Built into Vercatus
The system inside Vercatus that learns every day. It reads every verdict, every page change and every answer, and turns them into what no single customer could know: which plays win, how much demand sits behind a topic, and what just changed.
Each is learned from many sites at once, kept as counts and effects, and shown only above a privacy floor. You meet it inside your topics and plays, marked with its badge.
Every frozen verdict sets what to expect from the next play of its kind, and each play shows the record of plays like it.
The AI Demand Index estimates how many people ask AI about a topic, as a range with a grade, sharper as connected sites calibrate it.
When a page engines cite changes and the answers move, Vercatus measures it like a verdict, on your topics and on competitors' pages.
Once for each site and once across all of them. Nobody tunes it: every shipped play, page change and verdict makes the next recommendation better informed.
No engine publishes how often it is asked. The index estimates AI conversations a month per topic and market from sourced signals, always as a range with a grade, and calibrates itself against first-party AI data from connected sites. A topic with too little evidence gets no number at all: not estimated beats a guess.
On our own site, 2 October 2026: 9 topics estimated, grade C; 11 not estimated yet.
21 English questions in 3 markets, asked of 6 engines with search on and off, and every page the answers cited read. Run on 1 October 2026.
100 vs 26
So the poll asks many questions, few times each.
For the same 100 asks, 100 different questions were worth 100 independent answers; 20 questions asked 5 times were worth 26. Two asks of one question shared 68% of their brands. Method
30 to 58%
So Why has a Memory lens beside Sources and Search.
That is how much an engine's brands with search off overlap with its brands with search on. ChatGPT named 63 brands from memory alone and 82 only when it searched. Method
3×
So off-site plays aim at the sources that feed the answers.
A brand none of the cited pages name is named in 11.9% of answers; six or more, in 34.9%. The most-read pages are listicles. Method
66% vs 31%
So Why has a Search lens. Perplexity and Le Chat expose no searches.
Of the brands an answer recommends, 66.3% rank in the top ten for one of its searches, against 30.7% of those it left out. "best" is in 45%, "2026" is in 30% of the 299 searches exposed. Method
4%
So ChatGPT is read through its app.
On the same 10 questions, ChatGPT's app and its API shared 30% of the brands they named and 4% of the sites they cited. Method
3.7%
So Engine Briefs are shown as perception, and tested.
Asked what makes a brand their first choice, engines name the product (62%) and price (15%), and almost never reviews (3.7%) or third-party coverage (2.2%). The source data says the opposite. Method
33 of 64
The 64-site study. 33 of 64 well-known sites serve every AI crawler tested on their home page; 17 publish an llms.txt. Measured on 18 September 2026, every observation published.
Read the studyA line is published here only in aggregate, once enough sites and verdicts stand behind it. Today the library holds 187 graded practices. The knowledge base
It holds counts and effects, never a workspace, a site or a question.
Nothing is shown, and nothing sets a play's gain, until at least three sites stand behind it.
Analytics, Search Console data and crawler logs stay in your workspace; only ratios are pooled.
Run from scripts against the engines only, with no customer data, on a budget the founder set.
Create your account and add your domain. Every topic and every play starts from what it already knows.
Cancel any time · Data stored in the EU
Vercatus