The short version
Google's Generative AI report, live for every Search Console property since 11 August 2026, reports one metric: impressions inside AI Overviews and AI Mode. It carries no clicks, no click-through rate, no average position and no query data. Clicks from those surfaces do exist and are counted - they land in the standard Performance report, mixed in with blue-link clicks, where nothing marks them as having come from an AI answer. So the report can tell you that Google is pulling your pages into AI answers, and it cannot tell you what that is worth. Until clicks arrive, the only honest denominator a contractor has is booked jobs.
What actually shipped, and when
Google announced Search generative AI performance reports on 3 June 2026 and finished pushing them to every property on 11 August. They sit under Performance in Search Console as a view separate from Search results. Inside, you get impressions broken out by page, country, device and date, covering both AI Overviews and AI Mode, and an export button. That is the whole feature.
It is worth being precise about why this is a genuine change rather than a cosmetic one. Before it, AI-surface visibility was inferred - from third-party scrapes, from unexplained impression movement, from clients forwarding screenshots of an answer they had been quoted in. This is the first time the number comes from Google's own logs. It is also the first time the number arrives stripped of every field that would let you value it.
The one number it gives you, and the four it withholds
Impressions is the metric. There is no click count, so no click-through rate can be calculated inside the report. There is no average position. And there are no queries: Google does not expose the prompts behind AI activity, which removes the one dimension every SEO habitually reaches for first.
The missing query data is the limitation people underestimate. Page-level impressions tell you which of your pages the model reached for. They do not tell you what question it was answering, so you cannot tell a page that was cited as the definitive answer from one that was cited as a passing reference in an answer about something else.
Your AI clicks are not missing - they are unlabelled
This is the part most people have backwards. A click on an external link inside an AI Overview or inside AI Mode is a click, and Google counts it. It is simply reported in the standard Performance report rather than in the generative AI view, with nothing to distinguish it from a click on a tenth-place blue link.
So the correct mental model is not that AI traffic is invisible. It is that AI impressions have been separated out while AI clicks have not - which is precisely the combination that makes a ratio impossible to build. You have a numerator in one report and a denominator that has been dissolved into another.
Why your click-through rate looks like it fell off a cliff
Here is the mechanism that was not in anybody's release notes. In AI Mode, a follow-up question is treated as a new query: the impressions, position and clicks in that response are attributed to the new question, not to the one that opened the session. A homeowner who asks why their furnace is short cycling, then asks what a fix costs, then asks who does it in Calgary has generated three queries where the old results page generated one.
Impressions scale with queries. Clicks are still made by people, and there is still only one homeowner. Multiply that across a market and site-wide impressions rise while clicks stay flat, so the ratio between them falls. That is arithmetic, not a demotion. Respond to it by rewriting pages that were never downranked and you will spend a quarter fixing a number instead of a problem.
Average position now blends two different rulers
AI Overviews and AI Mode do not measure position the same way. An AI Overview occupies a single position in the results, and every link inside it is assigned that same position. AI Mode uses the ordinary rules, element by element, so its links carry ordinary positions.
Your site-wide average position is therefore an average across two measurement systems that were never meant to be pooled. Treat page-level position as directional, and stop putting the site-wide figure in front of anyone who is going to make a decision with it.
The two readings we keep talking clients out of
The first is triumphant: AI impressions are climbing, so the AI work is paying off. It might be. But an impression with no click attached to it is a reach number, and reach has never been the thing a contractor is buying.
The second is panicked: our click-through rate collapsed, so Google has demoted us. Usually it has not, for the reason above. Check clicks in absolute terms against the same window last year before concluding anything, because clicks are the one number that did not change definition this year.
The trap inside the report: it has no memory
As of the rollout this data is user-interface only. It is not in the Search Analytics API and not in the BigQuery bulk export, so no dashboard tool can pull it and no vendor can rebuild it for you after the fact. Whatever you do not export, you do not have.
The window is short to begin with. Properties are showing generative AI data from partway through 2026 rather than the sixteen months Search Console gives you everywhere else, so your earliest month is the start of the record, not the start of the trend. We now take a dated export of the generative AI view on the first of every month for every property we manage, filed beside the standard Performance export. It is a two-minute job that will turn out to be the only history anyone has of this period.
What this looks like on a contractor's site
Split your pages in two. Diagnostic pages - why a furnace short cycles, what hail damage looks like on asphalt shingles, why a breaker trips when the air conditioning starts - are the ones that will show AI impressions. Money pages - furnace repair in Calgary, emergency plumber in Airdrie - mostly will not, because those queries still resolve to a map pack and blue links rather than to a generated answer.
That split is not a defect to fix. It is the shape of the market, and it tells you what each page is for: the diagnostic pages are buying entity strength and citations, the money pages are buying jobs. What it changes is how you judge them. Do not hold a diagnostic page to a lead target it was never going to hit, and do not congratulate a service page for AI impressions it does not need.
What we changed in our own reporting
Three things, all unglamorous. AI impressions moved into the visibility section of the monthly report and out of the performance section, because they are not performance. Site-wide click-through rate came out of the report entirely and was replaced with clicks and booked jobs.
And there is one comparison we now run every month: the ten pages with the most AI impressions against the ten pages that produce booked work, checked for overlap. Today, on a typical trades site, that overlap is zero. The month it stops being zero is the month AI surfaces have started touching money queries in this market - and that is the signal worth reacting to, rather than to an impression count with nothing attached to it.
Check it on your own site
Export the report before you read it
Search Console, then Performance, then the generative AI view. Set the date range as wide as it goes and hit Export. Values displayed as a tilde or a dash come through as zeros in the download, so when small numbers matter, trust the exported file rather than the chart.
Write down your top ten pages by AI impressions
Then write down what they have in common. On most contractor sites the honest answer is that they answer a question rather than sell a service. That single sentence is most of your content plan for the next quarter.
Pull the same pages out of the standard Performance report
Same date range, same URLs, clicks in absolute numbers. This is the closest thing to a denominator you can construct, and it is only an approximation, because those clicks include ordinary blue-link clicks too.
Compare clicks year on year - never the rate
Clicks kept their definition through this change; impressions and click-through rate did not. If clicks are flat and impressions have doubled, nothing has happened to you that needs fixing.
Check the overlap with the pages that make you money
List the five pages that actually generate booked jobs and see how many appear in the AI impressions list. Most contractors will find no overlap at all today. Re-run it quarterly: the day that number stops being zero is the day this report starts predicting revenue.