Your brand is 'mentioned' by AI. That doesn't mean what you think.

Your brand is "mentioned" by AI. That doesn't mean what you think.
Last week I scanned how 6 AI assistants — ChatGPT, Gemini, Perplexity, Qwen, GLM and Kimi — answer questions about 5 brands. The results upended a number I'd been quoting casually.

The headline: a brand can be "mentioned" 80% of the time and have a verifiable source behind less than a third of those mentions.

Here's the raw data, pulled from our monitoring runs this week:

Model Mention rate Citations returned
GPT ~98% ~98%
Perplexity ~98% ~98%
Gemini ~60% ~60%
Qwen low 0%
Kimi low 0%
Across all brands and models combined: 73% of mentions carried a real source URL. 27% didn't — and for Qwen and Kimi, it's 100% of the time.

What "mentioned" actually means
When ChatGPT names your brand in an answer, one of three things happened:

It has a real source it can cite — your site, a review, a directory listing. This is the "good" mention. It means the model found evidence and used it.
It's generating from training data — your brand exists in its weights from months ago. The answer may be right, but there's no live source to check.
It's pattern-matching a plausible answer — and here's the uncomfortable part: for a brand with little indexed footprint, this is indistinguishable from hallucination until you check the sources.
We tested this with a pre-launch brand. Its AI coverage said "80% — known by all 6 AIs." Sounds great. But 31% of those mentions had a source. The model was willing to say the name — that's not the same as the brand being established in its knowledge.

Why this matters more than mention rate
A founder's real question isn't "does ChatGPT know my brand?"

It's: "when a prospect asks AI about my category, does the AI describe me correctly — with evidence?"

A brand can be mentioned 5/5 and lose every decision, because:

the mention is a hallucinated "yes" with no substance, or
the AI quotes a source that's wrong (we found a model citing a different company with the same name), or
the mention is a caveat ("not independently verified")
Mention rate tells you if you're visible. Citation rate tells you why — and whether it's real.

What we're doing about it
We're building this in layers, and the first one shipped this week:

Every mention now reports a citation rate — what % of your mentions come with a verifiable source
Models that can return citations (GPT, Perplexity, Gemini) are forced to via web search
Models that can't (Qwen, Kimi) are flagged — their mentions are labeled "may be from memory rather than evidence"
Source freshness is next — the "AI quoted a 5-year-old thread" problem
The honest takeaway for founders: stop quoting your mention rate as if it's awareness. Ask your tool for the citation rate instead. If your AI-visibility report can't tell you which mentions have receipts, it's telling you a number, not the truth.

We're BrandScope — we measure how AI assistants actually talk about your brand, with receipts. Run a free scan of your brand at brandscope.dev.