Guide · Chapter 6 of 6
Measure your AI visibility, don't guess it
Why one search in ChatGPT tells you almost nothing, and how to measure whether AI assistants recommend your business: repeated, dated and by engine.
Published 3 min readBy Visibility
Short answer
AI answers vary from one run to the next, between assistants, and between an assistant's own knowledge and its web search, so a single search tells you very little. To know whether you are recommended, ask the same fixed set of questions across several assistants, repeatedly, and record every answer with its date.
On this page
Why can't I just ask ChatGPT myself?
You can, and it is a good start. But one answer is a sample of one. Ask again and the list may change; ask a different assistant and it will change more. Your chat history and location can also shape what you see. See why AI assistants give different answers each time.
What should I measure?
| Measure | What it tells you |
|---|---|
| Mention rate | How often your business is named in answers to the questions that matter to you |
| Citation rate | How often an assistant links to your own website as a source |
| Position | Where you appear when you are named, first or fifth |
| By assistant | Whether you do well in one assistant and are missing from another |
| Sources cited | Which other sites assistants rely on for your sector |
Being named and being cited are different facts. You can be named because a directory mentions you without your own site ever being cited, or the other way round. Keep them apart. Our methodology sets out exactly how we define each.
How do I measure it properly?
- 1Fix the questions. Write them the way customers ask them, and keep them the same from one month to the next so results are comparable.
- 2Ask several assistants. At least ChatGPT, Gemini, Perplexity and Google's AI answers, since customers use all of them.
- 3Separate knowledge from search where an assistant offers both.
- 4Record every answer with its date and model. AI answers can shift with a model update alone, so a figure without a date cannot be compared with anything.
- 5Compare against competitors asked the same questions at the same time.
Visibility does this for Gibraltar's sectors every month and publishes the questions and results in the directory. The free visibility check runs a few questions live; paid plans track your own questions over time.
How should I read the results?
Look for patterns rather than single answers. Missing from one assistant but named by the others points at something that assistant cannot read or trust. Named but never cited suggests other sites are doing the work your own site should. Named only in web-search answers suggests the model's trained knowledge has not caught up with you yet.
A low position is a finding, not a verdict on your business. Assistants miss good businesses all the time; the point of measuring is to find out why.
Common questions
How often should I measure?
Monthly is enough for most businesses to see real movement without chasing noise. Daily tracking helps when you are testing specific changes or in a fast-moving sector.
Is there an AI visibility score?
Some tools combine everything into one score. We do not publish one, because a single number hides whether you were named, cited, or both, and in which assistant. We report each measure separately.
Why do my results differ from what I see in ChatGPT?
Your own chats are shaped by your history, settings and location, and each answer varies anyway. A measurement uses fixed questions with no personal history, repeated, so it describes the typical answer rather than one conversation.