How many answers are enough for a comparison?

“We measure a thousand prompts” sounds thorough. It does not tell you whether the results answer a useful business question. A thousand near-identical formulations may reveal less than a smaller selection covering different purchasing decisions.

Coverage and repetition do different jobs

Adding questions broadens the subject being examined. Repeating a question shows how answers vary under similar conditions. Neither replaces the other.

For illustration, a brand mentioned in six out of ten answers has an observed mention rate of 60 per cent. That describes those ten answers. It is neither a fixed rate for future responses nor a finding about every question in the industry.

Record the measurement setup

• Which questions were examined, and how were they chosen?
• Which providers and models were involved?
• When and how often was each question asked?
• Which answers were missing or could not be analysed?

A larger panel does not repair a biased selection. If every question already contains the brand name, the results say little about presence in open provider questions.

Record panel changes when comparing periods too. A higher rate could simply reflect the removal of difficult questions.

There is no universal minimum sample size to take from this example. It depends on the variability of answers and the decision the measurement needs to support. A transparent setup with visible limitations is more useful than an impressive headline count.