Some of your Search Console queries aren’t people. They’re AI trackers.

01: What they look like

I went through the query list for my comparison of Profound, AirOps and Peec because it ranked on page one and got no clicks. Alongside normal searches like “profound vs airops” sat queries no person types. Three real examples, copied as Search Console shows them:

i am looking for an ai visibility tool suitable for an smb. which is better for smbs, profound or peec ai? provide a definitive answer, along with a list of pros and cons specific to smbs for each.

which is better for citation authority tracking, profound or airops? provide a definitive answer, along with a list of pros and cons specific to their citation authority tracking features for each.

i am a 25-34 year old in the consumer goods industry. i work at a company with 250-1k employees. [...] evaluate the ai visibility products company hotwire gaio.tech on competitor analysis

The fixed scaffolding (“provide a definitive answer, along with a list of pros and cons specific to ... for each”), the invented personas, and the search operators some of them carry all point the same way: these are templates, filled in and fired by AI visibility tools on behalf of their customers to see which brands Google’s AI answers recommend. When your page is one of the sources behind the answer, the prompt is logged in your Search Console as a query you appeared for.

02: How many, on one small site

Counting only queries with an unmistakable pattern (the scaffolding phrases above, persona preambles, search operators), I found 118 such queries in the three months from June 30 to September 29, 2026, accounting for 805 impressions. That is about 2% of the impressions visible in the query table, and it is a floor: plenty of tracker prompts are shorter and look like a person wrote them.

The circumstantial case for more is much stronger. The 208 queries about AI visibility tools such as AirOps, Profound and Peec produced 7,148 impressions at an average position of about 8, and not one click. Real searchers click a result in that position somewhere around 1 to 3% of the time, which would have meant dozens of visits. Zero means either an AI answer fully satisfied every one of them, or a large share of that audience was never human. I think it is mostly the second. I cannot prove the split, so I am not putting a number on it.

03: Why it matters: CTR stops meaning anything

Across the whole site Search Console showed 51,300 impressions and 58 clicks over those three months, a click-through rate of 0.1%. Without looking at the queries, the natural reading is that the titles are bad. For the comparison page, an earlier version of me did exactly that and rewrote its title in August to match how people phrase the search. Some of the audience it was rewritten for was probably software.

The rule I now follow, and would give any team: if a page ranks well and gets no clicks, read its query list before touching its title. If the queries are long, templated and comparison-shaped, the fix is not on the page.

04: How to filter them out

In Search Console’s Performance report, add a filter on Query, choose Custom (regex), set it to Doesn’t match regex, and paste:

provide a definitive answer|pros and cons specific to|evaluate the .+ products company|-site:

It is deliberately conservative. It removes the unmistakable ones and leaves borderline long questions in, because some of those are real people and throwing them away would hide genuine demand. Flip the filter to Matches regex to see what it catches on your own site; the list is often more interesting than the rest of the report.

05: They probably inflate AI citations too

This is a follow-up to my earlier study of 21,700 Copilot citations, which covered April to July and found about 91% of this site’s citations on one old essay about the 10,000-hour rule. Bing’s AI Performance report for the following three months, read October 2, 2026, shows about 8,100 citations with the same concentration, slightly diluted:

Bing Webmaster Tools, AI Performance (beta), Pages tab, three-month view. Bing describes these figures as a sample of overall activity. Top five of 18 cited pages.

The essay now takes about 85% instead of 91%, and the comparison page has become the second most cited page with 903. Here is the connection to the rest of this post: the grounding queries Bing lists for that comparison page include phrasings like “Profound vs Share of Model automated content workflows comparison” and “Profound vs Share of Model real-time misinformation detection”, in long runs of near-identical variations. That is the same machine-generated shape as the Search Console prompts.

No dashboard separates tracker-driven citations from human-driven ones, so I cannot say how many of those 903 came from software. But the reasonable reading is that for any page comparing AI visibility tools, part of its “AI citations” are the tools’ own trackers asking about themselves and their competitors. That is worth knowing before anyone sells you a citation count as proof of reach. Google’s separate Generative AI features report, for the same window, shows the same two pages on top: the essay with 1,320 of 2,930 AI-feature impressions, and the comparison page with 814.

06: What I changed

Not the title. The comparison page’s contentwas the problem worth fixing: whoever is asking, AI answers quote it, so anything wrong on it gets repeated with this site’s name attached. Re-reading the three vendors’ pricing pages on October 2, 2026 found that Profound had removed the self-serve plans the page was built around and Peec had added a tier, so it has been rewritten against what the vendors say today.

Measuring the business pages themselves, as distinct from what a dashboard reports about them, needs repeated runs of real prompts, for the reasons in how to judge AI visibility tool accuracy and on the methodology page.

07: Frequently asked questions

Hami Tahm is an AI visibility consultant based in Toronto.