What actually decides whether AI names your brand

When someone asks an assistant which tool to buy, they get two or three names and a short justification. There is no second page. Either you are in that answer or you are invisible, and you will never see the deal you lost because it never became a lead.

The obvious response is to write more content about yourself. The citation data suggests that is close to the least effective thing you can do.

Assistants cite other people talking about you

When we look at which domains get cited in answers about a software category, a consistent pattern appears. Review platforms, community threads, and analyst sites dominate. Vendor websites appear, but well down the list, and usually only for the vendor being asked about by name.

Asked to recommend, a model leans on sources that read as independent assessment rather than self-description. Your own product page is not evidence that your product is good. A comparison thread where three people argue about it is.

The engines disagree with each other, sharply

Treating “AI visibility” as one number hides the most useful thing in the data. The same brand, measured in the same week, can be strong on one assistant and weak on another because each has different training data, retrieval behaviour, and access to the live web.

This turns a vague goal into a specific one. “Improve our AI visibility” is not a project. “We are third on ChatGPT and level with the market leader on Perplexity” gives you somewhere to look.

Ask the same question repeatedly or you are measuring noise

Assistant answers vary between runs. Ask the same question twice and you can get different brands named. Any measurement based on asking once is a coin flip presented as a finding.

You might see your brand once and conclude you are visible. Ask the same question ten times and you may be named three times. Both observations are real; only one is a measurement.

What to actually do

• Find the questions you lose. Questions where a competitor is named and you are not are the shortest path to a useful intervention.
• Look at what those answers cite. The same few domains often decide a category.
• Find a lever that moves several questions. One source cited across five answers is more valuable than five disconnected pieces of content.
• Record the action and check. Note the date, then inspect what changed in the connected questions.

What nobody can honestly promise you

Anyone claiming to prove that AI visibility drove revenue is overselling. AI-influenced buyers usually arrive as direct traffic or branded search, so the original touchpoint is missing.

What is honest: you can measure whether visibility moved, in which questions, on which engines, after a specific action. That claim is weaker than attribution and considerably more useful than a number with no context.