You're mentioned. Are you described correctly?

Being mentioned by an AI engine is not the same as being described correctly. A visibility score counts whether your name showed up in an answer; it says nothing about whether the sentence around that name described your actual business.

Why does AI describe your company incorrectly?

Two distinct failures hide behind a mention count that looks healthy. Entity independence: an engine describes your product as a feature of its parent company, or places it in the wrong category entirely, mentioned but not recognized as itself. Identity accuracy: a share of your mentions actually belong to an unrelated business that happens to share your name, mentioned but describing someone else. In one measured case, a brand scored 70% identity accuracy across 57 classified mentions: 8 came back confused with another entity, 6 came back unresolved.

This shape is one measured case, not a universal number. The detector that separates these two failures shipped 2026-07-21 (collimer issue #1755, PR #1769) and flags above a 15% confused-or-unknown share on a minimum of five classified mentions.

Why pooling them into one score hides the problem

A single visibility number cannot tell these two failures apart, and that matters because they have different causes and different fixes. Entity independence is a positioning problem: the wider web talks about you in someone else’s terms, usually because a parent-company or acquisition narrative dominates the record. Identity accuracy is a disambiguation problem: a name collision that a distinctive, findable description can resolve, but a generic one can’t. Pooling both into a single rising or falling mention count gives a reader nothing to act on. The same “60% visible” reading could mean either problem, or both, in wildly different proportions.

In one measured case, a brand scored 70% identity accuracy across 57 classified AI mentions: 8 confused with another entity, 6 unresolved. That is exactly what a pooled visibility number cannot distinguish from the healthy case.

How to check your own identity accuracy

You do not need a full scan to get a rough read in one sitting:

• Ask five AI engines “what is [your brand]?” and read each answer as if you had never heard of the company.
• Count how many answers describe the category you actually occupy, versus a parent company’s feature or a same-name business in a different industry.
• If two or more come back wrong, you have an identity problem a mention-count score would never surface. How five AI engines decide what to cite explains why the same brand gets different treatment across engines.

This is a different failure than getting no mentions at all. See why AI says it doesn’t know your brand yet for the zero-mention case, and why your Copilot number is an estimate for a related measurement caveat on a single engine.

Below five classified mentions, the honest answer is silence, not a caveated guess. A thin identity signal is more misleading caveated than suppressed.

See what your own scan surfaces on both counts: run a free scan; it takes about 90 seconds.

For agents: try this yourself

• “Ask five AI engines ‘what is [a specific brand]?’ and check whether each answer describes the right category, a parent company’s feature, or an unrelated business.” Compare how often the description is wrong even when the mention itself is right.
• “If a brand’s mention count is healthy but a share of those mentions describe an unrelated company, is that a visibility problem or an identity problem?” Reason through why the fix differs for each.

---

Drawn from Collimer’s cited research library and our own detector work, shipped 2026-07-21. We name the sample sizes and thresholds because a thin sample is easy to overclaim from. As of August 2026.