Google's Search generative AI control manages whether a site can appear in AI Overviews, AI Mode and generative AI features in Discover. Google-Extended controls separate Gemini and Vertex AI uses; it does not control Search inclusion or ranking. Search Console's Generative AI performance report measures impressions in AI Overviews and AI Mode, not clicks or exact queries.
Updated 5 November 2026: Google has added a Search Console control for its generative Search features. It is separate from Google-Extended, the robots.txt token often confused with it.
IDJoy sells an AI search visibility service. This guide distinguishes Google’s documented controls from independent research; no visibility outcome is guaranteed.
The short answer
If you want to manage whether your site’s content can appear in Google Search’s generative AI features, use Search Console’s Search generative AI control at Settings → Search generative AI. Google says it rolled out worldwide on 31 August 2026. Inclusion is the default; child properties can inherit a parent property’s choice.
The control covers AI Overviews, AI Mode and generative AI features in Discover. Excluding a property means it will not appear in those features or receive traffic from them; it does not remove the property from ordinary Search results, change ranking, or control Google’s model training. Google says an exclusion usually takes a few days to apply, with most content excluded within one to two days after the change goes live; caching and system updates can take longer.
Google-Extended is a separate robots.txt token. It expresses a preference about certain Gemini and Vertex AI training and grounding uses. Google says it does not affect inclusion or ranking in Google Search. Neither setting guarantees that a particular page will appear in an answer.
Which Google setting or report matches your goal?
| Your goal | Control or report | What it does | What it does not do |
|---|---|---|---|
| Manage eligibility for AI Overviews, AI Mode and generative AI features in Discover | Search Console → Settings → Search generative AI | Includes or excludes the property from those generative Search features | Does not remove ordinary Search results, change ranking or control model training |
| Manage Search crawling, indexing or result previews | Googlebot access and Search directives such as noindex, nosnippet, data-nosnippet and max-snippet |
Controls crawling, whether a page can be indexed, or what preview Google may show | Does not provide the same property-level AI feature control as Search Console |
| Set a preference for documented Gemini and Vertex AI uses | Google-Extended in robots.txt |
Signals whether content may be used to train specified models or support answers in Gemini Apps and Vertex AI | Does not control Google Search inclusion or ranking |
| Measure Google Search generative AI impressions | Search Console’s Generative AI performance report | Reports impressions from AI Overviews and AI Mode when enough data is available | Does not report exact queries, separate clicks, CTR or average position |
Google’s preview directives have different effects. noindex prevents a page from appearing in Search; nosnippet prevents a text snippet; data-nosnippet excludes selected parts of a page from snippets; and max-snippet limits snippet length. A page being technically eligible for a Search preview does not guarantee that Google will index or show it.
What Google-Extended covers
Google-Extended is not a separate crawler or HTTP user agent. Googlebot fetches pages; the token tells Google’s systems whether they may reuse site content for specific model training and answer-generation uses. In this context, “grounding” means using information from a page to support a generated answer. Google documents training uses for future Gemini Apps and Vertex AI API models, plus answer support in Gemini Apps and Google Search grounding on Vertex AI.
It is not the switch for AI Overviews or AI Mode. Google documents those as Search features and says Google-Extended does not affect Search inclusion or ranking. If you want to manage Search generative feature eligibility, use the Search Console control. If you want to set a preference for the documented training or grounding uses, check the current rules in your site’s root robots.txt before changing them. To disallow all paths for this token, use:
User-agent: Google-Extended
Disallow: /
Add or change these lines only if they match your intended preference; keep existing directives for other user agents intact. Blocking this token does not remove the site from Search AI features.
Google also documents user-triggered fetchers such as Google-GeminiNotebook and Google-Agent. These fetchers may retrieve pages after a user supplies a URL; they are not Search inclusion controls. Google says user-triggered fetchers generally ignore robots.txt because a person initiated the request. The former Google-NotebookLM fetcher was supported until August 2026; Google-GeminiNotebook is its successor for Gemini Notebook URLs.
What the research does and does not show
A SIGIR 2026 study compared Google Search, AI Overviews and Gemini Flash 2.5 on a benchmark of 11,500 queries. The researchers also used two time-sensitive query sets, bringing the overall collection to 14,212 queries. For the main comparison, they collected mobile results localized to Newark, New Jersey on 7–8 December 2025. The authors describe their findings as a snapshot, not a forecast for South African businesses.
In the study’s representative real-user query set, AI Overviews appeared for 51.5% of queries. On average, source overlap between the three products scored below 0.2 on the paper’s Jaccard measure, where 0 means no overlap and 1 means identical source sets. That is low average overlap in this sample; it does not mean the products never share sources. The study also found an association between sites disallowing Google-Extended and lower retrieval by AI Overviews. It does not establish that allowing this token causes better Search visibility, and Google’s documentation says it does not affect Search inclusion or ranking.
High organic rankings did not guarantee inclusion in the study’s generative results. The sample’s location, device, collection date and Gemini setup limit how far its results can be generalized. Treat it as evidence that the systems can return different sources, not a prediction of whether a specific South African business will be cited.
Google’s guidance for Search AI features
Google says AI Overviews and AI Mode have no additional technical requirements or special optimizations. Pages must be indexed and eligible to appear with a snippet in Search. Google’s AI optimization guide also says there is no need to create llms.txt, break pages into tiny chunks or target an ideal page length for its AI features.
Focus on the same foundations as Search: make important content crawlable and available as text, use clear internal links, provide a useful page experience, and keep structured data accurate and consistent with visible page content.
What Search Console measures
Google says its Generative AI performance report rolled out to websites worldwide on 31 August 2026. It reports impressions for AI Overviews and AI Mode, grouped by page, country, device and date. It does not provide clicks, exact queries, CTR or average position for those AI features. The report may not appear when a property has too few impressions or when its content is excluded by the Search generative AI control.
Clicks from AI features are included in the overall Search Console Web performance data, but that aggregate does not isolate them. The dedicated report measures impressions, not visits. Analytics and referral reports can show visits that arrive; they cannot show every answer where a business was cited without a click.
For other surfaces, manual checks are snapshots rather than market-wide visibility measures. Keep the query, date, country or location, device and sign-in state with each observation. Repeat the same checks over time before drawing conclusions, and report visits and citations as different measures.
What the South African Gemini figures say
The South African Generative AI Roadmap 2025, produced by World Wide Worx with Dell Technologies and Intel, reported that Gemini usage among surveyed business decision-makers fell from 56% in 2024 to 41% in 2025. The report focused on mid-sized and large businesses, with participation weighted toward sectors including manufacturing, logistics, construction and retail.
Those figures describe reported workplace tool usage in that survey. They do not measure consumer search behaviour or whether Gemini recommends small South African businesses. Google’s supported-countries page lists South Africa for the Gemini web app. Availability does not establish how often local consumers use it or how the app answers a particular query.
A practical check
- In Search Console, review Settings → Search generative AI and confirm the property setting reflects your intended inclusion preference.
- Review the Generative AI performance report if it is available. Note that its impressions cover AI Overviews and AI Mode, not Gemini app citations or visits.
- Review
Google-Extendedseparately against your training and grounding preferences. Do not use it as a Search visibility toggle. - If you sample answers manually, record the query, date, location, device and sign-in state. Keep citation observations separate from analytics visits.
- Keep Search pages useful, crawlable and eligible for a preview.
If you need a documented comparison across Google and other AI search products, IDJoy’s AI search visibility service explains its scope, pricing and limits.