AI Video Ads for Skincare Brands: Cosmetics Claims Regulation Explained

UK skincare advertising operates under the Cosmetics Regulation 1223/2009 (retained in UK law), the Cosmetics Claims Regulation EU 655/2013, and CAP code section 12 on medicines, treatments, and health-related products. The intersection between these three frameworks is where most ASA rulings against AI-generated skincare ads sit. The product is regulated as a cosmetic. The advertising claim, if too strong, gets reclassified as medicinal. The AI tool does not know where the line is.

The DTC skincare brands shipping AI variants at sustained scale work to a defined claims register. Cosmetic-acceptable language is anchored to function (cleanses, moisturises, smooths the surface) rather than effect (treats, repairs, regenerates). The visual register is anchored to format (routine, ritual, application demonstration) rather than transformation (before/after, multi-week change, age reversal). The brief discipline encodes both anchors explicitly.

What follows is the cross-skincare-vertical framework, with category specifics covered in companion articles for serum, sunscreen, before-and-after format, and the rest of the segment.

The cosmetic vs medicinal claim line

A cosmetic, under retained EU law, is "any substance or mixture intended to be placed in contact with the external parts of the human body... with a view exclusively or mainly to cleaning them, perfuming them, changing their appearance, protecting them, keeping them in good condition or correcting body odours."

The functional verbs in that definition (clean, perfume, change appearance, protect, maintain, correct odour) are the cosmetic-acceptable register. Beyond that register, claims start to imply effect on physiological function or treatment of medical conditions, and the product becomes an unlicensed medicine under MHRA classification.

The line is well-mapped. "Moisturises" is acceptable. "Heals" is not. "Smooths the appearance of fine lines" is acceptable. "Reverses ageing" is not. "Helps protect against sun damage" is acceptable for an SPF product. "Treats sunburn" is not. The pattern across the entire skincare category is: appearance language acceptable, function language conditional, treatment language prohibited.

The Cosmetics Claims Regulation framework

EU 655/2013, retained in UK law, sets out six common criteria that all cosmetic claims have to satisfy:

• Legal compliance (no claims a cosmetic by definition cannot make).
• Truthfulness (no exaggeration or distortion of evidence).
• Evidential support (claim must be substantiated by adequate and verifiable evidence).
• Honesty (no implication of attributes the product does not have).
• Fairness (no denigration of competitors or unfair comparisons).
• Informed decision-making (claim should give consumers information they need).

The third point matters most for AI-generated skincare. The brand has to hold substantiation evidence for any claim made in the advertising, including in synthetic testimonials. The fact that the AI tool generated the script does not transfer the substantiation obligation. The brand still has to be able to back the claim if the ASA asks.

This intersects with the AI-generation question covered in AI before and after videos for skincare ASA compliant, where the visible-result claim is the most-cited trigger.

The ASA's specific stance on AI-generated skin transformation

The ASA has issued rulings on AI-generated skincare content with increasing frequency through 2025 and 2026. The pattern is consistent: synthetic transformations imply substantiation that the brand does not hold, because the transformation never occurred. A before/after generated through AI is not evidence of effect; it is a visual claim of effect, and a visual claim has to be substantiated like any other claim.

The position carries a clear operational implication. Skincare brands cannot use AI to generate before/after content that resembles a real result. They can use AI to generate content that demonstrates application, format, or routine. The line between the two is not always obvious to the AI tool, which is why the brief has to constrain the output explicitly: no transformation framing, no time-elapsed visual change, no claims about visible results.

Where AI tools default to over-claim

A vanilla skincare brief produces over-claim output reliably. The training data is heavy on US-market skincare content, where structure-function claims are looser and transformation framing is the category default. The model generates "smooths and tightens", "visibly reduces", "younger-looking skin within weeks" in the first sentence of the script.

The negative-constraint instruction in skincare needs three layers:

• Functional language only: cleanse, moisturise, smooth the appearance of, protect, soften.
• No transformation framing: no before/after, no time-elapsed reference, no "weeks" or "months" outcome wording.
• No medicinal language: no treat, heal, repair, regenerate, fix, eliminate.

With those three layers, model output enters the cosmetic envelope. Without them, the output is non-compliant on every pass.

The category-specific compliance brief patterns

Each skincare sub-category has its own compliance specifics, but the brief structure is consistent across them:

• Serum: substantiation rules around active-ingredient claims (retinol, vitamin C, peptides). Covered in AI testimonial videos for serum brands.
• Sunscreen: SPF substantiation, broad-spectrum claims, environmental claims. Covered in AI video ads for SPF and sunscreen brands.
• Before-and-after format: AI-specific risks documented in AI before and after videos for skincare ASA compliant.

Across all of them, the brief specifies talent profile, visual setting, product context, claim language anchored to the cosmetic-acceptable register, and explicit negative constraints on transformation and medicinal framings.

The hybrid AI + human creator pattern for skincare

Skincare DTC brands at scale run the same hybrid pattern as supplement brands. AI generates 80% to 90% of variant volume against a cosmetic-acceptable brief library. Human creators produce hero placements where authentic skin texture, lighting consistency across cuts, and emotional register justify the rate.

Hero placements in skincare have an additional consideration: the audience can sometimes detect AI-generated skin texture in close-up. Skin rendering is an area where AI models still produce subtle artefacts on close inspection, particularly under uneven lighting. Hero placements with high media spend behind them tend to use real talent for this reason, while AI handles the wider variant cycle where the resolution and viewing context are more forgiving.

For the per-model cinematography specifics, How to write AI video prompts for Veo 3.1 covers the brief structure that produces stronger skincare output.

FAQ

Can AI-generated skincare ads make any direct effect claims?

Yes, within the cosmetic-acceptable register. Functional claims around moisture, smoothing the appearance of fine lines, protecting against environmental factors, and improving feel are all acceptable. The line is at effect on physiological function or treatment of medical conditions.

What about anti-ageing claims?

"Reduces the appearance of fine lines" is acceptable as an appearance claim. "Reverses signs of ageing" implies a physiological effect and is a frequent ASA-trigger. The wording matters: appearance language is cosmetic, function language is borderline, transformation language is medicinal.

Are AI-generated before/after videos acceptable?

The ASA's working position is that synthetic before/after content implies substantiation the brand does not hold. The safer pattern is to use real-customer before/after content with clear documentation, or to avoid the format entirely for AI-generated variants. The category-specific guidance is in AI before and after videos for skincare ASA compliant.

Does substantiation evidence transfer between US and UK markets?

The substantiation evidence is jurisdiction-neutral; the wording requirements are jurisdiction-specific. Clinical trials supporting an active ingredient can support claims in both markets, but the language permissible in the US (structure-function claims) is wider than in the UK (cosmetic-acceptable register). Most brands shipping to both markets use UK wording as the baseline because it transfers cleanly.

How do brands handle the AI-skin-rendering issue at hero placements?

Most operate the hybrid pattern: AI for variant volume where the viewing context is forgiving, real talent for hero placements where the close-up rendering matters. The boundary is set by media spend; placements above £20,000 in concentrated spend usually justify real-talent production for the hero asset.

For the platform-aware tooling that handles UK skincare compliance specifically, see AI video tools that handle ASA compliance UK.

---

100 free credits to test how Tonic generates skincare-category briefs that hold inside the cosmetic-acceptable register: tonicstudio.ai/signup?promo=UGC100.