Why your Search Console data might look rosier than it really is

A lot of marketers are watching their Search Console impressions fall at the moment. So it might sound odd to hear that your Search Console data could still look rosier than it really is. But some of the impressions in there don’t come from people at all. They come from AI visibility tracking.

It’s worth being clear which data we mean, because there are two reports. The Performance report, the one most people check, covers Google search as a whole, including the ordinary listings. The Generative AI report is newer and only covers Google’s AI Overviews and AI Mode. Tracking can add impressions to both.

If part of your impressions come from tracking, fewer real people are seeing your site than Search Console shows. When impressions are falling, the real fall could be steeper than it looks. When they’re rising, some of that rise may not be people at all.

We know this because we run AI visibility tracking ourselves: thousands of checks a day across ChatGPT, Claude, Gemini and Google’s AI answers, for the businesses we work with. Since September, we’ve been watching what those checks leave behind in Search Console.

Impressions aren’t the best way to judge AI visibility anyway. The truest reading comes from a tracked prompt set: a fixed list of the questions your customers ask AI, checked the same way on every engine. Tracking can’t inflate it, and it isn’t tied to any one platform.

Why does tracking show up in Search Console?

Checking what Google’s AI says means running a real Google search. And Search Console counts that search the same way it counts one from a customer.

Our tracking runs through a data provider. It puts each question into the normal, consumer version of Google, in every market we track, every day, much like a person searching in a private browser window. Google shows its AI Overview or AI Mode answer, and Search Console records an impression for the sites that appear.

Tracking never clicks on anything. It’s only taking a reading of what Google’s AI says, but each of those readings can still end up in your impressions.

How tracking affects your Performance report

Some of the questions we track turn up, word for word, as queries in the Performance report. That’s because an AI Overview sits on top of an ordinary page of results, and each check loads the whole page, so the sites listed underneath can get an impression too.

We’ve seen tracked questions appear in our own Search Console data. They don’t look like what people normally type into Google: they’re long, conversational questions, the kind people ask AI.

You’ll only see some tracked questions in your queries. Search Console hides queries that very few people search, so many never appear by name, and the ones you can see are only part of the picture.

How tracking affects your Generative AI report

Tracking adds impressions to the Generative AI report whenever a tracked AI Overview or AI Mode answer links to your site. The report doesn’t show the searches behind its impressions, so you can’t pick the tracking out, but you can see it in the shape of the line.

For a B2B client we track in six markets, Generative AI impressions stepped up almost overnight on the day tracking began. When we strip the numbers back to their trajectories, the impression line follows the volume of answers we collect each day. We can’t label every impression as ours, but the two lines move together closely enough to draw the inference.

The Generative AI report only shows impressions. There are no clicks and no queries, so there’s nothing in it to check a change against. These AI impressions are also counted in your Performance report totals, so tracking can lift impressions there while clicks stay flat, which can drag your click-through rate down.

How much tracking are we talking about?

One brand’s tracking can mean anywhere from a few dozen to a few thousand Google searches a day, depending on how many questions and markets it covers. Each question is usually checked in both AI Overviews and AI Mode, so it counts twice.

A business tracking 100 questions in one market creates 200 Google AI checks a day. A business tracking 200 questions in six markets creates 2,400. That’s before the same questions go to ChatGPT, Claude and the other engines.

For our six-market client, 703 of the AI answers we collected in the UK in one week linked to its site. That’s about 100 a day from one market, and each one is a chance for Search Console to log an impression that no customer created.

Your competitors’ tracking ends up in your numbers too

Any tool checking questions in your market adds impressions for every site that appears in the answers, whoever’s paying for it. So your impressions can be inflated even if you don’t track anything yourself.

In one week of UK checks for our six-market client, only about one link in 20 pointed to the client’s own site. Its most-linked competitor appeared in 529 of the answers we collected, against 703 for the client. That’s three answers for the competitor for every four for our client.

The more businesses track their AI visibility, the more impressions everyone collects without a single extra customer. That can prop your numbers up even when real demand is falling, which is how they end up looking rosier than they are.

ChatGPT’s business tools would be a big step for attribution

If OpenAI launches the business tools it’s reported to be building, it’ll be a big step for anyone investing in AI search. For the first time, businesses could see how often ChatGPT mentions them and how often people click through to their site.

Early signals, reported in November 2026, point to a ChatGPT Business Hub where verified businesses can claim their presence and manage how they show up. A banner spotted in ChatGPT mentions three things. Businesses could track how often they’re mentioned in conversations, show customers what they offer and keep their details up to date, and set up a business agent that answers customer questions using information they control. OpenAI hasn’t announced any of it yet, so the details could change.

Better attribution is what we’d most like to see. Proving the return on AI search work, from tracking to content, means connecting visibility to visits, enquiries and revenue, and mention and click data straight from ChatGPT would make that much easier.

For reading visibility, though, ChatGPT’s figures look likely to have the same limits as Search Console. They’d only cover one platform. And early reports suggest they won’t show the questions behind each mention, so you couldn’t tell whether a mention came from a customer or from a tracking tool asking the same thing every morning, or see where a competitor was named instead.

Platform data and a tracked prompt set work best together: one for attribution, the other for visibility. If you update your business details or products in ChatGPT, a tracked prompt set is how you’d check whether the answers actually change, in ChatGPT and on every other engine.

Why a tracked prompt set gives the truest reading

A tracked prompt set is a fixed list of the questions your customers ask AI, checked on each engine and in each market on a schedule. It gives the truest reading of AI visibility, because tracking can’t inflate it and it works the same way on every engine.

A tracked prompt set can’t be inflated because it measures visibility as a share of answers: out of all the answers to your questions, how many named you. Running more checks gives you more readings, not more mentions. For our six-market client, UK AI Overview visibility across all 200 questions was 22.2% on the first day of tracking and 21.0% on day 50, after 9,967 checks of the same questions.

A tracked prompt set isn’t tied to one platform either. The same questions go to ChatGPT, Claude, Gemini and Google’s AI answers, so you get one reading you can compare across all of them. Search Console only covers Google, and OpenAI’s tool would only cover ChatGPT.

What actually moves the reading is content. On the questions one new article was written to answer, the share of answers naming our client went from 0.5% to 3.7% after it went live.

A tracked prompt set also shows what impression counts can’t: whether you’re named or just linked, how you’re described, which competitors appear next to you and which sources the engines rely on.

What makes a tracked prompt set worth trusting

A tracked prompt set is only as good as the questions in it. Build it on the wrong ones and you’ll get a very precise reading of the wrong thing, which is why we agree every client’s set with them before tracking starts.

• Ask what customers ask. The questions should match how real customers phrase things at each stage of the customer journey, not the keywords a business wants to rank for.
• Keep it mostly unbranded. If a question names your brand, the answer nearly always will too, so we aim for around nine in ten questions that don’t.
• Cover the markets and engines that matter. A question checked on one engine, in one country, only tells you about that engine in that country.
• Keep it stable. Asking the same questions in the same way is what makes this month’s reading comparable with last month’s.
• Read the trend, not the day. Answers change a little from day to day, so look at the line over several weeks.

If you’d like that kind of reading for your own brand, we can build a tracked prompt set around the questions your customers ask and show you where you stand on the engines that matter to you.

How to read your Search Console data now

Until the reports can separate people from software, treat impressions as a rough guide rather than a verdict. A few habits help.

• Note when tracking starts or changes. Search Console lets you add notes to dates on its charts, so any jump has its explanation right there.
• Look for your tracked questions in your queries. Exact matches with impressions and no clicks are tracking, not new demand.
• Check impressions against clicks, visits and enquiries. If impressions move and nothing else does, look at tracking first.
• Ask how many checks are run for you. Whoever runs your tracking should be able to tell you how many questions, on which engines and in which markets, every day.

How we looked at this

We used our own tracking data alongside Search Console for the sites we track, including our own, and we’ve been watching this since September 2026. The tracking figures cover the seven days to 5 November 2026, plus one client’s first 50 days of tracking.

The details on ChatGPT’s business tools come from early reports in November 2026, which OpenAI hasn’t confirmed. Client names, industries and competitors are anonymised. You can find more studies built on our own data in our research.

Frequently asked questions

Can AI visibility tracking inflate Search Console impressions?

Yes. Tools that check Google’s AI Overviews and AI Mode run real Google searches, so Search Console records impressions for the sites in those results. That can add impressions to both the Performance report and the Generative AI report without a single extra customer seeing your site.

If my impressions are falling, how can my Search Console data look rosier than it is?

Some of the impressions you’re still getting may come from tracking rather than people. If part of your impressions come from AI visibility tracking, fewer real people are seeing your site than Search Console shows, so the real fall could be steeper than the chart suggests. And as more businesses track their AI visibility, that share is likely to grow.

What’s the difference between the Performance report and the Generative AI report?

The Performance report covers Google search as a whole, including the ordinary listings, while the Generative AI report only covers Google’s AI Overviews and AI Mode. Tracking can inflate both, because AI impressions count in both reports and the ordinary listings under an AI Overview count in the Performance report. Only the Performance report shows queries, so it’s the only place tracked questions can appear by name.

Can competitors’ tracking affect your Search Console data?

Yes. Any tool checking questions in your market adds impressions for every site that appears in the answers, whoever’s paying for it. In our data, one client’s most-linked competitor appeared in three answers for every four that linked the client, so tracking spreads impressions across a whole market.

Will ChatGPT’s business tools help with attribution?

Yes, if they launch as reported. Early reports describe tools that show verified businesses how often they’re mentioned in ChatGPT conversations and how often people click through to their site, which would make it much easier to connect AI visibility to visits and revenue. They aren’t expected to show the questions behind each mention, though, so they’ll work best alongside a tracked prompt set rather than instead of one.

Why is a tracked prompt set the truest reading of AI visibility?

A tracked prompt set is a fixed list of the questions customers ask AI, checked on each engine and in each market on a schedule. It measures visibility as the share of answers that name a brand, so running more checks adds readings, not mentions, and tracking can’t inflate it. It also asks every engine the same questions, so you get one reading you can compare across ChatGPT, Claude, Gemini and Google’s AI answers.

Does asking AI the same question every day make it more likely to mention your brand?

No. Each check is a fresh reading of what the engine says, and asking again doesn’t teach it anything. In our data, one client’s visibility ended its first 50 days within about a point of where it started, after nearly 10,000 checks of the same questions, while content written for specific questions lifted visibility on those questions.