Choose useful questions
Start with questions about discovery, services, location, comparison, and audience. Write them as a real customer would, and review them before running them. A list of keyword variations is less useful than a small set of questions with a clear purpose.
It sounds like something a customer would ask, names a real need, and gives you a useful answer to read.
Record more than the mention
- The exact prompt, date, and time.
- The ChatGPT model or mode, when that information is available.
- Whether web search was used.
- The full response and source links, not just an excerpt.
- Whether the business was mentioned or its website was cited.
- Whether the answer explicitly ordered its recommendations.
- Which competitors and external sources appeared.
This gives you something to review when the answer changes. Without the prompt and response, a mention is too easy to misread.
Interpret variation honestly
ChatGPT output can vary with the model, prompt wording, time, and whether web search is enabled. Compare the same prompt under the same setup where possible. A change in one answer does not prove that an underlying ranking system changed, so read the original response before drawing a conclusion.
See how those details fit together in the example report or read the evidence model.