Why this site does not use comedogenic ratings
Nearly every ingredient checker prints a number between 0 and 5. This one does not, and the reason is not caution — it is that the number answers a question nobody is asking.
Where the 0–5 scale came from
The idea that cosmetic ingredients can provoke comedones was introduced in dermatology in 1972, in the paper that named acne cosmetica. The scale people quote today descends from a rabbit ear assay published in 1984, which screened therapeutic products, cosmetics and raw ingredients by applying them to rabbit ear skin and counting the comedones that resulted.
That study is real, it is peer-reviewed, and it is the origin of most ingredient-level comedogenicity lists still in circulation. Two things happened to it afterwards. The first is that its results were copied between sources for four decades, often losing the attribution along the way, until ingredients that were never in the study acquired ratings anyway. The second is that the literature moved on, and the lists did not.
What an animal model can and cannot tell you
A rabbit ear assay is a screening tool. It is a way of sorting materials quickly, not a measurement of what a cosmetic does to a human face. The methodology literature is explicit about the limitation: a human comedogenicity assay study reports that a number of studies have pointed out the rabbit ear model’s disadvantage of overreacting to comedogenic materials.
It would be dishonest to leave the argument there, so here is the part that cuts the other way. A 1982 human upper-back assay reported that substances rated moderately to strongly comedogenic in the rabbit ear model were capable of inducing comedones in the human model. That is a real correlation, and it means the rabbit ear results are not noise. What it is not is a general equivalence between the two models: it is agreement at the strong end of the scale, which says nothing about whether an ingredient rated 2 rather than 4 means anything at all.
The finding that breaks the whole approach
In 2006, a human study applied a modification of that upper-back assay to finished cosmetic products — not raw ingredients, but products as sold — formulated with ingredients previously identified as comedogenic. It reported that those finished products were not necessarily comedogenic.
This is the single most important source behind this site, and it is the reason no ingredient-level number appears anywhere on it. An ingredient rating is only useful if it predicts something about the product in your hand. The direct test of that assumption found it does not reliably hold. Every checker that converts a list of ingredient names into a verdict about a product is performing a step that was tested and did not survive.
The practical consequence is that concentration, the rest of the formula, exposure time and whether a product is rinsed off or left on all sit between an ingredient name and an outcome, and none of them are visible on a label.
What this site reports instead
Four classes, no scale, and no aggregation. They are deliberately not ordered from good to bad, and they are never combined into a score for a product.
- Reported concern
- At least one cited source reports a comedogenicity concern for this name. The tier says what kind of study that was, which is usually the part that matters most.
- Conflicting or context-dependent evidence
- The cited sources disagree, or the classification is actively contested in the published literature. This is not a midpoint on a scale; it is a description of an unresolved argument.
- No concern found in the included sources
- No source included here produced a concern for this name. It is not a finding of proven non-comedogenicity, and for many records it means the corpus simply contains no assessment either way.
- Not recognised or insufficient evidence
- Either the name was not recognised, or the evidence is not established. Ingredients that circulate widely on comedogenic lists but have no verifiable assessment in this corpus land here rather than being classified in either direction.
The strongest product-level sentence this site will produce is a count of matches followed by a statement of what that count does not establish. There is no percentage, no grade and no risk level anywhere in it.
Why every record carries an evidence tier
An animal-model finding and a human finished-product finding are answers to different questions, so they are never averaged or treated as interchangeable. Each record states which kind of evidence it rests on.
- A finished cosmetic product was tested on human skin.
- A finished cosmetic product was tested on human skin. This is the only tier that answers the question people are actually asking, and it is the rarest.
- The ingredient itself was tested on human skin, separately from any product containing it.
- The ingredient itself was tested on human skin, separately from any product containing it. That is closer to the question than an animal model, but it is still not a finished-product result.
- The material was applied to animal skin — in this dataset, a rabbit ear assay.
- The material was applied to animal skin — in this dataset, a rabbit ear assay. It is a screening model, and the human-assay literature reports that it overreacts to comedogenic materials relative to human testing.
- Laboratory work on cells or tissue models rather than on skin.
- Laboratory work on cells or tissue models rather than on skin. It can describe a mechanism, but it cannot describe what happens on a face.
- A reference or regulatory source rather than a study that measured comedogenicity.
- A reference or regulatory source rather than a study that measured comedogenicity. In this dataset that most often means the source establishes only that the name refers to a recognised cosmetic ingredient.
- No source in this dataset assesses this ingredient's comedogenicity.
- No source in this dataset assesses this ingredient's comedogenicity. The entry exists to record that the question is open, not to imply an answer to it.
What “no concern found” does not mean
This is the class most likely to be misread, so it is worth stating bluntly. It means that no source included in this dataset produced a concern for that name. It does not mean the ingredient has been shown to be non-comedogenic, and it does not mean anyone tested it.
For a large share of records, the only citation attached is a regulatory database entry establishing that the name refers to a recognised cosmetic ingredient. That is an identity citation, and this project never treats identity as evidence about comedogenicity. Where that is the case, the honest reading of the record is that the reviewed literature contains nothing about the ingredient in either direction — which the individual ingredient pages say in those words.
Rules the dataset has to satisfy
These are enforced by a validator that runs during the build. A record that breaks one of them fails the build rather than shipping.
- A record may only claim a comedogenicity concern, or an absence of concern, if a source in sources.json is cited for that claim.
- Claims are limited to what the cited source itself reports. Where only an abstract could be verified, the record states the family-level or study-level claim rather than inventing a per-ingredient rating.
- Ingredient identity citations (CosIng) establish that a name is a recognised cosmetic ingredient. They are never treated as evidence about comedogenicity.
- Ingredients that circulate on popular comedogenic lists but have no verified assessment in the included corpus are recorded as not_established rather than being classified in either direction.
- No record is copied from another checker's proprietary ingredient list.
The whole corpus, and what each source supports
This is every source behind every classification on this site. It is a small corpus, and saying so is part of reporting it accurately. Each entry records what it does and does not support.
Comedogenicity of current therapeutic products, cosmetics, and ingredients in the rabbit ear
Fulton JE Jr, Pay SR, Fulton JE 3rd. Journal of the American Academy of Dermatology, 1984. Peer-reviewed. doi:10.1016/s0190-9622(84)80050-x
Rabbit ear assay. This is the historical origin of most ingredient-level comedogenicity lists in circulation today. The assay is a screening model in animal skin, not a measurement of how a finished cosmetic behaves on a human face. Later work (see src-baek-2016) reports that this model overreacts relative to human testing.
A human model for assessing comedogenic substances
Mills OH Jr, Kligman AM. Archives of Dermatology, 1982. Peer-reviewed.
Describes a human upper-back assay for comedogenic potential. Reports that substances rated moderately to strongly comedogenic in the rabbit ear model were capable of inducing comedones in the human model, which is a correlation at the strong end of the scale rather than a general equivalence between the two models.
Kligman AM, Mills OH Jr. Archives of Dermatology, 1972. Peer-reviewed.
The paper that introduced the concept of acne cosmetica and linked the use of certain cosmetic ingredients to comedo formation. Included as historical context for why ingredient-level comedogenicity lists exist at all.
A re-evaluation of the comedogenicity concept
Draelos ZD, DiNardo JC. Journal of the American Academy of Dermatology, 2006. Peer-reviewed. doi:10.1016/j.jaad.2005.11.1058
Applied a modification of the Mills and Kligman human assay to finished cosmetic products formulated with ingredients previously identified as comedogenic. Reported that finished products using comedogenic ingredients are not necessarily comedogenic. This is the single most important source in this dataset: it is the direct evidence that an ingredient-level match cannot be converted into a finished-product conclusion.
DiNardo JC. Journal of Cosmetic Dermatology, 2005. Peer-reviewed. doi:10.1111/j.1473-2165.2005.00150.x
A short commentary in the cosmetic dermatology literature that explicitly re-opens the question of whether mineral oil deserves its comedogenic classification. PubMed lists no abstract for this item, so this dataset cites it only as evidence that the classification is contested in the literature, and does not attribute a specific conclusion to it.
Analysis of comedone, sebum and porphyrin on the face and body for comedogenicity assay
Baek JH, Ahn SM, Choi KM, Jung MK, Shin MK, Koh JS. Skin Research and Technology, 2016. Peer-reviewed. doi:10.1111/srt.12244
Human comedogenicity assay methodology study in 66 volunteers. States that the rabbit ear model can replace human testing for screening purposes but that a number of studies have pointed out its disadvantage of overreacting to comedogenic materials. This is the citation behind every 'animal-model evidence may overstate the result' limitation in this dataset.
Acne: Why won't my acne clear?
American Academy of Dermatology Association, 2026. Professional body.
Consumer guidance from a dermatology professional body. Advises choosing products labelled non-comedogenic while also noting such products can still cause acne in some people. Cited for interpretation guidance, not for ingredient-level classification.
Summary of Cosmetics Labeling Requirements
U.S. Food and Drug Administration, 2026. Regulator.
Cited for how ingredient declarations are ordered. Ingredients are generally listed in descending order of predominance, with exceptions including ingredients present at one percent or less and colour additives, which is why label position cannot be read as a concentration.
Cosmetic ingredient database (CosIng)
European Commission, Directorate-General for Internal Market, Industry, Entrepreneurship and SMEs, 2026. Regulator.
Used only to establish that an INCI name refers to a recognised cosmetic ingredient. Presence in CosIng does not mean an ingredient is approved, safe, or non-comedogenic, and this dataset never treats an identity citation as evidence about comedogenicity.
Limitations of this site itself
- This dataset has not been reviewed by a clinician. It was compiled from sources whose bibliographic details were individually verified, but no dermatologist or cosmetic chemist has signed off on it.
- The corpus is small — 9 sources across 69 ingredient records. An absence of evidence here frequently reflects the size of the corpus rather than the state of the science.
- Where only an abstract could be checked, records state the family-level or study-level claim rather than inventing a per-ingredient rating. One source has no abstract available at all, and is cited only as evidence that a classification is contested.
- Matching is exact. There is no fuzzy matching, so an unlisted spelling or a supplier’s trade name is reported as unrecognised rather than guessed at.
- Nothing here is medical advice, and nothing here diagnoses anything.
Corrections
Every record carries a stable identifier, a review date and its sources, so a disputed entry can be pointed at precisely. Corrections are made by changing the dataset and its version, not by adjusting wording in the interface.
Reviewed by: pore-checker prototype dataset compilation (not clinically reviewed).