Google Now Reports Your AI Search Impressions

If you run a dealership, a law firm, or a restaurant in this valley, someone has already emailed you about AI SEO. The pitch is always the same shape: AI Overviews and ChatGPT are eating your traffic, and you need a new kind of optimization, sold monthly, to survive them. Until this summer you had no way to check even part of that claim against your own site.

Now you do. Google finished rolling its generative AI performance report out to all sites on August 31, 2026, and it costs nothing. We opened ours. What follows is three months of our own numbers, which are published nowhere else, set beside the four public studies that get quoted at you, and then a line by line check of what vendors sell against what Google itself published in May 2026.

What the new report actually counts

Google announced the reports on June 3, 2026, describing them as dedicated views of your impressions within generative AI features on Search, such as AI Overviews and AI Mode, as well as generative AI features in Discover. A note on that post says the rollout to all sites finished on August 31.

The Search Console help page for the report is narrower and more useful. It says the report includes impressions for AI Overviews and AI Mode, and it defines an impression as how many times links to your site were shown to a user in a generative AI feature on Google Search. The breakdowns available are page, country, device and date.

Read that definition twice, because it decides what the number is worth. An impression here is a link to you appearing inside an AI answer. It is not your brand being named in the prose of that answer, and it is certainly not a booked job. Our reading of the help page is that the report carries impressions and nothing else: no clicks, no average position, no query list. If that reading is right, and we think it is, the report can prove you are being surfaced and can never tell you what you were surfaced for.

The same page also lists a reason your report may be missing entirely: your site has not received enough impressions in generative AI features. An empty report is still information. It means the feature is not yet a channel for you, whatever you are being charged to optimize for it.

What ours shows, beside what everyone else measured

Here is our data next to the four studies in circulation. Ours is a single small site, which is the point: it is the size of thing a local owner actually operates.

Our 423 impressions are concentrated, not scattered. Six pages carry 401 of them: our AI consulting page at 225, the homepage at 77, an AI strategy page at 49, an appointment scheduling page at 17, a blog post on TCPA texting at 17, and the receptionist cost post at 16. That is 94.8% of the total in six URLs.

We did nothing special to earn those. No llms.txt file, no bespoke schema, no AEO package. The pages say who we are, where we are, what things cost, and how to reach us. In the classic query report we can see roughly 40 long conversational searches of the kind people type into an AI box, including show me services that offer custom llms based in the las vegas area and i need a company for ai strategy development in las vegas, who should i call. Those queries appear in the classic report, not in the AI report, which lists no queries at all.

Five things vendors sell, checked against Google's own guide

On May 15, 2026 Google published a guide to optimizing for generative AI search. It is short and unusually blunt.

The guide also names the thing that does help a local business appear in AI responses: your Google Business Profile. That is free, it is the asset with the map pack behind it, and it is where we would send the budget.

The 27 percent of our impressions that were not people

The more useful finding in our data has nothing to do with AI. About 27% of our three month impressions, roughly 2,330 of 8,470, were not human.

One query, ai automation, produced 1,496 impressions at an average position of exactly 1.0, with zero clicks, 100% desktop, arriving only on Fridays in bursts of 284 to 315, then stopping dead after August 7. The query ai agents fires only on Fridays. The phrase it consulting las vegas fires only on Wednesdays. Our inference, stated as an inference, is that this is an automated rank tracking scan sweeping the local pack on a schedule, not customers searching.

We were, on paper, ranked number one for a commercial term for months. It was worth nothing. If you have ever been shown a screenshot of a proud number one with no leads attached, this is a likely explanation, and you can check it yourself in ninety seconds: open the query in Search Console, split by device, then by date. Human demand is messy across seven days and mixed across devices. Machines are punctual.

Against the cleaned figure, our real numbers look different and better. Strip the 2,330 and roughly 6,140 human impressions remain. Our 423 AI impressions are then about 6.9% of human search exposure, and our 36 clicks are a 0.59% click through rate on a site that ranks for terms nobody in this town searches often.

Working the arithmetic for two different shops

Take us first, at our actual scale. If our AI impressions kept their September pace of about 30 a day, that is roughly 900 a month. Apply the Pew finding that 1% of visits with an AI summary produced a click on a link inside it, and 900 impressions imply around 9 visits a month. Nine. That is a number you plan around honestly, not one you restructure a website for.

Now take a bigger operation, say a dealership or a mid sized firm with ten times our exposure. Assume 60,000 classic impressions a quarter. At our observed 5% ratio, that is about 3,000 generative AI impressions in the same three months, or 1,000 a month. Run the same 1% and you get roughly 10 visits a month from inside AI answers. Ten times the footprint produced the same order of traffic, because the constraint is the click rate, not your size.

Here is where it flips. If a vendor tells you AI search is costing you money, the Ahrefs figure is the one with teeth: an AI Overview present at position one correlates with a 58% lower click through rate. That is a loss on rankings you already own, not an opportunity in a new channel. For that dealership, the question is not how do I get into AI Overviews. It is which of my existing first page informational pages now sit under an AI Overview, and are those pages the ones that used to convert. Answer that with the classic report, which still gives you clicks, position and queries.

Where these numbers do not hold

Ours is one site, in one city, in one B2B niche, over thirteen weeks. It is not a benchmark. If you sell tacos, your AI impression share will look nothing like the AI consultancy that writes about AI. Treat our 5% as a shape, not a target.

The studies have limits worth saying out loud. Pew is browsing data from March 2025 and 900 US adults, not local intent, not 2026. Ahrefs is desktop and informational only, and Ahrefs itself uses the word correlates rather than causes. The Whitespark local study used city name queries in Houston, Phoenix and Denver, not Las Vegas, and it is from May 2025; we found no 2026 measurement of local AI Overview trigger rates, so nobody actually knows this year's number for local queries. The SOCi index covers multi location brands averaging roughly 127 locations each, which is not you, and SOCi sells the software that fixes what it measures.

The objection worth taking seriously

A fair pushback: all of this measures Google, and the argument is that customers are leaving Google for ChatGPT. Fine. Then look at the one number in circulation that measures assistants directly. SOCi found 1.2% of locations recommended by ChatGPT and 11.0% by Gemini, against 35.9% appearing in the Google 3 pack. Whitespark found the map pack showing on 93% of plain local intent queries, with AI Overviews on 15% of them.

So even granting the trend, the distribution today is lopsided in the same direction as your budget should be. The second objection is better: this is early, and early is when positioning is cheap. We agree, which is why we read the report monthly. The response to early is to watch a free report and keep your Business Profile and your prices current, not to buy a product against a channel producing single digit visits.

The honest summary is that the work which put us in AI answers was ordinary work. Clear pages, stated prices, a named city, a phone number. That is the same thing we tell people about business AI systems generally: the boring, verifiable layer earns more than the novel one.

What to do this week

Three things, none of which require us or anyone else.

1. Open Search Console and find the generative AI performance report. If it is missing, that means too few impressions, which is your answer for now. If it is there, note the total and which pages carry it, and write the number down so next month has a baseline.
2. Audit your classic query report for machines. Take your top three queries by impressions, split each by device and then by date. Any query that is 100% desktop, arrives on one weekday, and never gets clicked is a scan, not demand. Subtract it before you judge your own performance or anybody's reporting.
3. Spend the AI SEO budget on your Google Business Profile instead, since Google's own guide names it as what helps local businesses appear in AI responses. Hours, categories, services, photos, current phone number. Then read our verified list of local AI companies before you hire anyone for the rest.

If you want a second set of eyes on what your own report is showing, our free 15 minute audit is here.