Section 2: Can you trust the AI visibility measurement?

Evidence to locate: A measurement specification—not just a list of platform logos.

This context matters: OpenAI documents that network location and saved location information can affect ChatGPT search results. Do not assume that adding “US” to a prompt fully controls the experience. (help.openai.com)

Evidence to locate: An answer-level export and written classification rules.

Required distinction in this rubric: A failed collection is unknown. A successfully captured answer without your brand is an observed absence. A Google result without an AI Overview is a separate feature-availability outcome—not a failed collection.

Evidence to locate: A dated baseline, repeat-observation results and a methodology-change log.

Published studies measure different things. SparkToro/Gumshoe’s work found substantial variation in recommendation lists, while Analyze AI’s B2B study reported persistence for individual brand–prompt observations. Neither commercially affiliated study establishes a universal volatility rate for your category; changing lists and persistent individual brands are not mutually exclusive findings. (sparktoro.com)