Monitoring AI visibility without a good question set is like measuring SEO with a single keyword. The set is 80% of the outcome.
Basic rule: customer prompts, not marketing prompts
Do not ask "is brand X any good?". Ask what someone who does not know you yet would ask. If your name is in the prompt, the answer is contaminated.
Templates by intent
Discovery (5-7 prompts)
- "Best [category] tools for [company type] in 2026"
- "What software should I use for [specific problem]?"
- "Alternatives to [market leader]"
Comparison (4-6 prompts)
- "[Competitor A] vs [Competitor B]: which one for [use case]?"
- "[Category] tools with [differentiating feature]"
Price and decision (3-4 prompts)
- "How much does [service category] cost per month?"
- "[Category] options under $X per month"
Trust and objections (3-4 prompts)
- "Is it safe to use [category] with customer data?"
- "Common problems with [category]"
Local or vertical (2-3 prompts)
- "Best [category] in [city or country]"
- "[Category] for the [industry] sector"
How many and how often
Fifteen to twenty-five prompts cover a category well. Fewer than ten gives noisy readings; more than thirty dilutes focus and raises cost. Repeat them weekly on the same day: models vary, and only a time series separates a trend from noise.
What to record on every run
- Whether you are mentioned (yes/no).
- Your position in the list.
- Which competitors appear and how often.
- Which sources the answer cites.
The most expensive mistake
Changing the prompt set every month. Change the questions and you lose comparability and start over. Freeze the set, measure it for a quarter, and only then extend it by adding prompts without removing the old ones.