EmpCo & AI Product Recommendations for Beauty Ecommerce

Beauty brands have spent years differentiating products through claims such as natural, organic, vegan, cruelty-free, recyclable and sustainable.

From 27 September 2026, Directive (EU) 2024/825 on Empowering Consumers for the Green Transition, usually shortened to EmpCo, changes the rules around some environmental claims and sustainability labels. It is sometimes described informally as the EU greenwashing directive.

For beauty brands, this matters because these attributes often sit inside the product proposition. They influence why a customer chooses one serum, cream or shampoo over another.

Not every claim in that list falls under EmpCo in the same way. The relevance depends on what is being claimed, the evidence behind it and how the brand communicates it. EmpCo is also separate from the proposed EU Green Claims Directive.

At the same time, customers are increasingly asking AI systems to help them compare products using many of these same criteria.

A shopper might ask:

‘Recommend a facial serum for mature skin that is:

• Vegan
• Cruelty-free
• Fragrance-free
• Available in Italy
• Supported by packaging claims I can verify’

The task is no longer only to persuade a customer. The product also needs to be understandable to a system trying to compare it against the conditions in the request.

A human may accept an attractive badge or a broad product promise. An AI system may look for product-level evidence across brand pages, retailer listings, certification sources, structured data and other accessible sources.

That creates a new commercial reality:

Between the beauty product and the customer, there is now another evaluator.

Key takeaways

• EmpCo changes the rules around some environmental claims and sustainability labels. For beauty brands, the issue is not only legal compliance. Environmental and ethical product attributes often influence the commercial proposition.
• AI systems are becoming part of beauty discovery and consideration. Shoppers increasingly ask them to filter products against several conditions at once, including product type, suitability, vegan status, cruelty-free status, ingredients, packaging and local availability.
• A claim can be true inside the business and still be difficult for an AI system to use. If the claim, product, evidence, SKU, certification and market availability are disconnected, the system may qualify the answer, express uncertainty or select a competitor instead.
• EmpCo does not create an automatic AI-recommendation advantage. It creates stronger pressure to publish specific, bounded and supportable information. This information can also make products easier for external systems to interpret.
• The priority is product-information governance, not more sustainability copy. Connect each important claim to its scope, evidence, relevant product variant, approved wording and distribution across product pages, retailer pages, marketplaces, product feeds and structured data.

From sustainability copy to recommendation evidence

Many of the claims used to differentiate beauty products also create an environmental impression: ‘clean’, ‘conscious’, ‘natural’, ‘responsible’ and ‘planet-friendly’.

Some terms describe a product preference. Others imply a measurable environmental benefit. The difference matters.

EmpCo strengthens consumer-protection rules on misleading environmental claims, sustainability labels and certain climate-neutrality claims based on offsetting. Read Directive (EU) 2024/825.

For e-commerce teams, the practical shift is straightforward:

Replace broad environmental impressions with specific, scoped and evidenced product facts.

This does not require brands to stop communicating progress. It requires them to say what changed, where it applies and how they know.

A product page should not force a customer, regulator or AI system to guess whether a claim applies to the formula, an ingredient, the bottle, the outer carton, a refill format or the company as a whole.

EmpCo is not a new rule for every beauty claim.

It is particularly relevant where brands make environmental claims or use sustainability labels. Claims about ingredients, product safety, efficacy, vegan status or cruelty-free status may sit under other EU rules, including the existing framework for cosmetic claims.

The commercial point remains broader. Shoppers and AI systems may use environmental, ethical and product-performance attributes together when comparing products. Each claim therefore needs the right evidence, the right scope and the right product connection.

Why AI product recommendations are affected

EmpCo does not regulate an AI system’s ranking algorithm in the abstract.

It regulates consumer-facing commercial practices. But AI systems increasingly retrieve, compare and summarise commercial information from product pages, retailer listings, feeds, packaging, FAQs, reviews, campaign content and third-party sources.

Brand claim → Product information → AI interpretation → Shopper decision

If the original claim is vague, later representations can become less reliable.

Consider the difference.

Illustrative examples: the percentages and claims show how specific wording can work. They are not findings about named products.

The middle column is an interpretation, not a guarantee. AI systems differ. They may draw on incomplete sources, apply their own language, omit caveats or decide not to recommend a product.

But the commercial implication remains practical: specific evidence gives AI systems less room to turn a narrow fact into a broad environmental conclusion.

This is also a question of how brand differentiation survives AI-mediated buying. A distinctive claim needs to retain its meaning when another system explains it.

AI systems evaluate several product conditions

A beauty product is rarely selected on one attribute alone.

A shopper asking for a product ‘for sensitive skin’ may also expect it to be fragrance-free, vegan, cruelty-free, appropriate for a particular age group, refillable, available locally and within a particular price range.

To produce a useful answer, an AI system may need to establish whether the product is the right category, whether it meets every requested attribute, whether each claim applies to the exact SKU or variant, whether a certification covers the relevant product or component, whether the item is available in the shopper’s market and what evidence supports the recommendation.

A brand can hold credible evidence for each attribute and still lose the recommendation if the evidence is scattered, contradictory, inaccessible or impossible for a system to connect to the correct product variant.

The new task is not simply to make claims.

It is to make product characteristics verifiable together.

Product identity + relevant properties + evidence + availability + market context = recommendation readiness

This does not mean every AI system will follow the same process or reach the same conclusion. AI systems use different sources, retrieval methods and confidence thresholds. No brand can guarantee a recommendation.

But a product becomes easier to consider when its relevant properties are clear, current, connected and supported by evidence.

When AI questions a product attribute

In one German-language test, I asked ChatGPT for the best baby cream for sensitive skin. Its displayed activity considered Weleda Calendula, then introduced uncertainty around fragrance before narrowing its shortlist. Weleda did not appear in the final eight-product list. The shopper had not asked for a fragrance-free product. The system introduced that consideration itself.

Fragrance is not an EmpCo environmental claim. This snapshot illustrates the wider selection problem: an AI system may introduce an attribute, look for supporting information and narrow a shortlist when its confidence is limited.

A single response does not establish a general ranking pattern, product quality or the reason for the final list. It does show why brands need to test how their attributes are represented in real buying situations.

A product can be present online, well known and commercially credible. Yet it may still be absent from an AI-generated shortlist if the system cannot confidently verify the combination of characteristics the shopper asked for.

The issue is not always product quality. It can be the relationship between a product claim, its evidence and the sources an external system can interpret.

Beauty claims with the highest risk

Beauty teams should begin with claims that compress several ideas into one attractive phrase.

‘Clean’ and ‘conscious’ beauty

‘Clean’ does not automatically make an environmental claim. But it often becomes a container for statements about ingredient safety, natural origin, ethical sourcing, biodegradability and reduced environmental impact.

Those are separate propositions.

A brand may substantiate an ingredient-exclusion policy. It may also hold evidence about natural-origin content. Neither point automatically supports a broader claim that the product is environmentally better.

Create separate fields, evidence and approved wording for:

• Ingredient exclusions
• Formula characteristics
• Natural-origin calculation
• Environmental attributes
• Packaging composition
• Certification status
• Corporate targets

This protects consumers from confusion. It also protects brands from their own shorthand.

Refill claims

Refill is a strong commercial story in beauty. It can become a weak claim when the comparison is unclear.

A refill claim should answer:

• What component is reused?
• What does the consumer need to keep?
• What is being reduced: material, weight, emissions or waste?
• What is the comparison baseline?
• Does the claim include or exclude caps, pumps, cartridges, sachets and delivery packaging?

‘Refillable’ can be a factual statement. ‘A more sustainable refill’ needs a defined basis.

Natural-origin stories

Natural origin, vegan status, cruelty-free commitments, ingredient sourcing and environmental impact are not interchangeable.

A beauty shopper may value all of them. But an AI system cannot responsibly treat them as the same attribute, and neither should a brand.

Use precise language:

‘Contains 95% ingredients of natural origin, calculated using ISO 16128.’

Avoid turning that fact into:

‘A lower-impact natural formula.’

The first statement can be checked. The second makes a wider environmental judgement.

Carbon and climate language

Product-level claims such as ‘carbon-neutral’, ‘climate-positive’ and ‘net-zero product’ are especially sensitive where they rely on offsetting outside the product’s value chain. The Directive addresses this category directly.

Beauty brands should separate:

• Measured reductions
• Reduction targets
• Residual emissions
• Compensation or offsetting activity
• Product-level and company-level claims

The stronger commercial message is not always the broadest one. A narrow, evidenced reduction can build more trust than a sweeping climate promise.

Evidence must travel across commerce surfaces

A product page is no longer the only commercial record that matters.

AI systems may encounter product information through a brand site, a retailer page, a marketplace listing, structured data, Merchant Center feeds, editorial coverage, certification databases, social content or commercial protocols used by emerging agents.

If information differs across these surfaces, the system faces a conflict. If evidence appears only in a PDF, an image or an isolated campaign page, it may be difficult to retrieve or connect to the relevant product.

For brands preparing for agentic commerce, claim approval is only the first step.

The practical work is to keep every approved attribute, its evidence and its commercial context connected to the correct product and variant wherever that product is published or made available for transaction.

For example, a verified refill claim becomes less useful if the product feed points to a non-refillable SKU, a marketplace uses outdated packaging imagery or the product is unavailable in the customer’s market.

The information needs to remain coherent through discovery, comparison, recommendation and transaction.

Google documents product structured data, merchant listings and product variants. These Google surfaces are useful examples of why product identity, offer and variant information need to remain consistent. They do not establish how every AI system selects products.

This is why an evidence record cannot stay with legal or sustainability teams. It needs to flow through every system publishing, syndicating and updating commercial information.

What beauty e-commerce teams need to do

The six steps below describe the full process.

To begin, choose one priority product range. Check the claims on its product pages, main retailer listings and marketplace records. Then test whether AI systems can represent those attributes accurately in the buying situations your customers use.

Start with one bounded category. Learn from the evidence before expanding the programme.

Marketing teams should treat every claim as a retrievable product attribute, not only as conversion copy.

1. Audit claims where shoppers see them

Start with high-traffic and high-conversion surfaces:

• Product detail pages
• Collection pages
• Search filters and product badges
• Product imagery and packaging zooms
• Retailer product feeds
• Paid ads and affiliate copy
• CRM journeys
• Social templates
• Brand impact pages
• Chatbots and AI-assisted customer-service content

Do not clean the product page while old retailer copy, feed fields or paid creative continue making the same unsupported claim.

2. Turn claims into governed data

For every environmental statement, create one central evidence record.

This is not bureaucracy for its own sake. It creates a source of truth for content teams, retailer partners, agencies, AI tools and customer-care teams.

3. Replace broad claims before creating new content

Prioritise claims without a clear answer to three questions:

• What exactly does this refer to?
• What evidence supports it?
• Can the qualification sit beside the claim where the consumer sees it?

If the answer is unclear, rewrite, qualify or remove the claim before it spreads through advertising, feeds and AI-generated content.

4. Give AI tools boundaries

Most marketing teams now use AI to draft PDPs, briefs, ads, FAQs and customer-service responses. Those systems should not be free to convert evidence into adjectives.

Build simple rules into prompt libraries, content workflows and product-data systems.

Illustrative wording for AI-assisted content workflows.

The goal is not to make product communication robotic. It is to make it accurate enough for people and useful enough for machines.

5. Test product representation, not only page compliance

Do not assume approved copy produces accurate AI representation.

Test the questions shoppers ask when they compare products:

• ‘Which fragrance-free baby creams are available in Germany?’
• ‘Which vegan and cruelty-free serum is suitable for mature skin?’
• ‘Which refillable cleanser has independently verifiable packaging information?’
• ‘Which moisturiser avoids [specific ingredient] and is available in Italy?’

Record:

• Whether the brand appears
• Which attributes the system repeats
• Which evidence it cites or appears to use
• Which attributes it doubts or cannot establish
• Which competitors enter the shortlist instead
• Whether results differ across systems, countries, languages or prompts

Treat results as observations, not final diagnoses. A missing recommendation may involve availability, source access, brand authority, product data, model behaviour or other factors.

But repeated patterns give teams a practical place to investigate.

For related measurement questions, see what consumer brands should measure in AI product recommendations.

6. Prepare claims for agentic-commerce pathways

Agentic commerce increases the importance of data systems can connect across the buying process.

A future purchasing agent may need more than product benefits. It may need to establish:

• Is this the right product for the stated need?
• Are required attributes supported?
• Is the exact SKU available in the customer’s market?
• Can the price, delivery and return conditions be confirmed?
• Is the product compatible with the shopper’s preferences or restrictions?
• Can the product be purchased through an approved route?

For this reason, claims should link to:

• The correct SKU and variant
• Supporting documentation or certification
• Product identifiers
• Market availability
• Price and fulfilment data
• Returns and commercial terms where relevant
• Product schema
• Merchant feeds
• Marketplace records
• Any protocol or structured interface used for product discovery and transaction

The strategic objective is not a perfect product page.

It is a coherent commercial record that can survive comparison, verification, recommendation and transaction.

The commercial decision

EmpCo is not an AI visibility tactic.

For beauty brands selling across the EU, sustainability now reaches beyond brand storytelling. It is becoming part of product-information architecture, regulatory compliance, marketplace content, data strategy and AI-mediated product selection.

EmpCo does not create an AI recommendation advantage by itself. It creates a regulatory reason to make environmental claims specific, bounded and supportable.

AI-mediated shopping creates the commercial reason. When an attribute becomes part of the buying criteria, an external system needs evidence it can find, connect and interpret.

Do not make your product look more sustainable. Make its environmental claims easier to verify.