Choose questions before inspecting favorable answers
Choose questions that reflect actual buyer decisions and keep the scope documented. A prompt set designed only to produce favorable answers provides a weak basis for marketing decisions.
A prompt set is a defined sample of buyer questions. Selecting only questions that already produce a brand mention biases the result. Include relevant discovery and evaluation tasks, with explicit treatment of brand names.
A worked example
A hypothetical set contains ten questions asking why ZeroDark is useful and ten unbranded questions about reporting options. The first group tests brand-aware explanation; the second tests discovery. Combining them without disclosure exaggerates unprompted visibility.
How to apply it
- Define audience, market, language and buying tasks before collecting answers.
- Separate branded, unbranded and competitor-comparison questions.
- Fix the initial set and version additions rather than silently replacing weak results.
What the result can establish
Record answer availability and failures instead of selecting only completed favorable runs. A prompt-set mention rate describes that sample and protocol, not a share of all AI users or demand. Avoid treating estimated keyword volume as known prompt volume.
Reference: Google: AI features and websites.
Supporting documentation
Google: AI features and websitesNIST: AI Risk Management FrameworkGoogle: AI features and your website