How do you spot AI tracking tools in your Search Console?

The short answer

A query that reads like a prompt, repeats like a schedule and never clicks is probably a tool. You cannot prove it from Search Console alone, because Google does not tell you who searched. You can get close by looking at shape, timing and history together, and you should, because counting tool traffic as demand will steer your content plan somewhere nobody is standing.

What we saw

These are the rows from our own Search Console that prompted this piece, taken from screenshots we saved between 17 and 27 September 2026. The queries are verbatim.

The first two rows are the interesting ones. Both open with a job title. Both are full sentences with punctuation. Both ask about measuring AI citations, which is the exact job the tools that run them are sold to do. And the first one was already in our Search Console before we had published anything that answered it, which means something was asking it independently of us.

Why a tool shows up as a search

Three documented facts, and one inference that joins them.

• You cannot see an AI Overview without running the search. AI Overviews and AI Mode only exist on Google's results page. A tool that reports whether a brand appears in them has to put the prompt into Google and read what comes back. Several of the platforms we compared track exactly these two surfaces.
• Automated searches have inflated Search Console before. When Google disabled the num=100 results parameter in September 2025, which let a single request pull a hundred results at once, sites widely reported impressions falling and average position improving overnight. The explanation practitioners settled on was that rank-tracking tools had been generating impressions at deep positions that no person ever scrolled to.
• Persona prompts are a sold feature. Persona-based prompt tracking, running the same question as a budget owner, an evaluator or an executive, is described as a standard practice by vendors in this category, because answers change with the asker's framing.

Put together: a tool running "i'm a paid media director..." against Google every day, on behalf of a customer who wants to know whether their brand is cited, would produce exactly the row we saw. That is our inference, not something Google has confirmed, and the next section is about what we still cannot square.

Six signals

No single signal is proof. Three or more together are a strong hint.

1. Persona voice. The query opens with a role: "i'm a", "we are a", "as an". People rarely introduce themselves to a search box. Prompt templates do it constantly.
2. Sentence length. Twelve words or more, with full stops or question marks inside. Average search queries are a few words long, and we published our own numbers on that.
3. Recurrence at one impression. The identical string appears day after day, one impression at a time. Nobody retypes the same 23-word sentence on Monday, Tuesday and Wednesday.
4. No clicks at a good position. A result at position one to three with impressions but no clicks, over weeks, suggests the reader is software that records the page and leaves.
5. It predates your page. The query was already there, ranking you somewhere deep, before you published anything aimed at it. Someone was asking it on their own schedule.
6. Tooling words. "test", "prompt", or the name of a category a tool vendor would track, sitting inside an otherwise human-looking query.

The part we cannot explain

Google says the queries it hides are the rare ones. Its documentation describes anonymised queries as those not issued by more than a few dozen users over a two-to-three month period, and they never appear in the query table.

So by that definition, a 23-word sentence that does appear in the table was issued by more than a few dozen users. A person retyping it would not clear that bar. A tracking service running the prompt for a customer from many addresses might, if Google counts each origin as a user. It might also mean something about the prompt is common across many customers of the same tool. We do not know how Google counts users for automated traffic, and we have not found anyone who does.

This is the most interesting open question in the whole piece, and it cuts both ways. It makes a single curious human very unlikely. It does not tell us which tool, or how many customers are behind it.

How to separate them in Search Console

Search Console's query filter accepts regular expressions, matches anywhere in the query by default, and ignores case. Choose "Custom (regex)", then "Doesn't match regex" to take the suspects out of your view.

Run each as "Matches regex" first and read what it catches before you exclude anything. The twelve-word rule will also catch genuine long questions from real people, and those are exactly the ones you do not want to lose.

One caution about totals. When you filter by query, Search Console drops anonymised queries from the chart as well as the table, so the chart total will change for reasons unrelated to the tools. Compare filtered views with filtered views.

What they are still good for

Tool traffic is not demand, but it is not worthless either.

• Someone is paying to watch that question. A prompt in a tracking tool is a question a business cares enough about to monitor every day. That is market research you did not pay for.
• You are probably in their report. If your page ranks first for a prompt a tool is tracking, your URL is likely showing up in that customer's dashboard as a source. We cannot verify this from our side, and we would not build a strategy on it, but it is a form of visibility to exactly the people who buy this kind of work.
• It works the other way too. If you run a tracking tool yourself, your prompts may be appearing in the Search Console of every site that ranks for them. Treat anything you put in a prompt as something a stranger could read.

What we got wrong

We read these as people. When the second persona query arrived, we wrote in our own plan that two questions from political and advocacy communications looked like a sector signal, and that a third would justify building a section for that audience.

If the queries are one customer's prompt list, that reasoning collapses. Three prompts from one tool are one customer, not a market. We have dropped that rule from our plan. What survives is narrower: a page that answers a specific question directly can rank first for it, whoever is asking. That is still worth knowing. It is just not evidence of demand.

Limits

• Search Console does not show who ran a query, so everything here is inference from shape, timing and history. We have not identified the tool behind any of these rows.
• Our data is one small site over ten days, read from saved screenshots rather than a full export. A daily export by query would make the recurrence pattern exact rather than approximate.
• The num=100 episode shows automated searches can reach Search Console. It does not show that every AI tracking tool's searches do, or that Google counts them the same way today.
• Some long, persona-style questions are genuinely typed by people, particularly now that people have learned to talk to search boxes the way they talk to chat assistants. The filters above will catch some of them.
• We sell SEO and AI visibility work, and we do not sell or resell any tracking tool. We compared the main ones separately.