Prompt Set Design

Layer it by buying stage: awareness (what is this), interest (how does it work), decision (how do I choose, versus whom), purchase (pricing and rollout). Testing only decision questions overstates visibility; testing only awareness questions hides commercial value.

One set per language. Chinese and English questions differ in phrasing and in which sources get cited, so machine-translating the English set does not produce a usable Chinese one.

The set does not need to be large, it needs to be frozen. Thirty-five to forty questions is a common size that keeps the denominator clean and comparable month over month.

Definitional and mechanism questions. In testing these produce the lowest brand hit rate, so including them permanently depresses mention rate — and their value is in being cited, not in naming brands. Keep them as content topics, out of the monitoring denominator.

Questions that name a brand in the prompt. Asking 'A or B, which should I choose' steers the answer toward those two brands and measures nothing; naming a competitor simply hands them the mention.

Log every change: which questions, when, and what the previous definition was. Without that record no later trend can be explained.

Match re-measurement cadence to how fast the data moves. Weekly is enough to see mention rate shift; daily produces sampling noise.