Definition
The Search generative AI control is a Google Search Console setting that lets a site owner include or exclude a property's links and content from Google's generative AI search features: AI Overviews, AI Mode and generative AI features in Google Discover. Excluding a site removes it both as a cited link and as grounding material for AI responses, while ordinary search results stay as they were.
Before 2026 there was no clean way to do this. Snippet directives such as nosnippet limit AI Overviews but also strip regular snippets, and noindex removes a page from Search entirely. Google-Extended controls training of Gemini models and grounding in the Gemini app and Vertex AI, not inclusion in AI Overviews or AI Mode.
The control came out of UK regulation. The Competition and Markets Authority designated Google as having strategic market status in general search in October 2025, and on June 3, 2026 it imposed a publisher conduct requirement that it called a world first: publishers must be able to stop their content being used to power AI features in Google Search, and to opt out of its use for fine-tuning AI models, without being penalized in regular results (CMA announcement). Google announced the toggle the same day, tested it with a subset of UK properties, began honoring choices on June 17, 2026, extended it outside the UK in July, and says it reached all websites worldwide on August 31, 2026.
The setting lives under Settings > Search generative AI and has three values (Google's help page):
- Include: the default for all properties, so content can appear as links and help ground AI answers.
- Exclude: links and content are kept out of these features, and the site gets no impressions or traffic from them.
- Inherit from parent: a property follows its closest parent property that has a configured value, falling back to the top-level domain property.
A child property, such as a URL-prefix property for one directory, can override its parent, and the domain owner sees a notification when that happens. Google says exclusion generally takes effect within a few days, with some content taking longer because of caching.
The limits are worth reading closely. The control is not used as a ranking signal elsewhere in Search. It does not override participation in Merchant Center or Google Ads, and it does not affect AI training, which still runs through Google-Extended. Google may still use the content to power Search in general, for example to understand language. And it is property level only: under the CMA timeline Google has until about March 2027 to provide more granular page-level controls.
For GEO, the control turns AI Overviews visibility into an explicit decision. Most brands want to be cited, since excluded sites simply hand the answer to competitors whose content says something similar. Publishers who sell content, or who see AI answers replacing visits, may decide otherwise. Base the decision on data: export AI impressions from the generative AI performance report, compare them with clicks for the same pages, and consider whether a licensing deal or other TDM rights reservation better fits the goal. News publishers should also weigh a NewzDash finding, reported by Search Engine Journal in August 2026, that Top Stories carousels sometimes render inside AI Overviews. Google's documentation does not say how the control treats them.
Examples of Search Generative AI Control
- A UK news publisher switches its domain property to Exclude in June 2026. Within a few days its AI impressions drop to zero, while rankings for its articles in the regular blue links hold steady.
- A university leaves its main domain included, but the owner of the https://example.edu/research/ URL-prefix property excludes that directory, and the domain owner receives a notification about the override.
- A retailer excludes its blog from AI features after licensing that content elsewhere. Its product listings keep appearing in Shopping placements because Merchant Center participation is a separate choice.
- A publisher that only wants to stop model training leaves the control on Include and disallows Google-Extended in robots.txt instead, keeping its AI Overview citations.
