Lipsticks are named Scandal, Chocolate Kiss, Desert Rose. Reds get sin and danger. Nudes get coffee and skin.
If a shade is called Sinner, is it actually a deep red? If it's called Chocolate, is it actually brown? Or are the names just words?
To check, every lip product Sephora sells was collected (lipsticks, glosses, liners, balms) and the exact colour of each shade was measured from its swatch image. Then the names were tested against the colours.
This is what was found.
The data comes from Sephora's catalogue. It covers — lip products, which between them offer — individual shades.
To compare the colours, each shade was measured for lightness: a scale from 0 (black) to 100 (white) that says how light or dark a colour is, ignoring what hue it is. A dark burgundy might score around 30; a pale sheer pink around 70.
Averaging by category shows a pattern. Lipstick shades are the darkest, averaging 46. Lip balm averages 58 and lip plumper 62. In other words, the more a product is about adding colour, the darker its shades tend to run; the more it is about caring for the lips, the paler they are.
Lightness is only half of a colour. The other half is hue: where a colour sits on the colour wheel, measured in degrees from 0 to 360. Red is near 25 degrees, green near 140, blue near 260.
Plotting all — shades by hue across and lightness up produces the shape on the left. It is a narrow band.
64° is the arc of the wheel that holds 95% of all lip makeupThis is an arc, not a position. The small wheel shows it. Of the full 360 degrees, lip makeup uses a wedge of about 64, running from berry through red to orange and brown. That is roughly a sixth of the wheel, and the rest of it is essentially unused.
Blue and green lip makeup does exist, but barely. 144 coloured shades fall outside the arc, about 5% of the total, and they include novelties like Dior's Icy Blue and Sephora's Very Blue.
A further 126 shades are colourless. A grey has no hue at all, so they cannot be placed on this axis and are set aside.
This matters for the rest of the page. If the colours barely vary, there is very little for a name to describe...
Turn the question around. Instead of asking what colour a name is, take a colour and ask what everybody calls it.
The shades are grouped by measured colour only, never by name. A group is only kept if every shade in it sits within 6 dE of the group centre, so a row really is one colour rather than a broad region. Two colours already look different at about 2.3 dE. That tolerance is strict and it excludes two thirds of the catalogue.
Each row is one colour, built from its own real shades. Beside it are the words the brands actually use for it.
One colour, sold by 32 brands, is called clear, vanilla, glass, mint, rosy and outlook. Another, sold by 41, is called nude, wherever, walnut, talk and mauve.
The bar asks whether those names mean similar things, judged by a language model that knows raspberry and candy are related but raspberry and espresso are not. Zero means the names are no more related than a random handful of shades.
Mostly, they are not. Of the fifteen colours, one agrees more than chance allows and two disagree more than chance allows. The other twelve show no signal at all. Brands sell the same colour and reach for unrelated words.