Undeclared substances in 98% of products: what the study says

A study published on August 11, 2026, in the American Chemical Society journal Environment & Health analyzed 113 personal care and household cleaning products sold in the United States. In 98% of cases, it found at least one substance that was not listed on the label. In 85%, it found at least one substance associated with a known or suspected hazard. Within days, the figure became a headline: almost all cosmetics hide something.

The number is real. Its translation is not. That study did not compare ingredients with ingredients: it compared the names written on labels with molecules detectable in the laboratory, which are two different units of measurement. The truly serious figure is not 98%. It is the 26% of products that contradicted the promises printed on their own packaging.

[INDICE:h2]

What they actually did

The researchers used a non-targeted analysis. In traditional testing, a list of substances is decided in advance; here, the instrument records everything it can detect, and only afterward is an attempt made to assign names to the signals.

The method used was two-dimensional gas chromatography coupled with mass spectrometry, supported by machine learning for identification. A total of 3,141 distinct chemical signals were recorded, of which 303 received a confirmed, probable, or merely tentative identification.

That confidence scale is the first thing the headlines threw away. A peak on a chromatogram is not automatically a substance identified with certainty, and fewer than one-tenth of the 3,141 signals had a name, with varying degrees of confidence.

Sixteen names on a label and 51 molecules in a test tube are not the same thing

The comparison that generated the headline is this: an average of 16 declared ingredients versus 51 detected substances. Three times as many. It looks like proof of concealment.

In reality, the two figures are not comparable, and anyone who formulates products knows that perfectly well. INCI lists the ingredients the formulator puts in; the laboratory detects the molecules those ingredients contain. A plant extract is a single ingredient and takes up one line on the label: inside it, there may be hundreds of different molecules, because it is a plant. A fragrance takes up a single line, and inside it there may be dozens of molecules.

Added to this are traces from raw materials, technically unavoidable residues, molecules formed through degradation or interactions between ingredients, and substances that migrate from the container. None of this is an ingredient added secretly. The authors themselves write that the differences may come from raw materials, equipment, processing, storage, packaging, chemical reactions, or changes of supplier.

“A label tells the story of the formula someone decided on. A sensitive analysis also captures everything that entered by other routes.”

Dangerous and risky are not synonyms

85% of the products contained at least one undeclared substance associated with a known or suspected hazard. This is the figure that is most alarming, and it is also the one that needs to be read most carefully.

Hazard is a substance’s inherent ability to cause harm under certain conditions. Risk depends on the dose, the route of contact, the frequency, the duration, and how much is actually absorbed. These are two different questions, and only the second concerns the person using the product.

The study detected the presence of substances, not precise quantities: the estimates were semiquantitative. It did not measure how much of those substances passed through the skin, and it observed no health effects in anyone. It says they were there. It does not say how much, or with what consequences.

That does not justify the opposite reaction, namely dismissing everything as irrelevant. Even a minimal trace matters if the substance is banned, sensitizing, or tends to accumulate. Simply put: finding that out requires figures that this study was not designed to produce.

The figure I would have put in the headline

Not 98%. The 26%more than one in four products contained substances that directly contradicted the claims on the packaging. “Phthalate-free” products that contained phthalates. Products presented as natural or “non-toxic” that contained synthetic contaminants. Overall, 57.5% failed the check for undeclared contaminants.

There is no ambiguity about units of measurement here. Anyone who buys a product that promises the absence of something is buying a promise about the finished product, not about the good intentions behind the initial formula. The origin may be accidental—as it almost always is—but if you declare an absence, that absence must also be verified at the output stage, not just on the supplier's paperwork.

That is why, in my work, I distrust communications built around negations. A formula should be judged by what it contains and how it was tested, not by the list of things it claims not to contain. I wrote about this when discussing how the fear of chemistry made its way into sunscreens: it is the same mechanism, seen from the other side.

Three limitations that do not appear in the headlines

1. The products were purchased in 2021 and 2022. The data is released today, but it reflects shelves from four or five years ago. During that time, several formulas were revised, and some regulations changed as well. Almost no one mentions this.
2. It is not a representative sample. One hundred and thirteen products selected to cover different categories do not describe an entire market, let alone a market different from the one in which they were purchased.
3. They are not just cosmetics. The sample also includes household and laundry detergents, which have different uses, exposures, and regulations. The headline “98% of cosmetics” rewrites the original sample.

I should add that the product names were not published and are expected to appear in a subsequent report. It is therefore not currently possible to verify which formula produced which result, and trying to guess the brands involved is a waste of time.

Why those figures cannot be transferred to Europe as they stand

In the United States, an ingredient list is mandatory, but fragrance compositions may be declared collectively under a single term, and there is some leeway for incidental ingredients and trade secrets.

In Europe, Regulation (EC) 1223/2009 requires a responsible person, a signed safety assessment, a product information file, and good manufacturing practices. Even here, however, the label does not contain every detectable molecule: fragrance appears as parfum, and individual allergens must be listed when they exceed the prescribed thresholds. Technically unavoidable impurities are not intentional ingredients, but they must be compatible with product safety.

I’m not saying that one continent is perfect and the other isn’t. I’m saying that a percentage obtained from products purchased in another market, with different labeling rules, does not automatically describe what you have in your bathroom.

What really changes for shoppers

Nothing here suggests throwing away the products you are using: there is nothing in this study that justifies it. If anything, change the way you read a package.

• A short list is not automatically safer, and a long one is not automatically worse.
• Apps that rate individual ingredients assess neither the concentration nor the formula as a whole: they are answering a different question from the one you are asking.
• Absolute claims — “natural,” “clean,” “non-toxic” — are not regulated definitions. When they replace an explanation, they deserve more skepticism than a long chemical name.
• If your skin reacts, the useful data is not on the label: it is in your reaction. Stop using it, return to a minimal routine, and reintroduce one product at a time.

On skin that is often irritated, incidentally, the variable that matters most is almost never a trace contaminant: it is fragrance, which remains the leading cause of contact sensitization in cosmetics. That is why a cream for reactive skin is formulated by taking things out, not adding them.

The part that concerns me directly

This study does not reinforce the fear of chemicals. It reinforces something far less spectacular and far more useful: the need to know your supply chain.

When I work on a formula, the INCI is the last thing to be written. First come the raw material specifications, the supplier’s documentation, the declared purity and its consistency from one batch to the next, compatibility with the container, and stability tests. The same ingredient is not the same thing depending on its origin, process, and any change of supplier — and that is exactly the point on which this study, regardless of the headlines, is right.

It’s the same reasoning I use when choosing the active ingredients for LeLang formulas: the question is never whether an ingredient is good in the abstract, but which version, from whom, and with what checks.

And this is why we chose pharmacies rather than another channel: because it is the place where someone can read a label with you and tell you which part matters for your skin. At the pharmacies that distribute LeLang, you can also bring in the products you already use and have them examined: it is a question no app can answer for you.

What I really think

Perhaps the right question is not “can we still trust cosmetics?”, which is a question that leads nowhere. The right question is how thorough the control of the finished product is, beyond the correctness of the formula on paper.

It is a less frightening and much more uncomfortable question, because it calls into question those who produce instead of frightening those who buy. Untargeted analyses will not replace traditional checks and cannot assess any risk on their own, but as a monitoring tool — to detect what you were not looking for — they are the most interesting thing to come out of this research. Much more so than the 98%.