How I find the questions worth tracking
You don't have to curate a list of questions for me to track — I generate it. On a schedule, I read what your site offers, the questions your visitors actually asked, and the rivals AI cites instead of you, then write the category questions a potential customer would ask an assistant about a business like yours — phrased the way someone who hasn't found you yet would ask, so they never name you. That set is what I probe the answer engines with.
Each discovery pass feeds three inputs to a language model and asks for the questions a buyer would put to an AI assistant. This is pure text generation — the same model that powers your chat, with no web search. The proposals are then de-duplicated and checked against your own choices before anything lands in the tracked set.
Page text comes from crawling; chat themes from demand; the finished set is consumed by citation.
I tag each proposed question by where it came from, and show that tag under the question — so you can see whether it reflects your own pages, real chat demand, or a place a rival wins. The tag is provenance, not priority: a not-cited question's opportunity is set by what the engines actually do — high when a rival is cited for it (or it mirrors chat demand), med when no one is cited yet.
Discovery never clobbers your decisions. A pass keeps every question you've pinned or muted and replaces only the rest with fresh proposals, and it will not re-create a question that matches one you already kept.
A pass generates up to a target of 30 questions — the lever on probe cost, since every tracked question costs engine calls — and re-runs about monthly, because the question set changes slowly. I only run discovery for verified, good-standing domains that already have crawled pages.