User story As a partner using the Snapshot for sales, I want the AI section to ask realistic questions and score real AI readiness, so that the report is credible. Problem / context Generated questions are often absurd ("digital nomad with dog"). A site with llms.txt, JSON-LD everywhere, FAQ, NLWeb/WebMCP endpoints and PageSpeed 90/100/100/100 gets a C, mainly for missing social links. Acceptance criteria Questions are suggested but editable before the run. Criteria include llms.txt, structured data, FAQ, machine endpoints. Grade weighting documented; social links do not dominate.