Author: Lance Singbeil
Published:
Your boss keeps asking how well the business is showing up in AI searches. It’s a hard question to answer, and you’ll be asked it more often. At the moment, AI assistants like ChatGPT, Claude, Grok and Perplexity don’t provide reporting. So they have no record of how visible your business is. There’s only one place you can measure it, and that’s Google.
AI Overviews are the AI-generated summaries at the top of the results page, built from web sources Google picks and links to.
Two things make Google worth watching. It handles about 91% of worldwide search activity, so AI Overviews reach more people than any other AI answers do. And Search Console’s generative AI performance report, under Performance, is the only place that reach gets recorded.
What you’re watching for is the trend. The report counts an impression each time Google cites one of your pages inside an AI Overview, broken out page by page. Rising impressions mean Google is citing that page more often. Falling impressions mean it’s citing it less.
A citation is worth having. Brands cited inside an Overview can earn roughly twice the organic clicks per impression of brands left out of it.
Take the trend you see in AI Overviews and apply it more widely. When an AI system gets a question, it runs a search, retrieves pages, and builds the answer from what came back. So the test is the same everywhere: does your page get retrieved, and does it answer the question well enough to quote. If Google is citing a page more, that’s your best available indication the AI assistants are too.
That makes AI Overviews the canary in the coal mine. It’s also the closest thing to an answer for your boss.
Google is committing to AI Overviews
Google seems to be committing to AI Overviews inside the traditional results page rather than making AI Mode the default search experience.
On August 28, Google confirmed that AI Overviews can dynamically expand for some queries. Rather than putting a shortened response behind a Show more button, it can display the fuller answer immediately, and the expanded version exposes the controls for continuing the search conversationally. The Overview is becoming the way into a conversational search from a page people already use.
The reporting is the other signal. Google built a dedicated generative AI performance report and completed the global rollout on August 31, which is a serious investment in a feature it intends to keep.
Trust may be the reason. A field experiment with 1,100 Google users tested an AI Mode-only experience against the ordinary results page. Users clicked through less and reported lower trust in what they found. The traditional results page gives searchers the AI answer alongside conventional results. Take the conventional results away and trust drops.
So the generated answer is being worked into the results page a step at a time. What’s driving the change is, perhaps, Google’s recognition of how search behaviour has shifted.
Searchers are describing their situation
People are getting used to asking questions in full, whether that’s a search engine like Google or an AI assistant. Every AI tool they touch rewards a detailed prompt with a better answer.
Someone looking for commercial insurance used to search:
commercial insurance Vancouver
Today, the same person now is more likely to describe the situation:
Which commercial insurance broker would be a good fit for a 40-person construction company in Vancouver that uses subcontractors and owns expensive equipment?
That second query carries the size of the firm, the exposure they’re worried about and the fact that someone is comparing rather than buying. None of it was available from three words and a city name.
Google doesn’t answer it in one search either. AI Overviews and AI Mode use query fan-outs, breaking a question into the several searches it implies. So your pages compete on the intent the system worked out from these queries.
Some of that intent is recorded in your Search Console query report, in your customers’ own words. Most of it won’t be there, since Search Console anonymizes low-volume queries and the longer a search gets the less likely it is to appear. You’ll spot a few of these just scrolling the report. You can use a filter to pull them all together and see how many there are and which pages they’re reaching.
Two filters that find conversational queries
Long queries surface conversational searches. Open the Performance report, add a Query filter, choose Custom (regex) and use (\b\w+\b\s){10,} to isolate searches of ten words or more. Google records AI Mode prompts as search queries, and John Mueller confirmed this when a practitioner asked why fragments like “yes, pricing” were appearing in her report. What comes back is a mix, since AI Overviews, AI Mode and ordinary search all report into the same place. Read it as how people are phrasing questions rather than as a count of any one feature.
Operator patterns may reveal ChatGPT fan-out. Lily Ray has been tracking how ChatGPT’s fan-out queries are evolving and noticed a surge in searches scoped with site: and modified with “official.” Her filter catches queries containing both. She’s careful about the claim, noting there’s no official ChatGPT Search Console and that Google’s own impression-data problems make the origin hard to confirm. Chris Long tested the filter and reports it works, mostly on sites with larger search footprints.
Treat the second one as a hypothesis you can check on your own property rather than a settled mechanism. Both are worth ten minutes.
Read the two reports together
Search Console’s generative AI report tells you which pages Google is citing and how that’s trending. The filtered Performance report tells you what people were asking when those pages came up. The way your pages are reported connects them.
Take a page whose generative impressions are climbing and look at the conversational queries attached to it. That gives you a working theory about which questions are earning the citation.
Here’s the other case worth looking for. Open the Performance report with the long-query filter on and find a page picking up more conversational queries than it did last month. Then check that same page in the generative AI report. Flat impressions there mean the page is showing up on the results page for those questions without being cited in the answer.
Search Console won’t tell you which query produced a generative impression, so you’re matching two datasets that share a page and nothing else. What you get is a page worth a closer look.
Tracking the trend
Neither report gives you a single number to watch, so the trend has to be built by hand.
Pick a fixed day each month and pull both reports on it. The generative AI report gives you impressions by page. The filtered Performance report gives you the conversational queries reaching those same pages. Record the date with each pull, because you’ll want to line them up later and Search Console won’t do it for you.
Watch a small set of pages rather than the whole site. The pages that matter are the ones carrying your most valuable customer tasks, and a short list is easier to read month over month than a full export.
What you’re looking for is direction. A page whose generative impressions climb across three consecutive pulls is being cited more. A single reading tells you nothing, since impressions move for reasons that have nothing to do with your content.
Two things constrain how fine a reading you can take. The generative AI report has no clicks and no queries in it, so page-level movement is the finest signal available. And it doesn’t separate AI Overviews from AI Mode, so a rising count won’t tell you which of the two is responsible. It’s still labelled Beta, so expect it to change.
Write to what the queries tell you
Take the page you found and read the conversational queries reaching it. Then run searches using those queries yourself.
The Overview will show you who Google cited. If it’s a competitor, open their page and compare it to yours. That’s the most direct evidence available of what Google considered a better source for that question, and it costs a few minutes. Worth doing regularly regardless, since it also shows you where your own page is landing on the results page.
Someone asking about brokers for a 40-person construction firm with subcontracted crews is telling you their size, their exposure and where they are in the decision. That’s a brief. The page either speaks to it, or it doesn’t.
A page saying a brokerage offers commercial insurance solutions for businesses gives a system nothing to match against that question. A page naming the kinds of construction businesses it works with and the risks it handles regularly gives it something to use.
State those facts plainly. “The brokerage ensures construction firms that use subcontracted crews” is a clear fact that can be attached to the business. “Comprehensive coverage tailored to your unique operational profile” is not.
Claims need support behind them. Case studies, credentials and third-party sources describing the business consistently are what make a page usable as evidence rather than as assertion. Then watch the same page in the generative AI report over the next few pulls and see whether the impressions move.
What the AI Overview canary can tell you
Search Console will tell you which of your pages Google is using to build AI answers. No AI assistant gives you anything like it.
The canary works on an assumption. A page Google finds useful is a page ChatGPT or Claude or Perplexity can use too, because they all answer a question the same way, by searching, retrieving and building from what comes back. Watch the citations in AI Overviews and you’re watching the one measurable version of something happening in a lot of places you can’t see.
Next time I’ll go through how to build a set of customer questions, run them across the assistants and record what comes back.
How visible is your business in AI search?
If you’re starting to track AI visibility and want help making sense of what Search Console, AI Overviews and your content are telling you, let’s talk. We can look at where you’re showing up now and where the evidence suggests you should focus next.
