Imagine this: you open your laptop at 11 PM, you don't have access to your supplier's database, and the deadline for a new formulation is tomorrow morning. In the past, this meant hours buried in stacks of textbooks or desperate searches through forums. Today, some cosmetic chemists simply open ChatGPT—and get a starting formula in three minutes. This isn't magic, and it isn't a threat to the profession. It is a tool that is already reshaping the industry—and the sooner you understand how it works (and where it lies), the better it will be for your formulas.
What AI can do in cosmetic chemistry right now
Talk of artificial intelligence in the beauty industry often boils down to personalized recommendations on retailer websites. But that is just the tip of the iceberg. At the level of formula development, AI is already performing several specific functions—and some of them are truly impressive.
Generating starting formulations
Large language models—GPT-4, Claude, Gemini—are trained on massive amounts of text, including patents, scientific articles, ingredient technical data sheets, and supplier formularies. This means that if you ask the model to suggest a basic recipe for a moisturizing shampoo, it will output something like: Sodium Laureth Sulfate 10–12%, Cocamidopropyl Betaine 3–5%, Sodium Chloride for viscosity adjustment, panthenol 0.5%, fragrance, preservative—and this will be a reasonable starting point. Not perfect, but a functional base from which you can move forward.
Specialized AI tools for cosmetics—such as Mindsync or GPT-based platforms with cosmetic datasets—go further: they can account for ingredient compatibility, pH ranges, and even regional regulatory restrictions. To understand why pH is so critical in any formula, I recommend reading our guide to pH in cosmetics—without this foundation, even the smartest neural network won't save you from an unstable formula.
Searching for ingredients and interactions
One of the most underrated use cases for AI is a quick search for ingredient interactions. Are you adding a new cationic conditioning agent to a formula and want to ensure it doesn't conflict with anionic thickeners? Previously, this required reading five technical data sheets. Now, it takes one prompt in GPT with a specification: "explain the interaction between Polyquaternium-10 and Carbomer in a shampoo with a pH of 5.5."
Of course, the answer must be verified. But as a starting point for research, it saves an hour of work. Especially when it comes to complex systems: for example, how gelling agents and thickeners behave under stress—more on this in our article about tribology, gums, and gelling agents.
Where AI frankly still falls short
This is where we start an honest conversation. AI is a probabilistic tool. It doesn't "know" chemistry in the way an experienced formulator does. It predicts the most likely next token in a text, relying on patterns in its training data. And this creates several systemic problems.
Hallucinations and outdated data
Models regularly "hallucinate"—they provide non-existent ingredients, incorrect concentrations, or outdated regulatory data with complete confidence. Ask ChatGPT about the maximum permitted concentration of a certain UV filter in the EU, and it might give you a figure that was relevant three years ago, before the latest amendments to Annex VI of Regulation 1223/2009. This is no small matter: it is a potential product recall.
The second painful point is the texture and aesthetics of a cream formula. AI doesn't feel how a cream sits on the skin. It doesn't know that your clients hate a "tacky" finish or that a specific batch of shea butter from Burkina Faso behaves differently than West African shea—even though climate really does affect the composition of plant oils, and that is a nuance that lives in the hands of the formulator, not in a dataset.
Lack of sensory expertise
Cosmetics are, to a huge extent, a sensory experience. Shampoo foam, the slip of a conditioner on wet hair, the feeling after rinsing—all of this is impossible to fully digitize into a training dataset. AI might suggest adding 1% Behentrimonium Chloride to improve detangling—and that is technically correct. But it won't tell you that in your specific system with a high cetyl alcohol content, this will result in an undesirable "waxy" tactile feel. That knowledge comes only from practice.
Personalization: where AI is truly changing the game
The most revolutionary application of AI in cosmetics is not generating cream formulas, but personalization on an industrial scale. This is something that was previously physically impossible.
Consumer data analysis
Companies like Prose or Function of Beauty use machine learning algorithms to analyse hundreds of parameters: hair porosity, water hardness in the region of residence, scalp type, color treatment, and climate. The result is a personalized formulation that is technically different from your neighbor's. This is not a marketing gimmick: it is backed by real variations in protein concentrations (for example, Hydrolyzed Keratin from 0.5% to 3%), types of silicones (Amodimethicone vs Dimethicone), and the ratio of moisturizing agents.
For a home-based formulator or a small brand, this level is currently unattainable, but it is important to understand where the industry is heading right now. Especially if you are just starting your journey — take a look at the about our school page, where you can get an idea of which skills will be in demand in the coming years.
Stability Prediction
Several startups — Evonik with its ACTIWAVE platform, Givaudan with AI tools for perfumery, and Unilever with its internal developments — are already using ML models to predict the stability of emulsions before real-world testing begins. The algorithm is trained on thousands of historical tests and can predict with a certain probability whether your emulsion will separate after 6 months at 40°C. This is not a replacement for real stability tests, but it is a way to filter out obviously failing combinations at the computer modeling stage.
Practical Guide: How to Use AI in Your Formulation Work Right Now
If you formulate — professionally or as a hobby — here are specific scenarios where AI really helps, and where caution is needed.
What You Should Entrust to a Neural Network
- Generating a starting formula. Ask ChatGPT or Claude to suggest a base — and use it as a draft, not as a final recipe. Specify: product type, target pH, desired viscosity, and ingredient restrictions.
- Explaining mechanisms of action. "Explain how Polyquaternium-7 adsorbs onto the surface of damaged hair" — this is an excellent prompt. The model will provide a clear explanation that you can verify against primary sources.
- Searching for INCI synonyms. When you need to find trade names for a specific INCI name or vice versa — AI handles this quickly.
- Brainstorming alternatives. "What can replace DMDM Hydantoin in this system while maintaining a broad spectrum of activity?" — this is a good question for generating ideas that you must then verify.
- Help with product descriptions. Copywriting for cosmetics is another strong suit of language models.
Where Human Expertise Is Required
- Final regulatory compliance check. Always check the current versions of regulations (EU 1223/2009, FDA, EAEU TR CU 009/2011) directly.
- Evaluation of texture and sensory properties. This requires hands, a nose, and skin — no algorithm can replace that.
- Stability testing. Real tests at 40°C, -10°C, and under freeze-thaw cycling conditions are mandatory. AI predictions are only an auxiliary tool.
- Working with new or rare ingredients. If an ingredient has appeared recently, it is almost non-existent in the model's training data — here, you need technical data sheets and direct contact with the supplier.
Ethics and the future: AI as a co-author, not a replacement
One of the most common fears I hear from aspiring cosmetic chemists is: "Why learn to formulate if AI will do it for me?" That is roughly like asking in the year 2000: "Why learn photography if there are digital cameras?" Technology changes the tools, but it does not eliminate the need to understand what you are doing.
AI does not bear responsibility for product safety. It does not know that a specific batch of jojoba oil from your supplier has a non-standard fatty acid profile. It does not feel that a cream formula is "almost right," but that something is off — that intuitive knowledge that comes after hundreds of experiments. That is precisely why the path from curiosity to professional formulation still requires real training — we wrote about this in the article how to become a cosmetic chemist.
The future that is already arriving is a hybrid model. AI takes on the routine: searching, generating options, documentation, and preliminary compatibility analysis. The human takes on judgment, sensory evaluation, ethics, and the final decision. Those who master both languages — chemistry and algorithms — will be in the strongest position.
If you want to build exactly that foundation — to understand chemistry well enough to be able to critically evaluate what any tool, including a neural network, suggests — take a look at the Walker Formulation Academy Club. It is a community of practitioners who analyse real cases, not theory in a vacuum.
Can you fully trust ChatGPT when developing a shampoo or hair conditioner formula?
No — and this is not a flaw of the tool, but its nature. ChatGPT generates probabilistic responses based on patterns in its training data. The starting formula it suggests may be a reasonable starting point, but it requires verification across several parameters: the relevance of regulatory restrictions, the compatibility of specific ingredient brands (not just INCI), and stability in your production system. Treat an AI-generated formula like a draft from an intern with a good theoretical background — it is useful, but it requires expert verification.
Which AI tools are cosmetic chemists actually using in 2024–2025?
Among publicly available tools, ChatGPT (GPT-4o) and Claude 3.5 are used for generating formulas, explaining mechanisms, and searching for ingredients. Among specialized tools, there are RAG-based (Retrieval-Augmented Generation) platforms trained on cosmetic datasets: some raw material suppliers (e.g., Evonik, BASF) have begun integrating AI assistants into their portals for technologists. For stability prediction and molecular modeling, more specialized tools like Schrödinger are used, but this is at the level of large R&D laboratories. For home crafters and small brands, GPT plus critical thinking is currently sufficient.
Will AI change the educational requirements for a cosmetic chemist?
It is more likely to shift the focus rather than eliminate the need for education. A basic understanding of emulsion chemistry, surfactants, pH balance, and stability will remain fundamental—precisely because it is impossible to evaluate the quality of what the algorithm suggests without it. However, the value of data skills, the ability to formulate precise queries (prompt engineering), and the critical evaluation of AI outputs will increase. A chemist who possesses both skill sets will be significantly more productive.
