In practice
Panels are built from how buyers actually speak, not from keyword lists reworded as questions. Sales call recordings, support tickets and customer interviews are the raw material, because the phrasing people use with an assistant differs substantially from what they type into a search box.
Running cold is the step most often skipped and the one that most affects validity. Personalisation, memory and conversation history all bias results toward whatever the tester has looked at recently — frequently including the client's own site. Sessions must be clean, and preferably automated.
The panel is an instrument, not a target list. Fixing it before work starts and changing it rarely is what makes any subsequent improvement meaningful, because a panel that moves alongside the work measures nothing.
Not to be confused with
- Keyword list
- Keyword lists have volume data and can be revised freely. Prompt panels have no volume data and must stay fixed, because you define the test.
Questions
Where do the prompts come from?
From how your buyers actually speak — sales calls, support tickets, customer interviews — rather than keyword lists reworded as questions. The phrasing difference is substantial and it changes the results.
Why does running cold matter?
Because personalisation and conversation history bias answers toward what the tester has looked at recently, often the client's own site. A signed-in session measures the tester rather than the market.
Is there volume data for prompts?
No reliable public source exists. Anyone presenting prompt volume figures is modelling or estimating, usually with more confidence than the method supports.