Google Search generative AI control: what Include and Exclude actually change

What changed in Google's guidance

On 21 October 2026, Google updated its Search Central documentation to surface the Search generative AI control alongside robots meta tags, data-nosnippet, X-Robots-Tag and other ways publishers manage what appears in Search. Google says the control had completed its worldwide rollout on 31 August 2026.

The setting is available in Search Console under Settings and Search generative AI. It applies to AI Overviews, AI Mode and generative AI features in Google Discover. Google expects the list of covered features to change over time, so site owners should review the current help page rather than relying on an old screenshot or checklist.

Include and Exclude are visibility choices

With Include selected, content remains eligible to appear as links and to help ground responses in the covered generative AI features. Google states that the site can receive impressions and traffic from those appearances. Eligibility is not a guarantee that any page will be shown or cited.

With Exclude selected, the site's content and links are prevented from being visible in those features and cannot be used as an input to generate an AI response or preview there. Google also states that the control is not used as a ranking or inclusion signal in other parts of Search.

Do not confuse Search visibility with AI training

Google explicitly separates the Search generative AI control from Google-Extended. The Search Console setting governs whether a site's content may appear in the covered Google Search features. Google-Extended is a robots.txt product token used to manage specified training and grounding uses for Gemini Apps and Vertex AI products.

That distinction matters because a publisher may want different policies for public search visibility and model-development use. Document the business reason for each choice, confirm which property or path it covers and avoid copying a robots rule from an unrelated AI crawler policy.

When Exclude may be reasonable

Exclusion can be a considered choice for publishers with contractual, licensing, regulatory or product reasons to keep content out of generative AI search experiences. It may also be relevant when a section has a different commercial model from the rest of the site. Search Console property inheritance allows parent and child properties to follow or override a configured parent setting.

Private information still needs real access control. Password protection, authentication and removal are stronger protections than a search visibility preference. Low-value or duplicate pages should be handled through information architecture, consolidation and appropriate indexing rules instead of excluding an otherwise useful site from AI features.

A measurement-first decision workflow

Before changing the setting, record the current value and export a baseline from the Generative AI performance report. Note the affected property, reporting window, cited or linked pages, impressions, visits and qualified enquiries where those are available. Keep the baseline separate from ordinary web results so a later comparison remains meaningful.

After a change, Google says exclusion generally takes a few days and most content is excluded within one or two days after the control goes live, although caching and propagation may take longer. Compare equivalent periods and document other site or campaign changes that could affect the result.

• Confirm the domain or URL-prefix property where the decision should apply.
• Check whether the property inherits a parent's setting or overrides it.
• Export a before-change baseline from the Generative AI performance report.
• Record the decision owner, business reason and review date.
• Recheck visibility, referral traffic and qualified outcomes after propagation.

RKD view: do not switch off discoverability blindly

For most service businesses trying to earn qualified discovery, the practical default is to keep useful public content eligible and improve its clarity, originality and evidence. Excluding an entire property without a licensing, privacy or business reason removes a discovery route before its value has been measured.

The better operating habit is to separate public marketing content from private material, publish direct and attributable answers, monitor the surfaces that send meaningful visits and review the control as the business model changes. This is a governance decision supported by search data, not a shortcut to better rankings or guaranteed AI citations.