Most companies have never checked what AI assistants say about them. Not once. They will happily pay for rank tracking on searches nobody performs any more, while the answers actually shaping their buyers' shortlists go completely unexamined.
Here is how to check, properly, in about half an hour, with free tools.
Step one: write the questions your buyers ask
Not keywords. Questions, phrased the way a person would type them into an assistant. Get fifteen to twenty of them. The best sources are your own sales calls, your support inbox and the first email a new enquiry sends you.
Aim for a spread across three types:
- Category questions. Who is the best provider of X in Y, what are the top options for Z.
- Problem questions. How do I solve the thing your product solves, without naming your category at all.
- Brand questions. What does your company do, is it any good, how does it compare to a named competitor.
Step two: ask them in a clean session
This matters. Assistants personalise based on your history, so asking from your usual logged in account gives you a flattering and useless result. Use a fresh session, a logged out window, or temporary chat mode. You want the answer a stranger gets.
Run the same questions through at least three engines. ChatGPT, Perplexity and Gemini will give you a genuine spread, and Perplexity has the advantage of showing its sources, which tells you where the others are probably drawing from too.
Step three: record it in a way you can compare
A spreadsheet with one row per question and one column per engine. In each cell record three things: whether you were named, who else was named, and which sources were cited if the engine shows them.
Do not summarise as you go. Paste the actual text. Patterns become obvious across twenty raw answers that are invisible in your impression of them.
Step four: read it for the three things that matter
Your share. Count the questions where you were named at all. That number, out of twenty, is your rough share of answer. Most companies scoring their own category for the first time land somewhere between one and four.
Who beats you. Tally which competitors appear most. Frequently the winner is not your biggest rival but a smaller company that publishes more clearly.
What the engines get wrong. This is the highest value output of the exercise. Wrong service lines, outdated locations, an old business name, a capability you dropped years ago, or a claim about your pricing that has never been true. Each error traces back to a source you can go and fix.
The errors are the gift. Every one is a specific, fixable reason an engine is hedging about you.
Step five: do it again next month
One run is an anecdote. The value is in the trend, and answers drift more than people expect. Same questions, same engines, same clean session, once a month, in the same spreadsheet.
If this sounds like something you would rather not maintain by hand, that is fair, and it is most of what our free audit does, at greater scale and across more engines. But do the thirty minute version first. It will tell you whether this is a problem worth paying anyone to solve.