问W
Before you ship a Copilot-drafted survey, test the wording and keep the instrument short. Read each item aloud to someone who did not write it. If they have to reinterpret it, rewrite it. Fluent AI copy is not the same as a validated question. One decision per survey: if you cannot name what the team will change when the score moves, do not field that item. In context, one rating plus an optional why is enough. A 5–8 item pulse is plenty for a trend. Matrix batteries mostly add fatigue. Pilot the exact wording with about five people and drop any item where two people hear a different question.