We read ingredient lists for a living. Here's what we've learned — evidence first, marketing last.
We read all 1,788 GDS hair dye shampoo reviews as a dataset: what the 88% five-star crowd actually praises, what the 83 unhappy reviews say, and the one complaint even happy buyers repeat.
The EU's safety committee adopted its tea tree oil opinion on 30 October 2025 and set ceilings running from 2.0% in shampoo down to 0.1% in face cream, with nothing sprayable. We hold seven products with tea tree in them, one of them a body mist, and we cannot check a single one of those ceilings because no label prints the percentage.
Our 56 products carry between 1 and 87 ingredients each, and the correlation between that count and the safety score we assign is 0.05. Nine PubMed queries later, we could not find a single paper testing whether ingredient count predicts irritation or allergy in cosmetics.
The 42× regional absorption figure is one steroid measured in 1967, and skin thickness does not explain the ranking it produced. When somebody finally measured a real cosmetic ingredient across seven body sites in 2025, the gap shrank to under two-fold.
Our eight hair products average 8.99 on our screening score while the other forty-six average 9.36, and six of our eight worst-graded ingredient entries sit in that category. Most of the gap turns out to be a limitation of how we score, not of the products.
Korea's cosmetic type list names 'essence' once as a bare example and never defines it, and the words 'ampoule' and 'serum' appear nowhere in Korean, EU or US cosmetic law. We counted our own catalogue too: five product names use those three words, and they land in five unrelated categories we invented ourselves.
We went looking for a July 2026 Korean labeling revision and found that it does not exist. What turned up instead: Korean law sorts every cosmetic into 13 types, our shop uses 54, and four of our eleven cleansers are not in the cleanser category at all.
Korea amended its Cosmetics Act to require safety assessments, phased in from 2028 to 2031. We tried to read the draft rule, got blocked at the server, and read the EU's equivalent instead.
We re-ran our own benchmark four different ways on the same 54 products and the same 117 rival entries, and the result flips from a 37-10 win to a loss depending only on which column we read. This is all four runs, plus the 18 rival scores where our two internal tables disagree with each other.
Parabens appear in none of our 54 products and phenoxyethanol in two. What actually holds the line in 42 of them is 1,2-hexanediol, which our own data cannot decide how to label.
We read all 30 ingredients in a gel whose name promises cica and found no Centella asiatica anywhere. Here is what we did about it, and what the naming rules actually say.
Scrubs, keratolytics and laser have very different evidence behind them for keratosis pilaris. We expected laser to be the weak one and had that backwards.
A probiotic is defined as a live microorganism, and a preserved cosmetic cannot keep one alive. We use the word on two of our own products, so this one starts at home.
Warm skin makes more sebum, and a controlled crossover trial puts a number on it. The famous figure of 10% per degree, though, turned out to be something we could not verify.
Nineteen of our products carry an alcohol-free chip and not one of them contains volatile alcohol. The real problem was that we spell the same idea eight different ways.
Antiperspirants and deodorants are legally different products in both the US and Korea. Remove the aluminium and you remove the only ingredient class approved to reduce sweat.
Eleven of our fifty-four products contain a ceramide, and five of those carry the same five-ceramide set. We counted where they sit on the label, checked them against the barrier-repair evidence, and found our own highest-scoring cream failing the test.
Korean cosmetics law never requires a pH figure on a label, and the functional-cosmetics review rules exclude pH from the standards they set. We checked the three weakly-acidic claims in our own catalogue against all fifty-six ingredient lists and could only verify one of them.
China's electronic-label pilot took effect on 1 February 2026, and on a pilot product's physical package only six items stay mandatory — the ingredient list is not one of them. Europe has not done this for cosmetics, Korea has only a filed bill, and our own 61-line dye shampoo shows why the printed block still matters.
Menthol binds TRPM8, the ion channel that reports cold, so the nerve fires without the temperature moving. We counted six menthol products in our fifty-six, found mint in three more under names that never say menthol, and found a bug in our own grading while we were looking.
Commission Regulation (EU) 2023/1545 added 45 entries to the EU fragrance allergen list, and the deadline for placing non-compliant product on the market passed on 31 July 2026. We read all fifty-six of our own ingredient lists and found sixteen generic fragrance declarations, ten names, and not one substance from the new entries.
A hair essence we sell says "100% Moroccan Argan Oil" on the front and lists argan sixth of thirteen on the back, behind three silicones. The claim is about a raw material, not the bottle, and Korean labelling rules stop ranking ingredients once you drop below one per cent.
Seasonal skin change is real and has been measured repeatedly, with water loss rising and hydration falling in winter across studies of 14 to 140 people. Every one of those studies compared whole seasons, so the cold, the dry air, the indoor heating and the hotter showers all moved together and none of them can be told apart.
Panthenol turns up in 24 of the 56 products we sell, which is more than niacinamide manages, and every one of those 24 rows is graded low. The problem is our own database: we filed the same molecule under eight different job descriptions, two of them in lower case.
