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.