The Allure of a Shortcut

AI is fast, confident, and available 24/7. Ask it to “formulate a brightening serum,” and in seconds it’ll give you a clean, professional-looking list of ingredients, complete with percentages.

It looks convincing. The language is polished. It even sounds scientific.

But when you try again, slightly rephrase the question, and ask it to “formulate a brightening serum with niacinamide,” you’ll get something completely different.

Different actives, different ratios, sometimes even ingredients that can’t coexist in the same phase.
That’s the problem. AI doesn’t understand formulation — it predicts text.

It’s not malicious. It’s not trying to trick you. It’s simply filling in blanks with probability, not precision.

Why AI Outputs Are Inconsistent

AI doesn’t have a “memory” of truth — it has patterns of language. So if 60% of online skincare formulas use glycerin at 5%, and 40% use it at 3%, AI might return either number at random. It doesn’t know which one actually works — because it’s never seen a stability test, a viscosity trial, or an emulsion fail in real life.

That’s what formulators do: they test, tweak, and interpret data.

AI just rearranges what it’s read. It can’t tell the difference between a peer-reviewed formula, a Reddit DIY post, or a mislabelled ingredient sheet.

That’s where hallucination creeps in — the confident presentation of information that looks plausible, but simply isn’t true.

But even beyond the data, formulation is sensory. It’s how a cleanser foams, how a serum absorbs, how a conditioner feels in the hair after rinsing. AI can’t tell if a product feels sticky, too oily, or not rich enough — or notice when a texture suddenly shifts during mixing because two ingredients aren’t playing well together. Chemists use their eyes, hands, and intuition as much as their instruments. Those micro-observations — the subtle colour change, the viscosity drop, the emulsion that just doesn’t look right — are what keep formulas safe, stable, and consistent. It’s a layer of judgment that can’t be coded, only learned.

The Problem with Hallucinated Confidence

If AI were hesitant, it would be harmless. But it’s not. It’s confident. It will happily write “safe for sensitive skin” beside an untested active, or list an ingredient that’s restricted under the AICIS schedule in Australia.

It can’t understand toxicology data, dermal limits, or international restrictions.
It also doesn’t know about cost, packaging compatibility, or sensory experience — the things that define real-world success.

AI-generated formulas look legitimate because they use real chemistry words.
But they lack the experience layer — the part that knows what happens next. A good formula isn’t just what’s on paper — it’s what survives mixing, scaling, and six months on a shelf.

The Hidden Chain of Real Formulation

Let’s unpack what happens between an idea and a bottle that sells:

Concept → What’s the problem we’re solving?

Formulation Design → Balancing actives, phases, and sensory goals.

Stability Testing → Does it separate, oxidise, or grow microbes?

Compliance Validation → AICIS registration, safety data, label accuracy.

Scale-Up → Does it behave the same in a 500 kg batch?

Cost Optimisation → Is it viable to manufacture and sell at scale?

AI doesn’t know these steps exist — it just outputs step two. A complete product is chemistry, regulation, economics, and logistics working together. A chatbot can help you brainstorm. But it can’t close that loop.

The Myth of “Just Take It to a Manufacturer”

Many founders believe they can ask ChatGPT for a formula, then give it to a contract manufacturer. In reality, that manufacturer will almost certainly reject it.

Here’s why:

• The ratios don’t balance.

• The preservative system is incomplete or unsafe.

• The actives exceed recommended usage limits.

• The ingredient list doesn’t match any known INCI standard.

• There’s no documentation, testing data, or proof of performance.

• And most importantly, any good manufacturer will need to revalidate a formula anyways... 

A manufacturer’s role is to produce, not reverse-engineer or fix an untested formula.
So while the AI draft might look like progress, it should be treated as a guide and point of discussion - not proprietary IP or a formulation that's ready for manufacturing. 

Where AI Can Add Value

Used properly, AI can be an incredible partner in product development and helping arrange your thoughts so you can get started. In short, It can:

✅ Summarise ingredient trends or consumer insights.
✅ Help translate INCI names.
✅ Inspire brainstorming during concept ideation.

What it can’t do is substitute verification. Every formulation must still undergo:

❌ Stability and preservative efficacy testing.
❌ Regulatory review for compliance and safety.
❌ Costing and scale-up validation.
❌ Commercial viability. 

AI can suggest, but it can’t validate.

The Responsible Use of AI in Formulation

The future isn’t about banning AI — it’s about understanding its place. The chemist of tomorrow might use AI as an analytical co-pilot to help model viscosity curves, cross-check ingredient interactions, or identify supply chain risks. But that chemist will still know that data doesn’t equal proof. Science always requires testing, and safety always requires accountability.

AI can make us faster. It just can’t make us right on its own.

The Bigger Truth

Formulation isn’t language — it’s outcome. It’s measurable, testable, physical. AI speaks confidently, but it doesn’t know — not in the human, hands-on sense of the word.
It hasn’t seen a batch split, watched a phase invert, or dealt with a preservative challenge test that fails on week eight. 

That’s what formulators bring: judgment built on evidence. AI can get you to ideas. But chemistry takes you to products.

Final Takeaway

AI isn’t a threat. It’s a tool. A remarkable one — but only when guided by expertise. Use it to research, plan, and ideate. But when it comes to formulation, trust data that’s been tested — not text that’s been generated.

-

References 

AICIS (2024) — Cosmetic Ingredient Regulation Overview
TGA (2024) — Cosmetic vs Therapeutic Product Guidance
Cosmetics Business (2025) — “AI and Formulation Integrity”
Mintel (2024) — “AI in Beauty Development: Potential & Pitfalls”