Most business owners have no idea whether AI search is doing anything for them. There are three ways to check, and one of them is sitting in a Google report you already have access to.
Nearly every business owner I speak to has the same blind spot. They’ve been told AI search matters, they might even be paying someone to work on it, and they have no way of telling whether any of it is working.
It’s a fair question to ask, and there are three ways to answer it. One is a Google report you already have access to and probably haven’t opened. Here’s how to read each one, and what the numbers should look like.
Google Search Console now reports on impressions from Google’s AI features, meaning AI Overviews and AI Mode, alongside the standard Performance report. It’s newer than most of Search Console, it’s still being rolled out and refined, and Google is upfront that the data is developing. It’s also the only direct read on this that Google gives anyone.
If you have Search Console set up for your site, open the Performance report and look for the AI features view. Compare it against your overall numbers over the same window, ideally three months, and export both so you can look at pages rather than just the headline total.
Two things to look at, in this order.
Is anything appearing at all? If the answer is zero across three months, your pages are not being pulled into AI answers, and no amount of talk about AI optimisation changes that starting point.
Which pages are appearing? This is the number that tells you something. On one of the sites we built, over the last three months there were 8,451 total search impressions, and 264 of those came through AI features. Around 3% of the total, which doesn’t sound like much until you look at where they landed: 19 different pages, spread across service pages, suburb pages, and blog posts, with at least one appearing on 75 of the 89 days reported.
That spread is the point. It means Google’s AI features have understood that the business does several specific things in several specific places, and it reaches for the right page depending on what was asked. That’s what a site built for this looks like when it’s working.
Now the contrast. On another site, launched much more recently, nearly every AI impression lands on the homepage and almost nothing else. Same country, same kind of local service business. That pattern says the AI knows the brand exists but hasn’t yet worked out what the business does in enough detail to cite a specific page. It’s an earlier stage, not a failure, and the fix is the ordinary work of building out and strengthening the service and location pages.
One limitation to keep in mind. An impression means your page turned up in an AI answer. It doesn’t tell you the person read it, and clicks from AI Overviews get counted inside your normal Search click totals rather than reported separately. So use this view for direction, not for revenue maths.
Your analytics will show some visits arriving from chatgpt.com, perplexity.ai and similar sources. Worth checking, with one important caveat: it undercounts.
A meaningful share of visits that begin in an AI tool arrive with no referrer attached, which means your analytics files them as direct traffic, right alongside people typing your web address from memory. There’s no way to separate those out after the fact.
So treat whatever number you see as a floor rather than a measurement. If it’s growing month over month, that’s a real signal. If it’s small, that doesn’t prove much on its own.
The third method needs no tools and takes ten minutes, and it’s the one most people do wrong.
Asking ChatGPT “what do you know about Smith Plumbing in Ipswich?” will nearly always produce something reassuring, because you’ve supplied the name. The model finds your site and summarises it. That tells you it can read your website, which was never really in question.
The test that matters is the one where you never mention your business. Ask the way a customer with a problem asks:
Then read the answer and note who gets named. If it’s three competitors and not you, that’s your answer, and it’s more useful than any dashboard. Run the same prompts across ChatGPT, Perplexity, Gemini and Claude, because they don’t agree with each other nearly as often as people assume.
Write down what you asked and what came back, with the date. Do it again in a month. That record is the closest thing to a before-and-after you’ll get on the AI side, and it costs nothing.
If you want to understand why the wording of these prompts changes the answer so much, fan-out queries explains what the models are doing behind the scenes.
Being honest about the limits here matters, because there’s a lot of confident nonsense in this space.
None of these three methods will tell you how many enquiries AI search produced for you. The attribution genuinely isn’t available, from Google or from anyone else. Someone can ask Perplexity for a recommendation, see your name, search you on Google an hour later, and call from the map listing. Every tool you have will credit that call to Google Maps.
What you can measure is whether you are present and getting more present. Whether your pages appear in AI answers, whether the number of pages appearing is growing, whether you get named when a customer asks without using your name. Those are the leading indicators, and they move before the phone does.
If nothing is appearing anywhere, start with the basics rather than anything labelled AI. A complete Google Business Profile, business details that match everywhere they appear, pages that answer specific questions instead of describing you in general terms, and some presence on sites other than your own. That’s the foundation, and how AI search works for local businesses covers why.
If your homepage is appearing but nothing else is, your service and location pages are too thin or too similar to each other for anything to reach for one specifically.
If a good spread of pages is already appearing, you’re in a better position than most businesses in the country, and the work is to keep adding depth where the questions are.
Not sure which of those three you’re in? That’s what our AI visibility audits are for, and a free snapshot will tell you the headline before you spend anything. Get in touch and we’ll run the checks and tell you plainly what we find.
Someone has probably told you that your website needs an llms.txt file to be found by AI. Google's own documentation says otherwise. There is one place it gets checked, though, and it isn't where most people think.
Before ChatGPT names a local business, it asks itself a series of sub-questions. Most businesses answer one or two. The businesses that get recommended answer four or five.
Your competitor isn't necessarily better than you. They're just more legible to AI. Here's what Perplexity is actually seeing when it chooses them over you.
Start with a free AI visibility snapshot and see exactly where your business stands across ChatGPT, Perplexity, and Google.
