Google has turned AI visibility from a debate into an operational task. On August 31, 2026, Google marked its Search Console generative-AI performance reports and Search generative AI control as available worldwide. For publishers, ecommerce teams and B2B sites, the immediate question is not whether AI search is good or bad. The question is who owns the setting, who reads the report, and what evidence is strong enough to change eligibility.
The wrong move is to treat the new control as a protest button or as a growth lever. It is neither. It is a property-level governance decision that can remove a site from AI Overviews, AI Mode and generative AI features in Discover. The report, meanwhile, is a visibility lens. It can show where pages appeared, but it does not yet answer every attribution question a CMO will ask.
What changed on August 31
Google’s Search Central post says the dedicated reports show impressions, pages, countries, devices and dates for generative AI features in Search and Discover. Search Engine Land also noted the practical limit: the AI performance report includes impressions and dimensions, but not click data. That makes it useful for visibility diagnosis, not a complete revenue model.
The related Search generative AI control sits on the eligibility side. Google says site owners can decide whether their links and content can appear in and help ground generative AI Search features. Google also says sites that opt out will not receive traffic or impressions from those features, while the setting is not used as a ranking signal outside generative AI Search surfaces.
What the report can and cannot prove
The report can answer four useful questions: which pages appear in AI features, which countries are exposed, whether mobile and desktop behave differently, and whether visibility changes after content, technical or authority work. That is enough to start a dashboard for AI search operations.
It cannot prove that an AI impression created a lead, sale or subscription. It also cannot separate brand value from lost clicks by itself. Teams should resist turning the first week of data into a budget argument. Use it as a baseline, then connect it with landing-page demand, branded search, assisted conversions and sales feedback.
The include-or-exclude decision model
Use four filters before changing the control. First, classify the property. A public product, service or editorial site normally wants discovery unless there is a clear legal or commercial reason to restrict it. Second, map the content risk. Regulated advice, licensed data and syndication-heavy content may require review before inclusion.
Third, check property inheritance. A subdomain or URL-prefix property may inherit a parent decision, so a legal choice made for a corporate property can affect a market site or store without the marketing team noticing. Fourth, define a measurement window. Do not change the control and rewrite the content strategy on the same day; you will not know which action moved the numbers.
How to run the first 30 days
Start with a baseline export by page, country and device. Tag pages by commercial role: lead generation, product category, editorial authority, support, legal or partner content. Then review pages with AI impressions but weak downstream evidence. They may need clearer structure, stronger images, better source depth or a more useful answer to the query.
For most companies, the default should be disciplined inclusion plus monitoring. Exclusion is a business decision for content where AI-surface exposure creates more risk than discovery value. If the team cannot explain the lost visibility it is willing to accept, it is not ready to use the control.
The CMO decision
The rollout gives marketing leaders a simple operating rule: SEO owns the report, legal and leadership own the exclusion threshold, and analytics owns the connection to business outcomes. A weekly review is enough at the start. The goal is not to chase every AI impression. The goal is to know where the brand is eligible, where it appears, and when the trade-off between visibility and control is worth making.
