Google's AI Overviews opt-out, explained — what the Search generative AI control does and whether your site should use it

Key takeaways

There is now a one-click opt-out from Google's AI answers. The Search generative AI control sits in Search Console under Settings → Search generative AI. Set it to Exclude and your pages stop appearing in AI Overviews, AI Mode and Discover's generative features, as links and as grounding material (Search Console Help).
It has been live for every site since 31 August 2026. On 21 October Google also added it to its main Control what you share with Google Search page, next to noindex, robots.txt and nosnippet (Search Engine Roundtable). It is now a standard option, not a beta.
Google says opting out does not hurt your classic rankings. It is "not a ranking or inclusion signal" for the rest of Search. It also does not stop AI training; that needs a separate Google-Extended rule in robots.txt.
Exclusion usually applies within 1–2 days. Caching can make some pages take longer.
Our view for most Indian businesses: keep it on Include, check the Generative AI performance report, and fix how your pages get cited. Opting out mainly suits publishers whose business is the answer itself.

1. What changed this week, and why it matters

Since 2024, site owners who wanted out of AI Overviews had only blunt tools. nosnippet also strips your normal search snippet, and noindex drops the page from Google completely. In June 2026 Google announced generative AI performance reports along with a dedicated control. The control reached every site worldwide on 31 August.

The 21 October documentation update is small but it settles the status. Google's own list of ways to keep content out of Search now has the AI control beside the established methods (PPC Land). That makes it a supported, long-term setting. Agencies and in-house SEO teams should expect clients to ask about it.

2. The three settings, in plain words

Setting	What happens	Default for
Include	Pages can appear in AI Overviews, AI Mode and Discover AI features as links and as grounding. You get impressions and traffic from them.	Top-level domain properties
Exclude	Pages are hidden from those features, both as links and as grounding. You get zero impressions or traffic from them.	—
Inherit from parent	The property follows the nearest parent property that has its own setting.	Child properties (subdomains, URL-prefix paths)

Google's help page makes two points that are easy to miss:

Excluding you does not remove the answer. Other sites' content is still used, and Google notes the result "may look similar" to what your page would have produced. You lose the citation, and the user still gets an AI answer from somewhere else.
Your content can still help Search in general, for example in understanding query and page language. Exclude is not a full privacy wall.

Section-level control through child properties

Reports at the global rollout described the setting as site-wide only (Relevant Audience). The help page is more precise. URL-prefix properties and subdomains can act as child properties with their own setting. So example.in/blog/ can be Exclude while example.in/services/ stays Include, provided both are verified as separate properties. If a child property overrides the parent, the owner of the top-level domain property gets a notification. Put that on your agency handover checklist.

3. How it compares with the controls you already have

Control	Where	Scope	Removes from AI Overviews / AI Mode	Removes normal snippet	Removes from Google index	Blocks AI training
Search generative AI control	Search Console	Property (child properties can override)	Yes	No	No	No
`nosnippet` / `max-snippet`	robots meta / X-Robots-Tag	Page	Yes (per Google, snippet limits also limit AI use)	Yes	No	No
`data-nosnippet`	HTML attribute	Part of a page	That text only	That text only	No	No
`noindex`	robots meta / X-Robots-Tag	Page	Yes	Yes	Yes	No
`Disallow` in robots.txt	robots.txt	Path	Not reliably: a blocked URL can still be indexed from links	—	Not reliably	No
`User-agent: Google-Extended` + `Disallow`	robots.txt	Path	No	No	No	Yes, for the Gemini models that power these features (Google crawlers list)

The takeaway: the Search Console setting and Google-Extended do different jobs. To stay out of AI answers and keep your content out of model training, you need both. Neither touches other AI companies' crawlers, which have their own robots.txt user-agents.

4. Measure before you decide

Google built the Generative AI performance report to go with the control, and it is the only first-party data you have. Per reporting on the launch, it shows:

Impressions only. It has no clicks, CTR, average position or triggering queries.
Dimensions: pages, countries, devices (Search only, not Discover) and date, with hourly to monthly granularity.
De-duplication: a URL that appears in an AI Overview and in the blue links on the same results page counts as one impression.

You cannot see clicks from AI features directly. A workable proxy: export the AI report's top pages, then compare those pages' clicks in the normal Performance report and their organic sessions in GA4 for the 28 days before and after any change. Do this per section, not site-wide.

5. Who should opt out, and who shouldn't

Usually keep Include:

Service businesses, agencies, SaaS, clinics, coaching institutes and D2C brands. A citation in an AI Overview for "best CA firm in Noida" or "GST registration documents" is free brand exposure. Your money pages convert on intent, not on the answer text. Excluding hands the citation to a competitor.
B2B companies chasing enterprise leads. If buyers start their research in AI Mode, not being named at all costs you more than a lost click.

Consider Exclude, ideally for one section first:

News and exam-prep publishers, recipe and lyrics-style sites, and calculators whose page is the answer. If an AI summary removes the reason to click and your revenue is ads or subscriptions, the AI report will show high impressions on pages whose clicks are falling.
Paid research or premium content that leaks into AI answers through free teaser pages.

A rupee sanity check: suppose a section brings 40,000 organic visits a month and earns ₹60 RPM from display ads. That is about ₹2,400 a month. If you suspect AI Overviews take away a quarter of those clicks, you are arguing over about ₹600 a month. That is not worth losing brand citations. If the same traffic feeds a ₹999 course funnel converting at 0.5%, the 200 sales a month (about ₹2 lakh) make the decision worth testing properly. Run the numbers with your own RPM and conversion rate. These figures only show the method.

6. A 30-minute checklist for this week

Verify a Domain property for the whole site if you only have URL-prefix ones. The setting inherits from the top-level domain.
Open Settings → Search generative AI and confirm it says Include (or whatever you chose deliberately). Write it down. Developers, plugins or agencies should not change it quietly.
Export the Generative AI performance report (top pages, last 3 months) and keep it as your baseline.
Audit stray nosnippet tags. Some SEO plugins and themes add them. They take you out of AI answers and kill your normal snippets. Search your templates for nosnippet and max-snippet:0.
Decide on training separately. If you don't want content used to train Gemini, add a Google-Extended block in robots.txt. That rule is independent of the Search setting.
If you exclude, test one section first using a URL-prefix child property. Compare 28 days before and after.
Note the hyphen guidance. Google's 23 October docs update also clarified that URL slugs should use hyphens, not underscores, to separate words (Search Central updates). Fix that on new pages; renaming old URLs is rarely worth the redirects.

Want a second pair of eyes on your Search Console settings, snippet tags and AI visibility? That is part of our digital marketing and SEO service. For robots.txt rules such as Google-Extended, start with the free .htaccess & robots.txt generator.