Why AI Is Not a Replacement for a Cosmetic Chemist
Sometimes I would love to put my feet up on the desk all day and let AI run my life. But, here is why I don’t think that’s going to happen right now.
In an age where algorithms write ad copy, AI tools draft formula suggestions, and entire skincare lines claim to be “AI-developed,” it’s tempting to believe that machines can replace the human minds behind beauty innovation. But here’s the truth, artificial intelligence is a tool, not a formulator. And when it comes to the actual craft of cosmetic formulation, AI is missing something critical. It’s missing communication, context, nuance, and lived scientific judgment.
What you ask of AI and how you ask it, determines the quality of the answers you get.
In many cases, if your input is vague, overly simplified, or rooted in faulty assumptions, AI will confidently return results that are either scientifically incomplete, functionally unusable, or just plain wrong. It seems to love agreeing with your preconceived notions. Mine anyway.
As cosmetic chemists, we don’t simply data input ingredients and spit out formulas. We interpret outcomes. We troubleshoot instability. We understand the why, not just the what. And that’s a kind of intelligence AI, Chat GPT, Perplexity and more just don’t have. We need to be adaptive, nuanced, cross-disciplinary and sometimes plain old practical. That is not a language model or algorithm that can be replicated.
Cosmetic formulation is a blend of chemistry, artistry, regulatory knowledge, sensory interpretation, experience, visual, olfactory and human psychology. AI tools trained on INCI lists, supplier data, and public-facing research can certainly simulate the structure of a formula. They can suggest surfactants for a given pH, preservative systems that match a target market, or emollient blends for a certain feel. But it can’t do it with critical thinking. And it’s not great and producing a definitive outcome.
AI Generated Cosmetic Formulations
You will also get a surface level interaction, it’s just regurgitating what already exists. And also what has been revealed. Cosmetic chemistry is notoriously protective of it’s trade secrets. There are lots of NDA’s, patents, and complex subjects that just don’t allow for an easy, reliable online tool to make your cosmetics.
A cosmetic formulator is thinking now and down the road. We are thinking of every situation this product might be in. From a hot shipping container to a customers -15C doorstep in December. We think about UV exposer, smell, colour and separation 6 months after opening. We are thinking of cost, scale up and packaging. It’s very dimensional. Artificial Intelligence does not lay awake at night thinking about your successful launch or rooting for your first sale.
These human interactions add safety to your business. They add joy to the journey. We make judgment calls, not simple data retrieval tasks. And we rely on real life knowledge and the things we learn from years in the lab, from failed prototypes, from understanding what consumers say vs. what they actually want. Our heads are filled with years of feed back and knowledge from a very nuanced job.
Another major flaw in the AI formulation narrative is the illusion of accuracy. I mean this in several ways. If your formulation totals 100 and you get wacky wild ranges to guess from, nothing is going to turn out. Not on paper and not in life.
Apps like Potion, Good Formulations, Chat GPT or any AI tool will also give you ranges for raw materials. These ranges can vary wildly. 30%-80% water? Thats a big guess. 10%-35% oil? That will change the whole formula. It can guide, but a chemist must do the work. And I believe that’s who a lot of these online apps are geared towards.
Obviously without hands and eyes its not going to have great feed back on colour, gloss, textural appearance etc. Can it choose the right mixing tool? High sheer or stir? 3 bladed propeller or dissolving stirrer? It has no idea. Will it do your freeze/thaw? Not at this point in time. It won’t tell you it looks like swamp water either (or smell like quats). You need a person.
Most people using AI tools as a cosmetic formulation engine, or a trend forecasting tool don’t realize that what you ask matters more than what you get.
Let’s say a brand founder types: “Create a natural anti-aging serum with niacinamide, peptides, and no preservatives.” Sounds simple, right? But this question is packed with contradictions. Niacinamide and peptides are notoriously tricky to stabilize together, especially in natural systems. “No preservatives” implies a microbiological risk unless the system is dry, anhydrous, or heavily glycol based. “Natural” is undefined and does that mean COSMOS compliant? Vegan? Free from petrochemicals? I know, do you?
AI won’t challenge your assumptions. It’ll generate a lovely sounding formula that ticks boxes on paper and that might lead the brand to think it’s viable, stable, and compliant when it’s none of those things. AI has CONFIDENCE! And it’s also quite a sweet talker. Even when I’m wrong it says ‘great point Annie”.
An actual chemist has likely experienced a LOT. And can walk you through the risks, alternatives and expectations from the start. I think a good one will also tell you if you’re idea is worth pursuing and how you can maximize it.
Sometimes I have a lot of intuition about cosmetic formulations. This comes from the communication with the client in certain instances. Like when they tell me “it needs more depth”. Depending on the client this could mean viscosity, fragrance, texture or something else. Without the relationships we develop with our clients, this can be a difficult one to figure out.
My intuition can also tell me “I don’t think this is going to hold together”. Sometimes you do everything right and a fragrance sheers the formula 24 hours later. Or the high oil content won’t stay in a lotion passed 2 weeks. Only experience will give you these predictors and help you choose the best fix.
A Cosmetic Chemist will Check and Double Check. Then Probably Triple Check.
Formulating skincare isn’t just mixing ingredients. It’s managing risk. What happens if the preservative system doesn’t cover gram-negative bacteria? I’m going to test for that before you go anywhere with this product. The client’s packaging choice leaches into the oil phase? I’m going to test that too. The final pH doesn’t match the supplier’s ideal range? Then you need some hands on chemists to help correct this.
AI doesn’t know. It doesn’t feel responsible. It just pulls data from its training set and serves it up confidently. Even when the confidence is misplaced. It’s still very cheery and seemingly on your side.
This is especially dangerous in formulation work, where aesthetics, safety, and shelf-life must be dialed in through careful trial, experience, and scientific rigor. Also your name is and brand is the public face. If something goes wrong out there, you have little support from AI.
AI and Cosmetic Formulations Apps are Simple Minded
Ask AI to generate a peptide serum, and you’ll get a nice, safe combo of Matrixyl, Argireline, and niacinamide, probably with a hyaluronic base. Fine. But not breakthrough. In fact it’s kinda boring. AI cosmetic formulation can not have the creativity of a good chemist and insight the client needs.
Ask a seasoned formulator the same thing, and we’ll ask: What delivery system are we using? Do we want a rapid effect or long term repair? Should we pair peptides with skin lipid support or enzymatic protection? We also want to know where you will be selling, how can we position your products to fit. What are the buzzy, upcoming forecasts for your category? Although AI, Chat GPT and Instagram may very well be listening to me talk on the Cosmoprof floor.
Then we build something novel and cool. A lamellar emulsion with a new neuropeptide and copper complex that supports fibroblast activity while improving skin barrier health. Not just functional. Memorable. New and exciting. Maybe a product that ties into your heritage, or gives a nod to a generation, or takes the lead from a brand you’ve already developed.
That kind of thinking is more lateral, iterative, tactile. It does not come from an AI or Chat GPT prompt. It comes from the mind of a chemist who understands not just the science, but the market, the skin, and your story.
My Experiences with AI Formulation
I’ve used AI as a tool myself. But honestly I’ve found more mistakes then I’ve had epiphanies. And when I’ve called them out it cheerfully says “good catch Annie”. So it’s still making me feel good and coming back for more lol. Or maybe I’m just teaching it for the next cosmetic chemist who asks the same question.
It’s useful for trend mapping, ingredient discovery, cross-referencing allergens and claims, generating rough concept drafts, and speeding up research for experienced chemists. However, it has not earned my full trust.
APL Experience with AI Formulation
Atomic Pom Labs has also seen an uptick in people sending us AI formulations for production. I know I’m not alone in this when I say, we can tell. Like, at first glance. It’s a very bad forging of your moms signature to get out of gym class. The formulas AI creates in comparison to what a lab needs for manufacturing are very very different. We are happy to assist a client with their vision and drafts, but you are really not saving money or discovering a short cut doing this. In fact it is harder to reformulate, reverse engineer or sort your mess out then it is to do it right from the beginning. Add those hours up and you’ve saved nothing.
When a contract manufacturer (CM) takes on a new project they will want a benchmark (sample of the product), formulation, phases, suppliers, testing etc. That can only be provided by your chemist.
I think AI in the cosmetic formulation space is helpful. Mostly helpful to a cosmetic chemist. Although I think it helps brands understand their product better. But we are in early stages and brands that want new and innovative will not find it in old, ‘been there done that” internet information. No matter how sure it is of itself.