Why we throw away most of what AI says about your competitors

Ask an AI tool who your competitors are and it will answer with total confidence. Ask it twice and you may get two different lists. When we built the competitor tracking inside WhoCanFindMe, this was the first problem we hit, and the way we solved it explains a lot about how honest AI measurement has to work.

The confident nonsense problem

Early in development we pointed our audit at a small business automation site. The AI we use to build a brand profile came back with a competitor list: Zapier, UiPath, Automation Anywhere, Workato, Blue Prism and Kofax.

Impressive names. Recognisable brands. Completely wrong.

The site in question named none of those companies. It did not sell what most of them sell. It competed with two local agencies and a freelancer, none of which appeared in the list. The AI had done what AI does when it is unsure: it reached for the most famous names in the general category and presented them as fact.

If we had shipped that list, every report would have shown a small business "losing" a share of voice battle against six enterprise software giants it has never competed with. The numbers would have looked rigorous. They would have meant nothing.

What we do instead: evidence or nothing

So we added a rule to the product that we still follow today. A competitor only appears in your report if one of two things is true.

First, you confirmed it yourself. When we build your brand profile, you see the list before any measurement runs, and you can strike anything that does not belong.

Second, we actually observed it. When we ask the AI engines the questions your buyers ask, we record which businesses get named in the real answers. A company that keeps appearing next to you in genuine engine responses is a real competitor for AI visibility, whatever your own market map says. A company the profiling step merely guessed, and that never shows up in a single answer, gets dropped.

Everything else goes in the bin. Most of what the AI guesses goes in the bin.

Why this matters beyond competitor lists

The competitor list is one example of a general rule we apply across the product: an AI claim is a hypothesis, not a fact, until something observable backs it up.

The same thinking is why a brand mention only counts when the engine actually talks about your business, not when it echoes your name back or answers about a lookalike company. We wrote about that in three ways a brand appears in an AI answer. It is why we suppress stability bands below three repetitions instead of decorating a single run with fake confidence. And it is why the share of voice number in your report is computed only from businesses observed in real answers, so the denominator is real.

Plenty of tools in this market do the opposite. They let a model guess, format the guess as a table, and ship it. The result looks like measurement. It is closer to astrology with a spreadsheet.

What you should take from this

If you use any AI visibility tool, ours included, ask one question about every number on the screen: what observation is behind this?

If the answer is "a model said so", treat it as a lead worth checking, nothing more. If the answer is "we asked real questions, recorded real answers, and counted", you can act on it.

And if you have never seen which businesses AI engines actually name next to yours, that is worth knowing before your next marketing decision. The answers are often surprising in both directions: competitors you obsess over that never appear, and businesses you have never heard of that are quietly winning the questions your buyers ask.

Run a free scan at whocanfindme.com. No signup. Takes about ten seconds.