Google has rolled out generative AI performance reports and a dedicated blocking setting to Search Console accounts worldwide, according to a report published by TechJuice. For a UK business owner, that changes a question that was previously guesswork into a reportable one. You can see which of your pages are being surfaced inside AI answers, and you can decide whether they stay there.
What follows sets out what the new reports contain, what they leave out, how the opt-out differs from the AI training controls you may already have set at your CDN, and the two situations where switching it on is defensible. If you are still setting up measurement in the first place, start with Google Search Console as an everyday visibility tool and add the generative reports on top.
What the new generative AI reports actually show
The new reporting section works as an expandable tab under the main performance dashboard. According to TechJuice's account of the rollout, it isolates AI Overviews, AI Mode and generative experiences within Discover, and lets you review impressions, identify the specific pages appearing in AI features, compare visibility across countries, and track historical performance in daily or weekly increments.
One omission matters more than the rest. The current version reports impressions and visibility only, and leaves out direct click data, as TechJuice reports. So you can establish presence, and you can establish geography, but you cannot yet establish what that presence returns to you in visits.
Google's John Mueller, posting on LinkedIn and quoted in the same report, set expectations plainly. "The goal is not a written-in-stone absolute truth for position counting (that's impossible), but rather to make something that's useful for site owners in terms of understanding how their site is shown," he wrote, adding that interpreting and comparing these metrics will require work at the site owner's end. Treat the numbers as directional, not as a ledger.
Two different switches that are constantly confused
The common assumption is that opting out of AI training also removes you from AI answers. It does not, and conflating the two is the most expensive mistake available here.
Search Engine Journal, reporting on Cloudflare's new Disallow AI Training setting, describes the split clearly. At Google, the training opt-out works through a Disallow rule for Google-Extended, the robots.txt token Google provides for opting content out of Gemini model training, and Google's crawler documentation says Google-Extended does not affect a site's inclusion in Google Search or its ranking. Appearance in AI Overviews, AI Mode and Discover's generative features is governed separately, through the Search Console setting, and Google's help page says that setting does not affect AI training.
There is a sharper hazard alongside it. Search Engine Journal reports that choosing Block on Cloudflare's Training control now stops Googlebot, Applebot and Bingbot completely, including for search, whereas Disallow AI Training keeps those crawlers allowed for search purposes. Before anyone touches the Google Search Console AI blocking control, check what has already been set at the network layer, because one of those settings is reversible housekeeping and the other removes you from search altogether.
On the Search Console setting itself, TechJuice reports that Google confirmed using the blocking toggle will not act as a negative ranking signal for core web search results outside generative features. That is a vendor statement rather than an independently measured outcome, and it is worth holding as such.
When brand protection justifies opting out
The strongest case for blocking is not about volume. It is about what an AI system says when it speaks for you.
Search Engine Journal's guidance on brand protection in AI search defines the problem as finding where a company or product is represented incorrectly, confused with something else, impersonated, or used to intercept branded demand. The listed risks include search or AI repeating an outdated or false claim, fake sites and accounts impersonating a business, and third parties defining a brand without accurate first-party context, with an AI system then consolidating those fragments into one apparently authoritative answer.
The same piece documents how far the problem now reaches into infrastructure. In one slopsquatting case, the malicious npm package "unused-imports" was registered in place of the legitimate "eslint-plugin-unused-imports", a name AI coding tools often reference. In another, a model invented a package name by combining two real tools, and the resulting command spread through 237 GitHub repositories containing AI-generated agent skills. Nobody was compromised, but the name was there to be claimed.
Scale of error is measurable too. Search Engine Journal cites a February 2026 evaluation of six chatbots across 2,100 news questions in which over 70% of inaccuracies came from retrieval failures rather than model reasoning.
Here is the limit of the remedy, and it is a real one. The Search Console setting stops your pages being used as grounding text and reference links inside Google's AI features. It does not stop a model describing your business from someone else's pages. Blocking can therefore remove your own accurate account of yourself from the answer while leaving the inaccurate third-party account in place. Opting out on brand grounds only makes sense once you know which version of your brand the systems are actually reproducing.
Audit first, then decide
Run the check before the toggle, in this order.
Open the generative AI tab in Search Console and list the pages appearing in AI features, with the countries where they appear. That is your exposure map.
Repeat the brand queries in a clean, logged-out browser. Search Engine Journal advises never running these checks from your own logged-in account, and recording the platform, date, location, language, device and login state for every query, because results still vary by location and device when signed out.
Cover the query types that carry commercial weight, including the exact name, the official website, login queries, ownership queries, reviews and complaints, support queries, comparisons and alternatives, and common misspellings.
Run direct prompts (what the business is, who owns it, whether it is legitimate, how to contact it) and decision prompts (whether to use it, how it compares, what the alternatives are) across the AI systems your customers use. Run each several times in a fresh conversation with memory disabled. Search Engine Journal's author reports seeing repeated runs of the same brand prompt produce different verdicts within the same hour.
Classify every claim in the answers as correct, partly correct, outdated, unsupported, false, about another entity, or drawn from an impersonating source.
Audit per market and per language. Five country and language combinations mean five audits, according to the same guidance.
Owners of multi-location businesses should run step 6 against every service area, which is also the point at which an in-house team and a local SEO company working on your listings need to be looking at the same record.
The click-through argument is reasoning, not a finding
Some publishers will block because they believe removing their grounding text pushes readers back to the blue links. The logic is coherent. It is also unverified, and the new reports cannot settle it, because they carry impressions and visibility but no direct click data. Anyone presenting an expected percentage recovery is estimating.
That uncertainty should shape the decision rather than end it. Treat the Google Search Console AI blocking control as reversible and test it narrowly. Block a defined section, keep the rest in place, and compare impression and query behaviour in Search Console against your own record of orders and enquiries over a period long enough to mean something. Rank tracking will not answer it either, since AI answer layouts differ from ranked results and position counting is, in Mueller's framing, not an absolute truth. Two questions are worth settling before you start. First, how comfortable the business would be absent from an answer a competitor still appears in. Second, does the disclosure and compliance position argue for a lighter footprint in generative surfaces, which is a live question for UK and EU-facing sites under the EU rules on AI content disclosure in search.
What to do this week
Open the generative AI tab and note which pages and which countries appear, because that baseline disappears the moment you change anything. Check your CDN settings next, and confirm you have a training preference rather than a full crawler block. Run the brand prompt set logged out, in every language you sell in, and write down what the systems get wrong. Only then consider the Google Search Console AI blocking control, and only on the pages where a bad answer costs you more than an absent one.
What remains unsettled is the commercial effect. Until the reports carry click data, nobody can tell you what blocking returns, and the honest position is that this is a brand governance decision with an unmeasured visibility cost attached.