Not medical advice. This is our personal experience and a decision-making method, not a treatment recommendation. Seborrheic dermatitis is a chronic condition that is managed, not cured, and everyone's skin reacts differently. Patch-test anything new, and see a dermatologist for a real diagnosis.
For months, seborrheic dermatitis was flaring across our skin — the classic pattern: red, flaky, faintly greasy patches around the nose, the eyebrows, the hairline. If you have dealt with it, you know the particular frustration: your skin is simultaneously irritated and in desperate need of moisture, and the wrong moisturizer does not just fail to help — it feeds the flare and makes everything worse.
That is the trap. You need to moisturize. But seborrheic dermatitis is linked to Malassezia, a yeast that lives on everyone's skin and, in susceptible people, drives the inflammation. And Malassezia eats certain lipids — many of the very oils, fatty acids, and esters that moisturizers are built from. So "just moisturize" quietly becomes "moisturize without feeding the thing that is inflaming your skin," and suddenly you are squinting at a thirty-ingredient INCI list trying to guess which esters are a problem.
We had narrowed it to two contenders, both beloved for sensitive skin: Illiyoon Ato Concentration Cream (a cult-favorite Korean ceramide cream) and Vanicream Moisturizing Cream (the famously bare-bones, fragrance-free American emollient). We could not tell, by looking, which was the safer bet for our particular problem. So we did not look. We asked Claude Code to.
Using an LLM as an ingredient-risk analyst
Here is the method, because the method is the part you can reuse for any reactive-skin decision — it matters more than which cream we landed on.
We pasted the full ingredient list (the INCI list) of each product into Claude Code and asked it to evaluate every single ingredient against the things that matter for seborrheic-dermatitis-prone skin, and to tally the result. The prompt was roughly:
For each ingredient in this list, flag whether it is a known risk for seborrheic-dermatitis / Malassezia-prone skin. Check four categories: (1) fragrance or essential oils, (2) fatty acids and esters in the C11–C24 range that Malassezia can metabolize, (3) common contact irritants, (4) heavy comedogenic occlusives. Give each ingredient a short verdict and a risk level, then total it up.
Doing this by hand is miserable and error-prone; you have to remember which ester chain lengths feed the yeast, recognize fragrance hidden under names like "parfum," and keep it all straight across two long lists. Doing it as a structured pass — same four questions applied to every ingredient — is exactly the kind of tedious, rule-based work a language model is genuinely good at.
What it found
The two creams came back with clearly different risk profiles.
Vanicream was the shorter, plainer list: no fragrance, no essential oils, no botanical extracts, and only a couple of ingredients worth a second look. Its whole design philosophy is subtraction — it exists specifically to not contain the things that provoke sensitive skin. Fewer ingredients simply meant fewer variables that could go wrong.
Illiyoon Ato is a lovely, more sophisticated formula — ceramides, a richer emollient system, more botanical support. But "more sophisticated" cut against us here: more ingredients meant more plant-derived oils and esters, and therefore more surface area for something to feed the flare. Nothing in it is bad — for most people it is a fantastic cream — it was simply the higher-variance choice for our specific, yeast-driven problem.
The verdict was not "Illiyoon is bad, Vanicream is good." It was a risk comparison: for skin where the failure mode is feeding Malassezia, the minimalist formula was the lower-risk bet. So that is the one we tried.
The one caveat that matters: verify the model
Before we switched, we did something you should always do with an AI's confident-sounding analysis: we checked its work. Language models are strong at applying a consistent rule across a list, but they can be wrong on a specific fact — misjudging an ester's chain length, or mislabeling an ingredient's function. So we cross-referenced the flagged ingredients against a dedicated skincare-ingredient checker (there are free "fungal-acne trigger" analyzers that encode the same Malassezia rules). The AI's job was to do the tedious first pass and explain the reasoning; a purpose-built database was the check on its facts. Treat the model as a fast analyst, not an oracle.
Did it work?
Yes — with honest caveats. Over the following few weeks on Vanicream, the flaring calmed down and our skin cleared up to a degree it had not in months. It was not overnight, and we want to be careful with the word "cure," because seborrheic dermatitis does not really get cured — it gets managed, and it can return. What we can say plainly is that swapping to the lower-risk moisturizer stopped our routine from actively working against us, and our skin has been dramatically better since.
The takeaway
The useful, transferable idea is not "buy Vanicream." Your skin is not our skin, and the right answer for you might be the Illiyoon cream, or neither. The transferable idea is the method: when you are choosing between products for reactive skin, an LLM can turn an unreadable wall of ingredients into a structured, side-by-side risk comparison in seconds — as long as you give it the right criteria to check and then verify its facts against a real database.
It is decision-support, not a diagnosis. But for the specific misery of standing in a pharmacy aisle unable to parse two INCI lists, it is a genuinely good use of the tool — and in our case, it pointed us at the thing that finally gave our skin a break.