Back to Blog

Measuring ChatGPT visibility: a repeatable protocol

A repeatable ChatGPT audit protocol: questions, search mode, language, sources and errors, with a completed example.

Editorial review:

Reviewed factual claims and sources; added distinct examples and acceptance criteria. Observations, methodological recommendations and unconfirmed causes are distinguished.

TL;DR: Key Takeaways

Reliable observation needs agreed conditions. A single screenshot and three repeated runs serve different purposes, and both have limits.

  • Test identifier: Allows comparison of the same test.
  • Question and context: Keep the exact wording in the internal record.
  • Fictional finding: Mention yes, owned citation no.

Ask precisely: for which buying questions does a statement about our company appear, and what visible source supports it? A general “Are we in ChatGPT?” mixes brand knowledge, search and different user contexts.

Context and scope

Define questions about service, location, exclusions and comparisons. Record the surface, visible model label, search mode, language, date and relevant account or memory context. A test using web search must be distinguished from an answer without visible search retrieval.

Classify the answer: correctly mentioned, misidentified, not mentioned or not answered. Source URLs need a separate column. A correctly named company without an owned source is not automatically a citation of its website.

One conversation is not a reliable baseline. Three repetitions on separate days can be a practical starting point; this is a chosen sample, not a statistical minimum. Sample size and uncertainty should reflect the business decision. Local Swiss markets need relevant location and language questions.

A completed observation record

The following record is entirely fictional and serves as a template. It does not report an actual ChatGPT mention. Save the raw answer as well, so a second reviewer can check the classification. Personal account information should be excluded from a public report.

Review pointExample / subjectInterpretation
Test identifierCH-DE-LOCAL-01 / version 1Allows comparison of the same test.
Question and contextOffice cleaning in Malters; DE; web search; 7 September 2026Keep the exact wording in the internal record.
Fictional findingBrand mentioned; no owned source URL; correct locationMention yes, owned citation no.
Next stepRepeat on two separate daysObserve stability; the cause remains unresolved.

How to apply this

  • Define the question set and categories before the first run.
  • Use fresh conversations with a recorded memory context.
  • Save the answer and sources together.
  • Show all runs and differences for each question.

Sources and further reading

The next step

Next step

The free short check provides an initial assessment of one domain, three buying questions and three observations. An introductory call establishes whether deeper analysis is useful.

Request an introductory call

Frequently Asked Questions

Are three repetitions mandatory?

No. They illustrate a small repeatable starting point. Important decisions may need a larger, better controlled sample.

Does a mention without a URL count as a citation?

No. A mention and a linked source are separate measures.

Why include local questions?

They reflect real decisions in a specific area and language that a generic English industry prompt may miss.