How to Run a ChatGPT Brand-Mention Test
No public checklist can guarantee that ChatGPT will mention or recommend a brand. A useful project combines a documented response sample with improvements that serve readers independently of assistant inclusion.
Step 1: select neutral questions
Use customer interviews, sales records, support themes, and site search to identify questions. Include category, comparison, constraint, and factual prompts. Avoid only asking branded questions, which do not test discovery.
Step 2: record collection conditions
- Save exact prompt and complete response.
- Record date, account state, region, model label, and enabled tools shown in the interface.
- Preserve any displayed source links.
- Do not assume another user will receive the same output.
Step 3: inspect relevant public facts
Check whether owned pages clearly state accurate products, audiences, limits, locations, policies, and dates. Correct contradictions. Where a response cites a source, inspect it directly rather than inferring how the system selected it.
Step 4: make and review bounded changes
Improve one defined content set, log releases, and repeat the frozen sample on a planned date. Report all runs and contextual changes. Treat differences as observations, not proof that one edit controlled the response.
Limits and next step
A mention is not necessarily an endorsement, and absence in a sample does not make a business invisible. Evaluate accuracy, relevance, linked evidence, and attributable business activity separately.