Five Questions That Dismantle Most Ingredient Claims in Under a Minute

The citation is an invitation, not a verdict

A reference on a product page performs a specific rhetorical job: it converts a claim into something that looks settled. Most people, quite reasonably, read the superscript and move on. The claim now has a study behind it, and arguing with a study feels like arguing with science.

But a citation is only a pointer to an experiment, and experiments vary enormously in what they can support. The same ingredient can have a paper showing it triples collagen output and still do nothing measurable on a face, without anybody having lied. The tripling happened in a dish, to cells that were never asked to sit under a stratum corneum first.

What follows is not a course in research methods. It is five questions, ordered so that the cheapest ones come first, and most claims fail before you reach the fourth. Running them on the back of a box takes under a minute once the habit sets in — and the Ingredient Decoder does the first half of the work by telling you what is actually in the formula and at roughly what position on the list.

Question one: what kind of experiment was this?

In vitro means cells in a dish. It is where mechanisms are discovered, and it is nearly useless for predicting what a product does on skin, because the experiment removes the barrier that is the entire problem. Fibroblasts bathed directly in a compound will respond to concentrations that could never reach them through skin. 'Stimulates collagen synthesis by 200% in vitro' is a real result about a real molecule and says nothing about the cream.

Ex vivo — excised human skin in a diffusion chamber — is a genuine step up, because the barrier is present. It answers penetration questions well and biological-outcome questions poorly, since the tissue is not alive in any interesting sense.

In vivo on humans is the only category that can support a claim about what happens to a face. Even here there is a hierarchy: instrumental measurements beat investigator grading, investigator grading beats subject self-report, and the thing at the bottom — 'in a study, 87% of women agreed their skin looked more radiant' — is a satisfaction survey wearing a lab coat. It has a sample size and a percentage and it is not evidence of a biological effect.

The practical filter: if a claim rests only on in vitro work, treat it as a hypothesis about the ingredient rather than a fact about the product.

Question two: could the molecule get in at all?

This one disqualifies faster than anything else and takes ten seconds. Human skin resists the penetration of large molecules, and the widely used working rule — argued in Experimental Dermatology from three independent lines of evidence — is that a compound must be under about 500 daltons to be absorbed. The supporting arguments are elegant: essentially all common contact allergens fall under 500 daltons, because larger molecules cannot penetrate and therefore cannot sensitise; the standard topical drugs are all under it; and so is every agent used in transdermal delivery systems.

Apply that to the categories with the loudest claims. Collagen as an ingredient is a protein of enormous molecular weight — whatever it does on skin, replacing dermal collagen is not it. Many ↗peptides sold for topical use are near or above the threshold, which is exactly why so many are supplied in palmitoylated form: attaching a fatty acid tail is a delivery strategy, and its presence on an INCI list is a tell that penetration was a problem someone had to solve. That does not make peptide products worthless — formulas like COSRX 6 Peptide Skin Booster Serum or CeraVe Skin Renewing Night Cream are perfectly good products — but it does mean the peptide is rarely the reason they are good, and a peptide claim deserves the penetration question before the efficacy question.

Small molecules pass this test easily and move on to the harder ones: niacinamide, ↗ascorbic acid, retinoids, azelaic acid, salicylic acid. Notice that this is more or less the list of ingredients with genuine outcome evidence, which is not a coincidence.

Question three: who paid, and does that end the argument?

Manufacturer-funded research is the default in cosmetic science, not the exception, so a policy of ignoring it would leave you with almost nothing to read. The question is how to weight it, and there is decent empirical work on exactly that.

A Cochrane methodology review pooling 75 papers found that industry-sponsored drug and device studies more often reported favourable efficacy results, with a risk ratio of 1.27, and more often reached favourable conclusions, at 1.34. That is a real and reproducible bias. But the same review found something more interesting: industry-sponsored studies were more likely to have low risk of bias from blinding, not less. Where they diverged was at the end — there was less agreement between the results and the conclusions in industry studies than in independently funded ones.

That is a precise and useful finding. The bias is concentrated less in fabricated data and more in interpretation: the trial is often run to a decent standard, and then the discussion section, the press release and the box each apply another coat of optimism. So the correct response to a manufacturer-funded study is not dismissal. It is to read the results and ignore the conclusions — and to be sceptical of any claim that exists only in the abstract's final sentence.

Question four: what was it compared against?

This is the flaw that survives in even quite respectable-looking cosmetic trials, and it is the one that most reliably produces impressive-sounding nonsense.

Almost any competent moisturiser will improve the appearance of fine lines within a fortnight, because hydrating the stratum corneum makes it swell and reflect light differently. That effect is real, immediate, and has nothing to do with whatever active is being tested. So a trial comparing an active serum against no treatment, or against a bare-skin baseline, is measuring moisturisation and attributing it to the ingredient. The comparator has to be the identical formula minus the active — a vehicle control — for the result to mean anything about the active.

Split-face designs handle a related problem, since they remove between-person variation by using each subject as their own control. And blinding matters more here than people expect: an unblinded assessor scoring photographs of skin knows which side got the expensive serum, and 'photographic assessment of improvement' is exactly the kind of soft endpoint that assessor expectation moves.

When you see a before-and-after, the questions are: against what, judged by whom, and did the person judging know? A large fraction of cosmetic claims never had a comparator at all, which is why ↗collagen as a topical continues to be sold enthusiastically decades after the physical chemistry settled the matter.

Question five: is the study even about the product?

This is the last question and the most damning, because it can defeat a genuinely excellent study.

Take the best-documented case in the category. Percutaneous absorption work on L-ascorbic acid established that it only enters skin when formulated below pH 3.5, that the maximum useful concentration is around 20%, that tissue levels saturate after three daily applications, and — the part almost nobody cites — that magnesium ascorbyl phosphate, ascorbyl-6-palmitate and dehydroascorbic acid did not increase skin levels of ascorbic acid in the same experiments. A product built to those specifications, like Timeless 20% Vitamin C + E Ferulic Acid Serum, is inheriting that evidence legitimately. A gentle, pH-neutral serum built on a derivative such as The Ordinary Ascorbyl Glucoside Solution 12% is a defensible product for people who cannot tolerate low pH, but it is not the thing the absorption studies were about, and citing them for it is a category error.

The same test applies everywhere. Was the studied concentration the one in the bottle? Was the vehicle comparable — an aqueous serum is not an emulsion is not an anhydrous oil, and penetration differs across them? Was the study duration long enough for the endpoint claimed? A well-run ↗niacinamide trial on a specific concentration in a specific base supports products in that neighbourhood, which is a fair reading for something like The Ordinary Niacinamide 10% + Zinc 1% — and not a licence for every formula that lists the ingredient somewhere after the preservative.

Which brings the whole thing back to replication, the boundary that actually matters. A single study — however well designed, whoever funded it — is a hypothesis that survived one test. Independent repetition, in a different laboratory, on different people, is what turns it into a reason to change what you buy. This is why the ingredients that genuinely earn confidence are the boring ones with decades of repeated work behind them, and why the exciting new molecule with one impressive paper should be treated as interesting rather than proven. That distinction is the entire basis on which we grade ingredient evidence, and it is the reason the grades move slowly.