Direct answer
On the query "what are the best car tire brands?", measured in a France configuration (French language, French market), a real measurement shows that Michelin is cited at the 1st rank by the AIs (average rank 1.0), with a 20% share of voice in a leading group of 6 brands. But the same report reveals things a simple search never shows: the brand goes from a presence of 35% to 100% depending on whether AI searches the web or not, the number of competitors swings from 25 to 76 depending on that same setting, and two AI providers disagree on the brand. That's what an AI visibility measurement measures — far beyond "am I cited."
The problem
"Does AI speak well of my brand?" Most people answer this question by opening ChatGPT, typing their category, and seeing whether their name comes up. If it does, they're reassured. If not, they worry.
The problem is that this snapshot tells you almost nothing. It doesn't tell you how stable it is, in which mode AI answered, who the hidden competitors are, or whether the different models agree with each other. Let's take a real case — a strong brand, Michelin, on its home turf — and look at what a real measurement reveals that the snapshot misses.
The idea to grasp
Here's what the measurement revealed, indicator by indicator. Every figure comes from the real report.
Position. On the tested query, Michelin comes out 1st, average rank 1.0, no ties, out of 85 cited brands. It's the only indicator the manual snapshot could have come close to. Everything else escapes it.
Presence by mode — the decisive point. Without web search (AI's memory), Michelin's presence with one provider is 35%. With web search, it rises to 100%. The same brand, the same question — a 65-point gap depending on an invisible setting. A snapshot taken "at random" in either mode would have shown one or the other, without warning.
The hidden competitors. From memory (model knowledge), AI surfaces 25 competitors. With the web (live web), 76 discovered competitors — those the models cite spontaneously, far beyond the 4 competitors the brand had declared itself (Continental, Pirelli, Bridgestone, Goodyear). Three players nobody was watching come out on top of the discovered ones: Hankook, Falken, Uniroyal.
Share of voice. Across all cited brands, Michelin captures 20% of citations. The landscape is a "tight pack": a leading group of 6 brands shares most of it.
Associated themes. The AIs mainly link Michelin to durability (74 citations), safety (36), and performance (15). Conversely, ecology (1), innovation (2), and price (2) are nearly absent. In other words: AI tells a story about the brand — and that story has holes.
Disagreement between AIs. The report rates the agreement between providers as low. On presence, one provider measures 35%, the other 100% — a 65-point gap. They do, however, agree on rank (1.0 across the board). So two AIs don't "see" the brand the same way.
None of these six insights comes out of a manual search. They come out of a measurement: repeated (20 queries per mode and per provider), compared across modes and across providers, and computed mechanically.
What you hear everywhere
"Michelin is the leader, of course AI speaks well of it." Partly. But AI barely associates it with innovation and price, and one provider only cites it spontaneously one time out of three without the web. "Leader" doesn't mean "well represented everywhere."
"If a big brand has blind spots, a small business has no chance." On the contrary: it's proof that the measurement reveals useful gaps even for the strongest. A small business often discovers there that it exists in one mode and not the other — directly actionable information.
"You just have to ask AI, you can clearly see what it says." This case flatly contradicts that. Depending on the setting, you would have seen 35% or 100% presence, 25 or 76 competitors. "What it says" depends on how you ask it — and the snapshot doesn't show that.
My underlying stance: facts only. And the fact here is that an absolute-leader brand still has six dimensions of visibility that no impression reveals. The measurement doesn't flatter, it doesn't darken: it states.
My vision: what a measurement gives that a search doesn't
From here on, the register changes: we describe the instrument.
This case illustrates what a methodically conducted AI visibility measurement produces:
- Repetition: 20 queries per mode and per provider, not a single isolated query.
- Separation of regimes: presence from memory (35%) vs. with web (100%) measured separately — the gap becomes the information.
- Mechanical counting: presence, average rank, share of voice, competitors, themes — everything is computed, never "judged" by an AI.
- Comparison across providers: the disagreement ("low" agreement) is itself a result.
- Dating and sealing: the report carries a certificate number and an HMAC-SHA256 seal, both verifiable — and therefore defensible.
The usual caveat. These figures describe what the AIs answer under known measurement conditions. They don't tell the "truth" about Michelin: they tell what the models say about it. mAIr measures what the AIs say, not the reality of the brand — and it doesn't sort true from false.