By Greg Nowak. Updated 1 September 2026.
Google’s Search generative AI control is no longer a limited UK experiment. As of 31 August 2026, Google says the control and its performance insights are available worldwide. That turns an interesting announcement into a live operational decision for website owners.
The default is inclusion. Your pages can appear as links and help ground responses in AI Overviews, AI Mode and generative AI features in Discover. Excluding the site removes that eligibility—and the impressions and visits it might produce.
Do not change the setting simply because AI Search feels promising or threatening. Start by establishing what the content does for the business, what the reports can actually prove and who has authority to make the decision.
What does Google’s AI Search control actually change?
The control sits under Settings > Search generative AI in Search Console. An owner can include or exclude an eligible property. Child properties inherit the closest manually configured parent setting by default, but their owners can override it.
Exclusion prevents the property’s links and content from appearing in the covered generative features or being used to generate their responses and previews. Google says the choice is not a ranking or inclusion signal elsewhere in Search. It also does not override participation in Google Ads or Merchant Center.
Three controls are easy to confuse:
- Search generative AI control: manages appearance and grounding in specified AI Search features.
- Google-Extended: manages certain training and grounding uses for Gemini products through
robots.txt; it does not control ordinary Google Search inclusion. - Noindex: removes a page from Google Search when Google can crawl and process the directive.
A governance record should name the exact control used. “We opted out of Google AI” is too broad to be useful and may be factually wrong.
Start with the decision, not the dashboard
A sensible test question is: Does inclusion give commercially important content enough valuable exposure to justify the organisation’s content-policy concerns?
That question forces the right people into the discussion. SEO or an agency can operate Search Console and interpret visibility. Operations can coordinate the test. Content owners can identify original or sensitive material. Legal or commercial leadership may need to assess licensing, attribution and bargaining concerns.
Classify pages before counting impressions. A service page, support article, paywalled analysis and original research asset do not have the same purpose or risk. Record whether each group primarily supports leads, sales, customer service, authority, subscriptions or licensed access.
| Content situation | Starting position | Evidence to review |
|---|---|---|
| Service and product discovery | Keep included while measuring | AI impressions, landing-page visits, qualified enquiries and assisted conversions |
| General educational content | Usually include initially | Visibility trend, engaged visits and contribution to relevant journeys |
| Licensed, premium or negotiation-sensitive material | Escalate before deciding | Contract terms, access model, attributable demand and commercial policy |
| Customer-support documentation | Assess usefulness and accuracy | Exposed pages, support journeys, outdated versions and governance ownership |
| No meaningful report data | Keep observing unless policy requires action | Longer reporting window, property scope and whether exclusion is already active |
This is not a universal recommendation to remain included. It is a way to avoid treating every URL—and every impression—as equally valuable.
What the performance report can and cannot prove
The Search generative AI performance report covers impressions from AI Overviews and AI Mode. Discover has a separate generative AI report. Data can be grouped by page, country, date and device, then exported.
An impression means a link to the site was shown. It does not prove a visit, lead or sale. The report does not include Search Labs experiments, and the usual Search Console constraints apply, including the 1,000-row table limit. Chart and table totals can also differ because property-level and page-level aggregation work differently.
Google says the insights have rolled out worldwide, although a property may show no report when it lacks sufficient impressions or has been excluded. Do not interpret a missing or nearly empty report as proof that AI Search has no business relevance.
Pair Search Console exports with trusted analytics and CRM outcomes for the same landing-page groups. You may not be able to attribute every session specifically to an AI-generated result, so avoid claiming precision the systems do not provide. The useful question is whether important pages gain exposure and whether their wider search and conversion performance moves in a commercially meaningful direction.
How to run a defensible test
- Record the baseline. Export AI impressions by page, country, device and date. Save ordinary Search performance, landing-page analytics and conversion data for the same period.
- Document the configuration. Record the property, current choice, inheritance status, owner, timestamp and business rationale.
- Choose the scope. If suitable content already lives in a verified child property, its override may allow a narrower test. Do not assume an arbitrary folder is independently controllable.
- Protect the measurement window. Avoid simultaneous template redesigns, major content rewrites, internal-link changes or tracking migrations. Log unavoidable changes.
- Allow for propagation. Google says exclusion generally takes a few days and should begin within one or two days, although caching may extend the transition. Mark those dates rather than forcing them into the before or after period.
- Review a full business cycle. Low-volume sites will need longer. Compare commercially meaningful page groups, not only the site-wide total.
A child-property comparison is useful operational evidence, but it is not automatically a clean A/B test: the content and audience may differ. Seasonality, campaigns and algorithm changes can also affect a before-and-after comparison.
Agree on the decision rules before seeing the result
Continue inclusion when priority pages receive sustained exposure and the surrounding business signals justify it. Consider exclusion when contractual, licensing or governance requirements outweigh the measurable discovery benefit. Extend or narrow the test when the evidence is too thin.
The control emerged partly from a publisher-governance debate, including intervention by the UK Competition and Markets Authority. It is therefore more than an SEO switch. Give the decision an accountable owner, an evidence trail and a review date.
If you need someone to connect Search Console, analytics, content priorities and stakeholder decisions, Greg can help you plan and run the test without turning it into an open-ended reporting exercise.
Related on GrN.dk
- Google’s AI Search Toggle Is a Publishing Decision, Not an SEO Setting
- When Google can call the business, your local data stops being cosmetic
- JavaScript-Heavy Service Pages Still Lose Leads: What to Audit in 2026
Need help with this kind of work?
Plan your AI Search test with Greg Get in touch with Greg.
