Build the sample from customer questions
Collect the questions buyers ask while learning about a category, comparing solutions and choosing a supplier. Group them by intent and keep a fixed version instead of changing the questions each time. The number of prompts is not the objective; covering the actual decision process is what matters.
Record answers and citation evidence
Keep the platform, query time, exact question, whether the brand appeared and the cited links. Context and system changes affect AI responses, so a single answer cannot demonstrate long-term performance. Store examples, simulated content and real observations separately.
Retest under the same conditions
First correct website content that is inaccurate, unclear or missing a source. Then repeat the original question set. Compare information accuracy, verifiable citations and intent coverage. A correlation does not establish causation; interpret changes alongside publication records and other channel results.
Before you start
Keep a fixed question set, record the platform and time, retain the original answers, preserve traceable citations, and use consistent retest conditions.
Using this guide
This guide is a reference for working methods. Apply it according to your business situation, platform rules and the actual scope of authorisation.