JTSTech Services
All articles

AI Visibility · September 1, 2026 · 8 min read

Google switched on AI visibility reporting for every website on 31 August — and parked an opt-out switch right beside it

Yesterday Google finished rolling the Search Console generative-AI performance report out to every website in the world. For the first time you can open a tool you already have a login for and see how often your pages appeared inside AI Overviews and AI Mode, broken down by page, country, device and date. Two things to understand before you read anything into the numbers. It is not new traffic — Google says that data was already inside your overall performance report, and all that changed is that you can now see it split out. And the same settings menu contains a switch that removes you from those surfaces entirely, which is a defensible choice for a publisher and almost never the right one for a store.

Written by Jeremy Souffir Founder, JTS Tech Services

The short version: as of 31 August 2026, Google has rolled the Search Generative AI performance reports out to all websites worldwide. That is a single line appended to the top of the announcement Google published back on 3 June, and it is the whole news. If you own a Search Console property, you can now open a report that tells you how often URLs from your site appeared inside AI Overviews and AI Mode, split by page, by country, by device and by date. For two years the honest answer to "are we showing up in Google's AI answers?" has been an experiment, a third-party estimate or a shrug. As of yesterday it is a report, for free, in a tool you already use. That is genuinely worth ten minutes of your week. It is also worth being precise about what it measures, because the two most natural readings of it are both wrong, and one of them is expensive.

What exactly went live?

The report is not new — it launched on 3 June to a limited set of sites while Google tested it. What happened yesterday is that the test ended and the gate opened for everyone. Here is what is actually in it, taken from Google's own announcement and help documentation rather than from anyone's summary of it.

  • It covers generative AI features on Search — AI Overviews and AI Mode — and there is a companion report for generative AI features in Discover. Search Labs experiments are excluded, so the numbers describe shipped surfaces rather than tests.
  • The metric is impressions: how often URLs from your site appeared in those features. Google describes it in exactly those terms. There is no click metric and no click-through rate in the report today.
  • You get four breakdowns: pages (grouped by canonical URL), countries, devices for the Search report, and dates — with hourly, daily, weekly and monthly granularity.
  • Google is explicit that this data was already being tracked in your overall performance report and will continue to be. What launched is a separate dedicated view of it, not a newly discovered channel.
  • The usual Search Console table limit applies: 1,000 rows. On a large catalogue that is a real constraint before you have begun.
  • Sites that have opted out of generative AI features do not appear in the report at all, which is the first hint that the switch and the report are two halves of one thing.
  • Google's closing line in the announcement is that it is "continuing to work with website owners to understand what insights and data would be most helpful… such as adding additional metrics over time". Read that as: clicks are a known gap, and there is no date attached.
What actually changed on 31 August. The flow on the left is unchanged — the same visits, counted the same way, landing in the same total on the right. What is new is the branch dropping out of the middle into its own panel: the AI-surface share of that total, previously blended into one number, now readable on its own. No new volume entered the system. A partition appeared inside it.
What actually changed on 31 August. The flow on the left is unchanged — the same visits, counted the same way, landing in the same total on the right. What is new is the branch dropping out of the middle into its own panel: the AI-surface share of that total, previously blended into one number, now readable on its own. No new volume entered the system. A partition appeared inside it.

Why does this report exist at all?

Worth knowing, because it tells you something about where the report is heading. Google's own announcement gives no reason beyond helpfulness, and we are not going to put words in its mouth. What is on the public record is that in October 2025 the UK's Competition and Markets Authority designated Google as having strategic market status in general search, and that on 4 June 2026 — the day after these reports launched — the CMA published a final decision imposing a publisher conduct requirement, obliging Google to give publishers a control over whether their content is used in AI-powered search features including AI Mode and AI Overviews, with up to nine months to implement it. Coverage at the time noted that the first tranche of the reports went to UK sites and read the timing against that decision. We would put it no more strongly than that: the sequence is documented, the causation is inference, and Google has not claimed it either way.

The practical reason to care about the provenance is what it implies about the roadmap. The CMA's materials on publisher controls have been read by trade press as expecting engagement data that goes further than impressions — click-throughs and click-through rates, separated out from ordinary organic search. That is reported interpretation rather than a Google commitment, and no public date binds it. But if you are deciding today whether to build a reporting habit around this, the reasonable bet is that the report gets richer rather than being withdrawn, and that a baseline started this week ages well.

What this report can and cannot tell you

Four limits, stated plainly, because every one of them will otherwise be discovered by someone in a meeting. First: impressions are not clicks and not visits. An impression means a link to your page was shown inside an AI feature. It does not tell you whether anyone followed it, and there is no click column to check. Second: there is no query-level view here, and per published testing as of mid-August the generative-AI data was not exposed through the Search Analytics API or the BigQuery bulk export — meaning it is a screen you read rather than a feed you pipe into your own reporting, and the 1,000-row cap bites on any real catalogue. Third: it tells you that you appeared, not how. Being cited in a footnote under a competitor's recommendation and being the answer both count as one impression, and the difference between those two is the entire commercial question. Fourth, and the one people forget: this is Google. It covers AI Overviews and AI Mode and nothing else. ChatGPT, Copilot, Perplexity and the Gemini app publish no equivalent, so a complete picture of your AI visibility is not something this report can give you no matter how carefully you read it.

The switch beside the report

In the same corner of Search Console, under Settings, sits the Search generative AI control. It has three positions — include, exclude, or inherit from a parent property — and it decides whether your content can appear in AI Overviews, AI Mode and the generative AI features in Discover. Include is the default. Some of what Google says about it is more reassuring than people expect, and some of it is the reason we are writing this section at all.

  • Excluding does not affect your ordinary Google Search results. Your blue links stay where they are. Google states this directly.
  • The setting is not used as a ranking or inclusion signal elsewhere in Search. There is no quiet penalty for opting out, and the CMA's conduct requirement includes an anti-retaliation element on exactly this point.
  • It is a blunt instrument by design. Per the CMA's published decision, a single control covers AI Mode, AI Overviews and other AI-powered search features together, and covers crawling, training and grounding together. You cannot decline to be summarised while remaining eligible to be recommended.
  • Exclusion takes a few days to take effect, and Google notes caching and propagation can make some content linger longer.
  • Google says plainly that similar content from other sites may still appear in those features even when yours is excluded. The surface does not go dark. You do.

For a news publisher whose complaint is that an AI summary answers the reader's question so completely that the click never happens, that trade is arguable and reasonable people are making it. For a store, it is almost always the wrong lever. Your pages are not the answer being given away — your products are the thing being chosen between, and the last bullet is the one that decides it. Opting out does not remove the AI answer from your category. It removes you from it and leaves your competitors in it.

What the opt-out actually does. The lit band is the AI surface where the answer is assembled; the blocks standing in it are the sites eligible to be drawn from. Flipping the switch moves one block out of the light and off to the side. The band does not go out, and it does not narrow — the remaining blocks simply fill it. That is the whole trade, and it is why the choice looks different for a publisher than for a store.
What the opt-out actually does. The lit band is the AI surface where the answer is assembled; the blocks standing in it are the sites eligible to be drawn from. Flipping the switch moves one block out of the light and off to the side. The band does not go out, and it does not narrow — the remaining blocks simply fill it. That is the whole trade, and it is why the choice looks different for a publisher than for a store.

The two conclusions that both get this wrong

The first wrong conclusion is that something happened to your traffic. It did not. Google's announcement says the generative-AI data was already inside your overall performance report and stays there; a dedicated view of it was added. So if you open the report on Monday and find a substantial impression count you had never seen, nothing arrived over the weekend — you were being shown in AI answers all along and were counting it without knowing which part of the total it was. This matters because the reflex is to present the number as growth, and it is not growth, it is a partition. We wrote almost this exact paragraph on 25 August about Merchant Center, for a different product and a different reason, which should tell you how routinely reporting changes get read as performance changes. The second wrong conclusion is the more costly one and it comes from the other end: seeing the impression count, deciding AI answers are cannibalising the clicks you deserve, and reaching for the exclude switch. Read the mechanics before you touch it. It is one switch across every AI surface and across crawling, training and grounding, your competitors stay in the answer either way, and you have no click data in this report with which to have actually established that cannibalisation was happening. Both mistakes share a root: treating a new measurement as a new event. Nothing changed yesterday except your ability to see it — and a measurement you have had for one week is not yet evidence for a decision that hard to reverse.

What would we actually do this week?

  • Open it and confirm you have it. Search Console, Performance, and look for the generative AI view. Coverage is worldwide now, but a site needs enough impressions to show anything, so an empty report is a finding rather than a bug.
  • Take the baseline and get it out of the tool. Export the page-level and country-level tables today and store them somewhere you control. With no API, no BigQuery export and a 1,000-row cap, a manual export is not a workaround, it is the mechanism, and the history you have in six months is exactly the history you chose to save.
  • Compare AI impressions to standard impressions per page, not in aggregate. The aggregate ratio tells you almost nothing. The page-level ratio tells you which of your page types the AI surfaces actually draw on — and if it is your guides and comparison pages rather than your product and category pages, that is the most useful thing this report will ever tell you.
  • Check the switch is set to include, on every property, and look at the inherit setting on child properties. This is a two-minute check and the failure mode is silent. If someone tidied settings in the spring, you want to know now rather than from a chart.
  • Annotate 31 August in whatever your team actually opens. The date a report changes should always be written down beside the curve, for the same reason we said it about 24 August a week ago: in eight weeks nobody will remember why the shape changed, and the person presenting it will not be the person who read this.
  • Do not rebase targets or restructure spend on one week of impressions. There is no click data, no query data and no history. Anyone proposing a budget change on the strength of this report today is proposing it on the strength of a number they met on Tuesday.
  • Check your edge is not blocking Google's crawlers, again. We repeat this in nearly every post on this subject because it keeps being the actual cause. A report showing zero impressions is far more often a bot rule added for good reasons in the spring than it is a content problem.

The genuinely useful part

Set the caveats aside for a paragraph, because the underlying development is good and it is rarer than it sounds. A closed system that decides whether your business gets mentioned has just started reporting, to you, first-party and at no cost, how often it mentioned you — and it is doing so in a tool your team already knows how to open. Compare that to the state of play in January, when the only options were to run manual prompts and count, or to buy a third-party estimate built from a sample of somebody else's queries. The report is thin, it is Google-only, and it is impressions rather than outcomes. It is still the first honest instrument any of us have had on this surface. And the work that improves the number is the same work that has improved every other number we have written about this year: pages that answer the question in their opening sentences, real prices and real delivery facts in plain text, comparison and use-case pages that exist at all, and crawlers that are not being turned away at your own front door. It has now been the right answer under every study, protocol, regulation and report we have covered. At some point that stops being a coincidence.

Where we fit

The reason a business would bring us in over this is not the report, which you can open yourself in a minute and should. It is the four-fifths of the picture the report does not contain. Google now tells you how often it showed a link to your page inside its own AI answers. It does not tell you what was said about your product, whether you were the recommendation or a footnote beneath somebody else's, or what happens on the four other assistants your customers are using — and it gives you no clicks, no queries and no export with which to work any of that out. That gap is what an AI Shopping Visibility audit is for. We run your best-selling and highest-margin products through ChatGPT, Gemini, Copilot, Perplexity and Google's AI surfaces and write down what actually comes back, we set that beside your new Search Console baseline so the two measurements check each other, we identify where a competitor is being recommended in your place, and we trace which of those gaps come from something in your own pages or your own edge configuration rather than something outside your control. Then we tell you which ones are worth fixing first. The report is the free half of the instrument. This is the half that tells you what the reading means.

Sources

  • Google Search Central — Introducing Search Generative AI performance reports in Search Console (the primary source for everything factual above: the 3 June launch, the metric and breakdown list, the hourly-to-monthly granularity, the statement that this data was already in the overall performance report, the note about additional metrics over time, and the line appended at the top of the post — "As of August 31, 2026, we've rolled out these insights to all websites worldwide")
  • Google Search Console Help — Generative AI performance report (Search) (the impressions definition, the page/country/device/date breakdowns, the 1,000-row table limit, the exclusion of Search Labs experiments, and the note that opted-out sites do not appear)
  • Google Search Console Help — Search generative AI control (the include/exclude/inherit settings, the surfaces affected, the statements that exclusion does not affect regular Search results and is not used as a ranking or inclusion signal, the few-days propagation delay, and the note that similar content from other sites may still appear)
  • Search Engine Journal — Google Search Console AI reports rolled out worldwide (independent confirmation of the 31 August worldwide rollout covering both the reports and the opt-out control, and the source for the reported expectation that click and click-through-rate data is meant to follow)
  • Search Engine Journal — How to track Google AI Mode traffic in Search Console (the published mid-August testing behind our statement that the generative-AI data was not exposed through the Search Analytics API or the BigQuery bulk export, plus the practical effect of the row cap and the anonymised-query pool)
  • Professional Publishers Association — CMA publishes final decision on Google Search publisher controls (the 4 June 2026 final decision, the requirement for a publisher control over content use in AI-powered search features, the fact that it is a single control across AI Mode, AI Overviews, crawling, training and grounding rather than a per-feature one, and the nine-month implementation window)
  • Competition and Markets Authority — Final decision: strategic market status investigation into Google's general search services (the underlying designation that makes those conduct requirements possible — the primary document, if you want the regulator's own words rather than ours)
  • JTS Tech Services — Google restated your Merchant Center numbers (the 24 August definitions change in a different Google product, and the closest recent example of a reporting change being read as a performance change)
  • JTS Tech Services — The EU just classified ChatGPT as a search engine (yesterday's post, and the other half of this week's pattern: disclosure prised out of an assistant by a regulator, arriving in December rather than today)
  • JTS Tech Services — What ChatGPT actually reads from your site (the mechanism underneath the page-level advice above — what an assistant reads, how much of it, and when)

Keep reading

Free AI Shopping Visibility audit

Google will now tell you that you appeared. Do you know what was said?

We run your best-selling products through ChatGPT, Gemini, Copilot, Perplexity and Google's AI surfaces, set the results beside your new Search Console baseline, show you where a competitor is being recommended in your place, and tell you which of those gaps trace back to your own pages.