We asked one model what deodorant to buy — fifty separate times — and checked every answer against what Target, Ulta and Walmart actually carry. The question isn't which brands AI likes. It's how clear the AI's picture of the category is. This week: not very. But that conflicting reality is more about where you buy your deodorant than what deodorant you buy.
No panel. No survey. No opinions — ours or anyone's. We asked one AI model the questions real shoppers ask: What's the best women's deodorant? I'm about to buy one — what should I get? Which one's safest? What are the top brands? If you were buying for me, what would you pick?
Fifty separate conversations, each one starting from scratch, so no answer could crib from another. Half asked about women's deodorant. Half asked the way people actually talk — antiperspirant for women. No web browsing, no memory, no follow-ups. Just the model and the question, same as it meets a shopper at midnight.
Then we counted. Who gets named. Who gets named first. How hard the answer pushes. And whether what it says is actually true — checked against store shelves, FDA records and the brands' own paperwork.
The score blends three things: does the AI answer the same way twice, is what it says true, and does its ranking match the real category. That's clarity, measured.
Here, ask the same question five times and you get a different top five almost every time — the answers agree with each other less than half the time. What they do say is mostly accurate. Translation: the model knows this category. It just hasn't picked its winners.
This is the first time we've measured deodorant, so there's no arrow. The arrow shows up when we come back.
Add it up. The model knows every brand we tested — even the ones it never mentions. Most of what it says is accurate. And still: ask five times, get five different lists, and a ranking that barely resembles the real aisle. That's what 65.1 means. Not a broken picture — an unsettled one.
And the blur has a shape. Brands the model knows cold but never says. Brands it files in the wrong aisle. Brands it describes as what they used to be. Brands it recommends that aren't in the store. Four different ways for a picture to be unclear, and every one of them showed up in a single category. And the biggest gaps all point the same direction: the AI describes the boutique aisle, while the brands it buries live in the mass one. The disagreement is less about what to buy than where.
This is the baseline. The point of an index is the second measurement: come back, run the same fifty conversations, and see whether the picture settles, drifts, or flips. The AI's picture of women's deodorant is still forming. Now there's a number on it.
This is the reality check inside the clarity score: does the AI's picture of the category match the real one? We counted every deodorant Target, Ulta and Walmart carry — brand by brand, from the retailers' own listings, three reads per count, the same day we ran the questions. That's the measuring stick, and it's mass retail on purpose: not Sephora, not Amazon, not anybody's own website. Remember that. It matters on both sides of this table.
Why Amazon isn't in the count: on a marketplace, nearly every brand is "carried" — by somebody, at prices the brand doesn't set. A count there measures listing chaos, not distribution. So read the zeros correctly: a brand at zero isn't unbuyable — Salt & Stone sells on Amazon, at Sephora, and on its own site. It's absent from the biggest aisles in mass retail. That's the gap being measured.
| Brand | Mass shelf | Named | Gap | |
|---|---|---|---|---|
| Each & Every | 8 SKUs · one retailer | 20 of 25 | +73.9 | |
| Salt & Stone | 0 SKUs · all three | 15 of 25 | +69.6 | |
| Lume | 76 | 25 of 25 | +50.0 | |
| Certain Dri | 54 | 23 of 25 | +47.8 | |
| Schmidt's | 46 | 16 of 25 | +43.5 |
| Brand | Mass shelf | Named | Gap | |
|---|---|---|---|---|
| Arrid | 125 | 0 of all 50 | −50.0 | |
| Dry Idea | 87 | 0 of all 50 | −37.0 | |
| Degree | 333 | 14 of 25 | −26.1 | |
| Right Guard | 196 | 2 of all 50 | −71.7 |
Right Guard is crossed out, not claimed — that story's below. Keep it in the math and the number is 65.1; grant its exclusion entirely and it's 65.8 — Unsettled either way. Arm & Hammer isn't in the table at all: it failed two of our checks and we couldn't finish its shelf count. If we can't measure it cleanly, we don't print it.
The model recommends Salt & Stone in 15 of 25 conversations. Target, Ulta and Walmart carry zero of its products. We counted that zero by hand, twice, because it seemed impossible for a brand this size — over $165 million in revenue last year, private equity bought control in March. But it's real: Salt & Stone lives at Sephora and online. The AI is recommending the boutique shelf to everyone — including the woman standing in a Target aisle, asking her phone what to buy, surrounded by brands the answers barely mention.
125 products on the mass shelf. Zero mentions in fifty conversations, either phrasing. Our first guess: the model just doesn't know a 1930s drugstore brand. So we asked it about Arrid directly. It knows everything — the era, both ad slogans, who owned it, who bought it in 2001. Asked point-blank if Arrid belongs in this category, it said yes, three out of three times. Then it went right back to never mentioning it. Arrid isn't missing from the model's memory. It's missing from its mouth. It's filed under history, not under options.
87 products, zero mentions in fifty. And here's the twist: its slogan is more famous than it is — never let them see you sweat. While we were testing, the model credited that line to the wrong brand before catching itself. The words made it. The product that paid for them didn't.
Before you decide the AI just likes new internet brands, look at Certain Dri. Older than most of the buried brands, about as glamorous as a tube sock — and it shows up in 23 of 25 conversations. Why? Because it's the standing answer to a question people actually say out loud: I sweat too much, what actually works? That's the one clear spot in the picture: a real question with a settled answer. Most of this category doesn't have one.
Right Guard should be the biggest story on this page — almost 200 products on the shelf, all but invisible in the answers. We're not claiming it, and here's why. When we tested the miss, the model refused, three out of three times, to call Right Guard a women's brand at all. It isn't being buried. It's been filed as a men's brand, no matter what aisle it sits in.
We tested every miss three ways: Does the model think the brand belongs in this category? Does it know the brand at all? Does it show up if you change the question? Three fresh runs per test, transcripts saved. One brand failed and got crossed out. One got held. The rest earned their spot.
The brand named in every single conversation is never the actual recommendation — and what the model believes about it stopped being true two years ago.