A practical AI visibility audit: how to check what AI says about you

An AI visibility audit means systematically asking ChatGPT, Claude, Gemini and Perplexity the same questions your customers would ask, and noting whether you are mentioned, how you are described, and who gets recommended instead. Do this in a structured way rather than randomly, and you will quickly see where the weaknesses lie. Below is the step by step approach.

Step 1: Build a list of realistic questions

Most businesses make this mistake first: they ask "who is the best [industry] in Norway" and are disappointed when they are not mentioned. That is rarely how real customers ask. Instead, write down ten to fifteen questions the way an actual customer would phrase them, ideally with location, budget or need specified. "Best accountant for sole proprietorships in Kristiansund" produces a completely different and more realistic answer than "best accountant in Norway".

Step 2: Test each question in all four assistants

Run the same question through ChatGPT, Claude, Gemini and Perplexity, and note the answer word for word each time. The models retrieve information differently and therefore often give different answers to the same question. One model may mention you, another may not, and a third may cite incorrect information. This variation is the whole point of testing broadly, not just in the tool you personally use most.

Step 3: Record four things per answer

• Whether you are mentioned at all, and if so, where in the answer (first, in the middle, only as a footnote).
• Whether the description of your services and your name is correct.
• Who gets mentioned instead of you, and how many times the same competitors show up.
• Whether the model cites any source for its claims, and whether that source matches what you actually publish.

Step 4: Look for patterns, not single answers

A single answer where you are not mentioned is not necessarily a problem, it could just be random variation in how the model generates text. A pattern where you are never mentioned across ten relevant questions, while one particular competitor keeps showing up, is a real signal that something structural is missing. Look for patterns in the theme, not single outcomes.

Step 5: Connect the findings to concrete actions

An audit without action is wasted time. If you are missing structured data, check the schema generator to get the right markup in place quickly. If you have a lot of content a model should be finding more easily, look into an llms.txt. If the problem is that no one but you talks about the business, third-party mentions are likely what is missing most, see third-party mentions in AI answers for how to build these up.

Step 6: Repeat the audit regularly

Answers from an AI model are not static. They change when the model is updated, when your competitors publish new content, and when your own visibility shifts over time. An audit done in January says little about how things look in June. If you want to avoid doing this process manually every month, ongoing monitoring is what actually alerts you when something changes, instead of relying on remembering to check yourself.