AI search visibility is often presented as if ChatGPT has a normal search results page and every company occupies a stable position.
That is not how the systems behave.
An answer can change because of the platform, model, user location, language, conversation history, search availability and time.
So I treat AI visibility as a sample that needs to be measured repeatedly.
The first step is building a useful prompt set.
The prompts should reflect real customer decisions.
For example:
- Which companies provide a specific service?
- Which providers are suitable for a particular country?
- What alternatives exist to a known competitor?
- Which agency has experience with a particular problem?
Then I separate different outcomes.
A company can be:
- Mentioned.
- Recommended.
- Cited.
- Described accurately.
- Described negatively.
- Omitted entirely.
Those are not the same thing.
I also record sources.
If a platform recommends a competitor, I want to know whether that recommendation came from the competitor's website, a review site, a directory, Reddit, a news article or another source.
Repeated testing matters.
One screenshot is evidence of one answer.
It is not evidence of a stable market position.
For important prompts I prefer several runs across fresh sessions and, where relevant, multiple languages or locations.
Finally I connect the monitoring to business data.
Are AI referrals appearing in analytics?
Are they converting?
Are customers saying they discovered the company through an AI system?
AI visibility is worth measuring.
It is not useful to turn it into another vanity metric.