Start with the question
An answer engine can mention a product, cite a page about it, or recommend it for a particular job. Those are different observations. Build an index that stores each separately so a visible mention does not become an unsupported endorsement.
A workflow you can check
Use the following steps as a small acceptance test. Keep the underlying observations with your notes so someone else can review the conclusion.
- Choose a fixed set of buyer questions and record why each belongs in the sample.
- Record engine, question, location, retrieval time, answer text and citation URLs.
- Label recommendation, neutral mention and citation independently; keep ambiguous cases for review.
- Report counts alongside the sample size and preserve unanswered queries.
Interpret the evidence
A useful index has a transparent denominator. Compare equivalent questions across runs and describe changes as observations within your sample. Do not call a small set of prompts a market share study, or infer which provider is best from how frequently it appears.
A prompt to adapt
Replace the inputs with your own approved examples. Ask the assistant to explain its scope before starting research.
Create a recommendation evidence sheet for these buyer questions. Keep mentions, recommendations and citations separate, include source URLs and flag missing evidence.