an afternoon. For each domain I fetched https://<domain>/robots.txt, parsed it into user-agent groups, and recorded whether any group naming Google-Extended carried a disallow. Then I fetched the homepage, extracted the highest-scoring article link, fetched that article unauthenticated, and searched the served HTML and response headers for nosnippet, max-snippet and data-nosnippet. Five robots.txt files refused me (metro.co.uk, inews.co.uk, newscientist.com, thetimes.com, mirror.co.uk), and 13 article fetches failed or returned a challenge page, so the real denominators are 30 and 22, not 35.

Google-Extended: 15 of 30 disallow it, every one of them with a blanket disallow: /. All seven Penske properties, plus dailymail.co.uk, nytimes.com, usatoday.com, nypost.com, cnn.com, nbcnews.com, npr.org and people.com. One site (telegraph.co.uk) names the token and allows it. Fourteen never mention it, including theguardian.com, wsj.com, washingtonpost.com, forbes.com and chegg.com.

Snippet controls: 0 of 22. Not one article carried nosnippet. Not one carried a numeric max-snippet limit. Not one carried a single data-nosnippet attribute. This is not a case of publishers having no opinion about their markup: 20 of the 22 serve a robots or googlebot meta tag, and 19 of 22 use it to carry max-image-preview:large, which asks Google for a larger preview. Newsweek's tag reads max-snippet:-1, which is the explicit instruction that there is no limit on snippet length.

The 2x2, across the 22 sites where I have both files:

	No snippet control	Snippet control
Disallows Google-Extended	12	0
Does not	10	0

An entire column is empty. The seven Penske sites add a flourish: each serves <meta name="robots" content="noai, noimageai">, a directive Google documents nowhere. The company suing Google over AI Overviews is running one control Google scopes to Gemini, one control Google has never recognized, and none of the three Google names for limiting what Search displays.

Why is "not documented" different from "proven ineffective"?

Because it is, and the distinction is the whole post. I did not run an experiment. I cannot tell you that a page with Google-Extended: disallow appears in AI Overviews at the same rate as one without, because measuring that requires query-level AI Overview data across a controlled sample, which I do not have. Google's wording is also deliberately soft: "some of Google's other systems" is not a list, and a token scoped to grounding in one surface could plausibly touch another.

What the record supports is the narrower claim: Google does not document Google-Extended as governing AI Overviews, and documents a different set of controls for what Search shows. A publisher who deployed Google-Extended believing it was an AI Overviews opt-out was reading a control Google described as being about Gemini training. That is a documentation-versus-belief gap, not a proven null effect, and anyone telling you they have measured the null effect should show you the sample.

What did Google ship in June 2026?

The strongest evidence is Google's behavior. On June 3, 2026, the UK's Competition and Markets Authority imposed a Publisher Conduct Requirement on Google Search, and the CMA's own description is that "in a world first, publishers will be able to opt out of their content being used to power AI features in Google search." Google responded the same day, writing that it was "now exploring updates to our controls to let sites specifically opt out of Search generative AI features."

You do not build a new opt-out for a thing your existing opt-out already covers. Google shipped a Search Console toggle to a subset of UK site owners rather than telling publishers to keep using Google-Extended, which is as close to a concession as this gets without a sentence saying so. The substantive obligations come into force on December 3, 2026, and the toggle is UK-first before any global rollout, so most of the 35 sites I checked could not have used it on the day I ran this even if they wanted to.

What should you check on your own site?

Decide which question you are answering, because they are three different switches. If you do not want your content training Gemini, Google-Extended is the documented control and it does not affect your ranking. If you do not want long extracts of your text displayed on a results page, nosnippet and max-snippet are the documented controls, and data-nosnippet fences off specific paragraphs while leaving the rest indexable, though the traffic loss they are meant to stop is smaller and narrower than the 58% figure everyone quotes. If you are in the UK, watch Search Console for the AI features toggle, which is the only control Google has described as doing what publishers wanted all along.

Then check what you actually shipped, because the gap between intent and markup is where this whole audit lives. The same gap shows up one layer down in which crawler tokens a blocklist really covers, and one layer over in why a site can rank on Google and stay invisible to ChatGPT. The paywall version of this problem is worse, because the defenses publishers built assume a crawler rather than an agent.

What would change my mind: a Google statement naming AI Overviews in the Google-Extended scope, or a controlled measurement showing AI Overview appearance rates differ with the token set. Neither exists today. Until one does, 15 publishers are running a control aimed at a different product, and none of them are running the ones aimed at this one. The reference case for what actually happens to traffic is Wikipedia, where human pageviews fell about 8% and the number only became visible after a bot-detection fix.