Your Search Console is full of machines.
Google AI Mode logs every turn of a chat as an ordinary query. We read 16 months of yours, tell you which rows were machines, and name who is running them.
Free · read-only · about 3 minutes
Before you connect: Google grants openid, email, and https://www.googleapis.com/auth/webmasters.readonly. We send a selected set of query rows to DeepSeek to write the report. Tokens, email, and reports expire after 30 days. You can disconnect sooner at any time.
See a finished report- nike running shoes12,904842
12,904 impressions · 842 clicks
- yes1500
150 impressions · 0 clicks
- act as a running coach and pick…4100
410 impressions · 0 clicks
- best {running shoes} for {flat feet}22,8470
22,847 impressions · 0 clicks
- ok show me cheaper options960
96 impressions · 0 clicks
- nike air max 908,221611
8,221 impressions · 611 clicks
The report
The report
Every machine watching your market, named.
nike.com · AI query report
Running shoe comparison prompts
Confidentbest {running shoes | sneakers} for {marathon training | flat feet} under {price}
38 template variants · 22,847 impressions · running 97 days · 0 clicks
Nike is named in the rankings this machine asks for.
Free · read-only · about 3 minutes
What the scan finds
What the scan finds
Three things your query report has been hiding.
Who is monitoring you
The prompt templates running against your market every day, who is behind them, the languages they run in, and whether you are ever in the lists they ask for.
“best ai visibility tools”
141 separate days · 0 clicks
What real buyers ask AI
The genuine human questions in your category, separated from the machine probes that imitate them. Taken from your own data, not a keyword tool.
“how can i tell if im being cited in gemini?”
982 impressions · position 4.1
Conversations you were pulled into
Replies and follow-ups that leaked out of real AI chats where your site was cited mid-answer, tied back together by the pages they landed on.
“ok show me cheaper options”
seen on 96 days · 0 clicks
How it works
How it works
Five steps, about three minutes.
- 01
Read
Your whole Search Console history, 25,000 rows a page.
- 02
Separate
Every row against 40+ AI conversation patterns, in five languages.
- 03
Follow
Daily cadence, cited pages and countries for every flagged query.
- 04
Build
The strongest evidence, curated down for the analyst.
- 05
Write
The slow, good part. A reasoning model reads the whole file.
What it flags as it reads
- forced ranking"…debes proporcionar un ranking forzado…"
- persona wrapper"as a marketing manager at a b2b company…"
- tracker probe"respond with \"…is working correctly\""
- reply fragment"yes" · 150 impressions across 96 days
Keep watching
Keep watching
One scan is a snapshot. The machines keep running.
Track one brand across all eight AI models every day: 50 buyer-intent prompts, a daily visibility score, alerts, competitors, citations and sentiment.
14-day trial. Agencies and multi-brand teams from $500/mo.
FAQ
FAQ
Before you connect.
Google AI Mode fires real searches for each turn of a conversation, and Search Console records those turns as ordinary queries whenever your site is surfaced or cited. That is why query reports now contain things like "yes please", long persona-style prompts, and instruction text such as "you must provide a ranking". The AI Query Scanner reads 16 months of your queries and separates those machine turns from real human searches.
The scanner flags conversational replies ("yes", "ok show me cheaper options"), instruction wrappers ("only respond with one of the options"), persona preambles ("as a marketing manager at a b2b company..."), tracker probes, pasted prompt fragments, and templated comparison prompts that repeat for weeks. Each flagged row keeps its impressions, clicks, active days, cited pages, and countries as evidence.
Often, with honest limits. When one brand appears in nearly every variant of a recurring prompt template, the pattern is consistent with a monitoring tool run by or for that brand, and the report says exactly that, with the real queries as evidence and a confidence label. It is inference from public data, not surveillance, and the report never claims proof it does not have.
Three rules the report never breaks. Every quote is verbatim from your own Search Console rows, never paraphrased. Every number is recomputed from those rows rather than repeated back from the model. Every claim about who is behind a program carries a confidence label, and the evidence sits underneath it so you can disagree with us.
The scan is free. Google asks for three exact OAuth scopes: openid, email, and https://www.googleapis.com/auth/webmasters.readonly. The first two verify the Google address that owns and receives the report. The third can only read Search Console data. You pick one verified property, and nothing in the Google account can be changed.
The Search Console scope is read-only. It cannot change settings, submit sitemaps, or read another Google product. We send a selected evidence set of query text and metrics to DeepSeek to write the analyst sections. Google tokens, your email, and the private report expire after 30 days. The report URL alone grants no access. Disconnect in the property picker to delete your private reports and scrub the grant sooner. Your Google address is used for report ownership and delivery, not automatic marketing enrollment.
Because nobody is on the other end. A monitoring script asks an AI for a ranking, the AI searches, your site is shown, and Search Console counts an impression. There is no human to click. That gap, thousands of impressions with zero clicks across months, is one of the clearest machine fingerprints the scanner uses.
The scanner looks backwards: it reads what already happened in your Search Console and shows who has been asking AI about your market. Trakkr looks forwards: it asks those questions across ChatGPT, Perplexity, Gemini and Claude every day and tracks whether you are in the answer. The scan ends with the exact prompt list worth tracking.
More free tools