How to Investigate Brand Mentions in AI Assistant Responses
Outside observers cannot infer a universal brand-selection formula from a few responses. Platform configurations, tools, prompts, and outputs may vary, and undocumented internal weighting should not be presented as fact.
Ask an answerable question
Instead of asking why all assistants choose brands, ask what appeared for a defined prompt set under recorded conditions. Decide whether you are studying mentions, ordering, source links, factual accuracy, or wording; each needs a coding rule.
Collect enough context
- Exact prompt and full response.
- Platform and displayed model or mode.
- Date, region, language, account state, and enabled tools.
- Displayed links or citations.
- Repeated runs if variation is part of the research question.
Evaluate possible explanations cautiously
Compare response statements with public sources and owned pages. You may identify conflicts or candidate sources, but do not claim a page, backlink, schema field, review, or crawler visit caused the response unless the evidence establishes that connection.
Publish the limits with the findings
- Show the denominator and missing results.
- Describe prompt selection and coding.
- Separate observations from hypotheses.
- Avoid extrapolating to market share, all users, or future behavior.
- Retain data so another reviewer can inspect the conclusion.
Limits and next step
This method can document responses and factual problems. It cannot reveal proprietary internals or provide a strategy that ensures recommendations.