When using Browse AI: never trust an export without spot-checking it against the source page, tasks marked successful can still have missing or wrong values. If a field looks off, re-check the robot's field selection rather than assuming the site changed. And do the robot-building on a desktop, the mobile signup works but custom robot creation is desktop-only.

Context: Web (Medium, ToolScope review, Sep 2026): a real hands-on test found data-quality gotchas. All four tasks were marked successful, but the exports still had missing values: Countries came out as "0" instead of "Anywhere in the World", and a "Top 100" tag was missing. The reviewer had to compare the CSV against the source page before trusting it. Also: custom robot creation is desktop-only, signing up on a phone works but building the robot required switching to a computer.