AI in Cosmetics 2026: 138 Verified Use Cases, and Where a Manufacturer Fits

Brands increasingly arrive at a manufacturer with a concept that took an afternoon to generate: a positioning, a hero active, sometimes a candidate formula sketch. That shift is real, and it is measurable. What is less often discussed is what it changes downstream — at the factory, where an idea has to become a batch that can be repeated, documented and legally sold.

This article summarises the most comprehensive public inventory of cosmetics AI we have seen to date, then looks at the part of the chain it points to next.

The evidence base: 138 verified cases, graded by source

The study — 2026 全球化妆品行业人工智能应用全景研究报告 ("2026 Global Cosmetics Industry AI Application Panorama"), published in Chinese on 18 September 2026 — collected and verified public disclosures from cosmetics groups, brands, ODM/OEM manufacturers, beauty retailers and fragrance/ingredient suppliers. As of its 18 September 2026 cut-off it records:

• 138 publicly verifiable AI cases
• 47 companies and groups, 62 brands or business units
• 21 distinct AI application categories with at least one real case
• 92.0% of the retained sample backed by high-confidence evidence after weak entries were removed

The methodology is the interesting part, and it is stricter than most "AI in beauty" roundups. A project only entered the main database if a company website, annual report, regulatory filing, paper or technology partner explicitly evidenced the use of artificial intelligence, machine learning, deep learning, generative AI, an LLM, computer vision or an AI agent. Projects that could only demonstrate AR, digitisation or "smart devices" — without evidence that AI sat behind the core algorithm — were excluded. Sources were then graded A+ (listed-company reports, ESG filings, exchange or regulatory disclosure), A (official websites, newsrooms), B (cross-verified company and technology-partner material) and C (papers, government sites, conference and mainstream media coverage). A and A+ together account for 124 of the cases, or 89.9%.

In other words: roughly one in thirteen "AI beauty" projects the authors looked at did not survive evidence review.

Three phases: seeing, understanding, creating

The study's central finding is a shift in where AI is applied.

Phase	Problem it solves	Typical applications
First	Seeing the consumer	Skin analysis, virtual try-on, shade matching
Second	Understanding the consumer	Recommendation, personalisation, beauty agents
Third — the one the study flags as most significant since 2024	Creating products	Discovering raw materials, designing molecules, generating formulations, predicting safety, laboratory robotics, smart manufacturing

The authors' framing is worth borrowing for sourcing decisions: the competitive question for a cosmetics company is no longer "do you use AI", but "has AI entered the system that does your scientific discovery and product creation".

Two named examples anchor that third phase. L'Oréal and IBM appear as an attempt to build a formulation foundation model — moving AI from retrieving formulas to generating them. On the measurement side, skin analysis is described as upgrading from reading photographs to reading biology, illustrated by FANCL's SKIN PATCH approach of reading corneocytes directly.

The study also dates the acceleration: 2025 is identified as the clear inflection point, driven by two forces — generative AI collapsing the deployment barrier, and a decade of enterprise digitisation starting to compound ("AI dividends" from data that already existed).

What this changes for anyone outsourcing production

If AI is heading into the creation side, the practical consequence for a brand is asymmetric — and mostly good news, with one caveat.

What gets easier: concept generation, market-fit hypotheses, ingredient shortlists, and the first draft of a formulation direction. Work that used to consume weeks of desk research is now an afternoon.

What does not get easier: everything that turns a formulation direction into a product on a shelf — pilot batching, stability and microbiological testing, packaging compatibility, fill-weight consistency, batch documentation, and the claim set the destination market will accept. Those depend on equipment, qualification and record-keeping, not on model output.

That is why a brand arriving with an AI-assisted brief usually finds the same bottleneck in the same place. The bottleneck moved from "do I have a formula idea?" to "can someone produce this repeatably, and can the paperwork survive an audit?"

The production side, in numbers you can check

The checklist above is only useful if the numbers behind it are public. These are ours — every figure here is either on a certificate with a number, or on a product page with a stated minimum order.

What	Detail
Formats we run in-house	Facial masks (hydrogel, jelly, crystal, bio-cellulose, clay) · eye patches and under-eye gels · forehead / neck / lip patches · serums and essences · creams and lotions · toner pads · foot and body masks · hair masks and scalp oils
Minimum order	500 units per SKU (private label on an existing formula) · 1,000 units (standard catalogue) · 3,000+ units (custom formulation)
Sampling	Branded samples produced in 15–20 business days
Bulk production	30–45 days after sample approval; 45–60 days from brief to shipment on standard projects
Formulation capacity	12 formulation scientists, 90+ mask formulas, 200+ active SKUs
Plant	3,886 m² GMP facility in Zhongshan, Guangdong · 8 production lines · in-house microbiology laboratory
Certificates with numbers	GMPC — SGS-issued ISO 22716:2007 (CN19/31476) and US FDA CFSAN Cosmetic GMP (CN19/31477); FDA facility registration FEI 3015316073 (MoCRA); NMPA cosmetics production licence 粤妆20160335; amfori BSCI social audit (25-0314051-1)
What we do not claim	Medical or device claims, and anything outside the certified production scope. Cosmetic claim wording for your market stays your regulatory advisor's call — we supply the technical file behind it

Two of those rows are the ones AI-assisted briefs hit first: which format is actually on the lines, and what minimum applies to the lane you need. Browse the formats with their real specifications on the product catalogue, or check every certificate number on the certifications page before you talk to anyone.

Have a concept that needs producing? Send it over — format, target market and first order quantity — and you will get formula options, MOQ, lead time and the documentation list back within 48 hours. Send your brief · Request a sample · Get an instant estimate

Six things to check in a factory when you arrive with an AI-assisted concept

1. Who owns the formula, and can it be produced at your volume. A generated direction still has to be either matched to an existing proven formula or developed properly — and the two have different minimums. In our own case, private label on an existing formula starts at 500 units per SKU, standard catalogue production at 1,000 units, and a genuinely new custom formulation at 3,000+ units per SKU.
2. Whether the format is actually on their lines. A formulation is only half the product; the other half is a hydrogel cast, a jelly sheet, a filled sachet, a patch die-cut or a tube. Ask which formats the factory runs itself and which are subcontracted.
3. How sampling works, and how long it takes. We produce branded samples in 15–20 business days, then run bulk 30–45 days after sample approval, with 45–60 days from brief to shipment on standard projects. A supplier quoting a week is usually quoting a different product.
4. What testing runs per batch, and by whom. Stability and microbiological testing on every batch, with COA, MSDS and the INCI list issued per SKU, is the baseline — not a premium option.
5. Which certificates exist, and whether you can verify them. Numbers matter more than logos. Ours include SGS-issued ISO 22716:2007 (certificate CN19/31476) and US FDA CFSAN Cosmetic GMP (CN19/31477), FDA facility registration FEI 3015316073 under MoCRA, the NMPA cosmetics production licence 粤妆20160335 and an amfori BSCI social audit — all published so a buyer can check them with the issuing body. For EU-bound products, the safety assessor still needs the technical file behind your CPSR, which is a factory document, not a model output.
6. What the supplier will not claim. This is the tell. AI can help draft a claim, but regulated claim language — anything implying a physiological effect rather than a cosmetic one — has to clear the destination market. A manufacturer that tells you where its claim boundary sits is easier to work with than one that agrees with everything.

A note on this article's own sources

The statistics above come from the Chinese-language study cited at the top, in its 18 September 2026 edition; we did not contribute to it and its case counts are not claims about our own capabilities. The manufacturing figures used in the checklist come from our own published specifications and certificates. Where the two meet is the point of the piece: AI is visibly entering how products are conceived, while the part that decides whether a brand can actually ship — qualification, consistency and documentation — still sits in the factory.

If you arrive with an AI-assisted concept and need it turned into a specification that can be produced and documented, send the brief: format, target market and first order quantity. Formula options, MOQ, lead time and the documentation list come back within 48 hours.

Related reading:

• Cosmetics R&D Toolchain 2026: Essential Platforms for Formulation, Compliance & Market Research
• How to Verify a Chinese Cosmetics Manufacturer in 2026
• Science-Backed Skincare 2026: Product Development Guide
• Private Label MOQ, Pricing and Compliance Guide