Who am I competing with in AI answers, and how does GEO MarketMap find out?

From the answers, not from a directory. Step 4 of the setup asks 3 assistants your tracked questions; every brand those answers name, other than you, is a candidate. Each candidate arrives with its evidence: the number of answers it was named in out of the answers collected, the questions, and the position when the answer ranked brands. A candidate named once in 60 answers is shown with that count, so you can drop it with reason.

A rival you add by hand is recorded as yours, and a rival you drop is remembered as dropped, so it is never proposed again. Each kept rival needs a site, because the measurement reads its pages.

1. The rival's own rows for the category's queries are bought once, up to 300 per rival, and never bought twice.
2. Every question is judged against your project: relevant, adjacent, or off-topic. Rules run first, then anchors, then a judge on the middle band.
3. The niche rows are written: for each question, whether you are on the answer, who is, and what the answer rests on.
4. The market map is redrawn with the rival on it. The niche card fills while this runs, 1 rival at a time, with a line saying which rival is being read.

A rival the Google report of step 2 already measured costs nothing to keep. The rates for the rest are on the pricing page.

1 map of the category: your project, the rivals you kept, every question the niche asks, and who is on the answer for each. Reading down a question shows who is named; reading across a brand shows which questions it holds. The gaps, questions where a rival is named and you are not, are the queue's first lane.

A visibility score tells you that you are losing. It does not tell you which question you are losing, who is being named instead, what would have to exist for that to change, or who writes it.

The market map feature page shows the shape of it with placeholder figures.