Your Search Console has impressions no person made
Two kinds of robot inflate your impressions: the ones that track rankings and the ones that measure AI visibility. Neither clicks. We checked it on our own two websites.
Search Console counts as impressions searches no person made. On joseredondo.es, between 1 July and 20 September 2026, 28.5% of query impressions came from rows with no clicks below position 50 or shaped like a GEO tool prompt. Without them, the average position goes from 28.9 to 22.6. The most reliable signal is clicks: a robot does not click.
- Sample
- Two websites of our own, Search Console web search export: joseredondo.es (544 queries, 10,954 impressions) and vilnux.com (27 pages, 15 queries)
- Period
- joseredondo.es: 1 July to 20 September 2026 · vilnux.com: 6 to 20 September 2026
- Method
- Any row with no clicks that either only shows up below position 50 or looks like a tool prompt (a persona such as "as a director of…", an instruction to the model, or 20 words or more) is flagged as a likely crawler. The average position, weighted by impressions, is then recalculated with and without those rows.
- Language
- Spanish and English
Where it comes from
On 22 September, Álvaro Peña de Luna published that on his website, mencoro.com, 86.5% of 90 days of impressions had been made by third-party rank trackers: 84,401 impressions, 136 clicks, and an average position of 36.6 that, filtered, was 18.2. His post (in Spanish) explains how he spotted it. Those figures are his, not our measurement.
Our engine reads Search Console in four places: to suggest questions to measure in AI, for quick wins, to cross-check the sitemap and for the classic search block. So we exported Search Console for our two websites and went through it row by row.
Two kinds of robot, not one
The first is the one Álvaro described: tools that track rankings. They ask for the top hundred results, so they leave impressions at positions no person sees, and they never click. On joseredondo.es that is 103 queries and 1,336 impressions below position 50 with no clicks.
The second we did not expect: AI visibility tools running their prompt batteries against Google AI Mode, which Search Console counts as web search. They are easy to recognise: "as the director of a real estate agency… what do G2 users say about…?", "as the manager of a medical clinic…", instructions such as "you must give a forced ranking". On joseredondo.es there are 160 such queries, 721 impressions and 0 clicks. They rank high — half of them at position 3 or better — so the position filter does not catch them.
Every click in joseredondo.es's query report comes from short searches. Not one query shaped like a tool prompt has a single click.
Is it us? A little, very little. We measure joseredondo.es with our own engine, including in Google AI Mode. Of its 544 queries, only 4 match our prompts (13 impressions). The batteries with personas and G2 reviews are not ours.
What came out on our two websites
| Impressions | Clicks | Suspicious rows | % of impressions | Average position | Without the suspicious rows | |
|---|---|---|---|---|---|---|
| joseredondo.es · queries | 7,225 | 14 | 263 of 544 | 28.5% | 28.9 | 22.6 |
| vilnux.com · pages | 96 | 1 | 5 of 27 | 61.5% | 57.4 | 9.3 |
| vilnux.com · queries | 52 | 0 | 15 of 15 | 100% | 90.2 | — |
Search Console's query report leaves out anonymised queries, which is why its clicks (14) are fewer than the website total (66).
On vilnux.com, a website launched in August, it is almost all noise: all 15 queries in the report sit below position 60 and none has a click, and 72.4% of impressions come from outside Spain. The query with the most impressions, "visibility checking tool", is exactly what visibility tool vendors track — in other words, the competition.
What we did with that data before looking at it this way
On 17 September we reoriented a vilnux.com page towards those queries, because it looked like demand already existed and ranking was what was missing. With this filter, no click backs that demand. The page stays, because it answers a real question in the sector, but the decision was made on data that was probably not human, and we are saying so.
On joseredondo.es, six of the queries our engine suggested as "quick wins" were persona prompts from other tools. They no longer appear.
How to spot it in your Search Console
- Filter by your country. Many of these queries come from other markets.
- Set aside rows below position 50 with no clicks. To see result 51, a person has to get past page five. A rank tracker always asks for the top hundred.
- Look for queries shaped like prompts: a persona ("as the director of…"), an instruction to the model, or a sentence of more than twenty words. If they have no clicks, they are measurement batteries.
- Compare weekends with weekdays over time. On joseredondo.es, weekends had 71% of a weekday's impressions at the start of the period and 119% at the end.
- Decide with clicks. A robot does not click.
What we changed in the engine
From today, Search Console is downloaded filtered to the client's market, with the daily series and the country breakdown, and the report the client uploads is used as well. Suspicious rows no longer count as demand when suggesting questions, for quick wins or when prioritising sitemap pages. And the dashboard warns when part of the impressions looks more like a robot than a person, with the criterion visible and the average position recalculated without them.
What this finding does NOT prove
- It does not prove those impressions come from robots. Search Console does not say who searched: the criteria are the ones least like a person, not proof.
- The vilnux.com report covers fifteen days and 96 page impressions: it shows the effect but is too small to measure its size precisely.
- These are two websites of our own, from one sector and in Spain. The share of robot impressions depends on how many tools track your keywords and your brand, which is higher in marketing and AI visibility niches like ours.
- Some long questions may come from a person using Google AI Mode like a chat. That is why the cut is at 20 words and at sentences with a persona or an instruction, and rows are only flagged if they have no clicks at all.
- It does not cover the Search Console generative AI report, which has no clicks or position: this filter cannot be applied there.
Frequently asked questions
Do rank trackers show up in Google Search Console?
They can. They query Google daily to see where each website ranks, and those searches generate impressions without clicks. On joseredondo.es, 103 queries with 1336 impressions only showed up below position 50 with no clicks.
Why does Search Console show queries like "as the director of… what do G2 users say about…"?
Because AI visibility tools run prompt batteries against Google AI Mode, and Search Console counts them as searches. On joseredondo.es there are 160 such queries, with 721 impressions and no clicks.
Why does my Search Console average position not add up?
Because it mixes impressions no person saw: rank trackers, far down the page, and tool prompts, sometimes near the top. Remove rows with no clicks below position 50 and rows shaped like prompts, and recalculate.
How do I filter robot traffic in Search Console?
Search Console does not let you exclude it directly. The practical route is to filter by your country, set aside rows with no clicks below position 50 or shaped like prompts, check whether weekends stop dropping, and decide with clicks rather than impressions.
Does this affect AI visibility measurement?
Not the measurement itself, which asks AI search engines directly. It does affect decisions made with Search Console, such as which questions to measure or which pages to prioritise; that is why we filter those rows before using them.