Prompt

Prompt wording affects which evidence an assistant retrieves and which businesses it names. "Best roofer in Phoenix" and "trustworthy roofer for an insurance claim in Phoenix" can produce different results. Publish the prompts so readers can compare and repeat the measurement.

Prompt design has some of the same risks as survey design. A leading prompt can steer the answer, a prompt naming one company encourages the assistant to discuss it, and industry language may retrieve different content than a homeowner's wording. Such results record what the assistant said under those conditions, but may not represent what buyers encounter.

Write prompts in language customers use. For local services, that often means describing a situation instead of entering a keyword fragment. Include a deliberate mix of commercial, urgent, price-related, and credential-related questions. Results can vary by phrasing, so the set should represent several common needs.

Context beyond the question also shapes the answer. Web search settings, supplied location, and businesses named earlier in the conversation can change the result. Hold these conditions fixed across a measurement run and publish them with the prompts.

Version the prompt set when wording changes. Keep older results tied to the version that produced them, and avoid comparing two periods as a trend when the questions or run conditions changed between them.

Routeless publishes six buyer questions per trade, including direct commercial and situational wording, on the methodology page. Public prompts let readers repeat the process and assess whether another sample produces a similar pattern.