Icecat

AI-Generated Product Content and the EU AI Act

In general, AI-generated product content does not need to be flagged under the EU AI Act.

While such labeling is therefore generally not required for compliance, Icecat is developing an AI-generated content layer for business reasons. This will give channel partners greater transparency and control, allowing them to distinguish between brand-validated, human-reviewed and AI-generated content and apply their own quality rules.

Does AI-generated product content need to be flagged?

Content typeAI label required?Why?
Product taxonomy – categories, feature names, feature values, units and identifiersNoShort, structured data (“data labels”) falls outside the scope of Article 50(2) and the text-labelling duty of Article 50(4).
Product descriptions – titles, marketing text and bullet pointsGenerally noThe Commission specifically identifies advertisements and product descriptions as generally outside the disclosure requirement. An important exception can apply to claims concerning areas such as health, consumer safety or sustainability, particularly where AI-generated content has not undergone substantive human review.
Product translationsNoAI-generated translations are explicitly included among standard editing activities exempt from the Article 50(2) marking requirement.
Product imageryGenerally no, provided the product itself is not materially misrepresentedAI-generated backgrounds, staging, colour correction, cut-outs and resizing generally do not constitute deepfakes. AI-generated imagery that materially changes or misrepresents the actual product may do so.

Do AI agents and chat assistants need to disclose that they are AI?

Yes. Article 50(1) requires users interacting directly with an AI system to be informed that they are interacting with AI, unless this is obvious.

Icecat AI assistants therefore disclose that they are AI at the start of the interaction.

Why Icecat is developing an AI-generated content layer

Icecat is developing a specific content layer identifying AI-generated content. This is being done for business reasons, because the AI Act generally does not require AI-generated product content to be labeled.

Channel partners increasingly want to prioritize brand-validated and human-reviewed content over AI-generated content within their own quality rules, marketplace listings and AI shopping assistants.

The Icecat AI-generated content layer gives them this control at source, across the catalogue and within the feeds they already consume.

Icecat’s interpretation of the AI Act

The analysis below is based on the European Commission’s Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act, C(2026) 5054 final, 20 July 2026, and Article 50 of Regulation (EU) 2024/1689.

Article 50 has applied since 2 August 2026. The AI Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) provides a transitional period until 2 December 2026 for the machine-readable marking obligation for providers of generative systems already on the market. The other Article 50 obligations were not delayed.

1. What Article 50 requires, and who is responsible

Article 50 contains four transparency obligations. Two apply to providers of AI systems and two to deployers using AI systems under their own authority.

  • Article 50(1) — providers: AI systems intended to interact directly with people must be designed so users are informed that they are interacting with an AI system, unless this is obvious.
  • Article 50(2) — providers: systems generating synthetic audio, image, video or text content must ensure that their outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. This does not apply where AI performs standard editing or does not substantially alter the input or its meaning.
  • Article 50(3) — deployers: concerns emotion recognition and biometric categorisation and is not relevant to Icecat’s product-content activities.
  • Article 50(4) — deployers: requires disclosure of deepfakes and certain AI-generated text published to inform the public about matters of public interest. An exception applies where the content has undergone human review or editorial control and a natural or legal person assumes editorial responsibility.

Icecat integrates third-party generative AI systems into its content processes. The Guidelines explain that the company providing such an AI system remains responsible for the provider obligations under Articles 50(1) and 50(2), including where its system is integrated into another company’s processes without the need for modifications or additional components.

Icecat acts as a deployer when it uses those systems under its own authority. The Article 50(2) machine-readable marking obligation therefore rests primarily with the AI-system provider. Actors further down the content value chain are encouraged to preserve those marks so that AI-generated content remains identifiable.

2. Product taxonomy: no AI flag required

The Guidelines explain that Article 50(2) is intended to help people distinguish AI-generated or manipulated content and thereby address risks of deception and manipulation.

They consequently exclude content that does not serve that purpose, including:

“Outputs generated in the form of a short sequence of numbers, symbols or letters (e.g. single words, image captions, alt-text, UI labels and icon-scale graphics, image captions and other data labels).”

Product category names, feature names, feature values, units and identifiers fall within this type of short, structured information.

For Article 50(4), the Guidelines similarly state that short texts that do not materially communicate knowledge, opinions or facts cannot be considered to inform the public.

Conclusion: Icecat product taxonomy does not require AI labeling under Article 50.

3. Product descriptions: generally no AI flag required

The Guidelines specifically identify the following as content that is not considered to be published for the purpose of informing the public on matters of public interest:

“AI-manipulated text that is part of a company’s advertisement or product descriptions (not including any claims related to e.g. health, consumer safety or sustainability).”

Product titles, marketing descriptions and bullet points therefore generally do not require an AI disclosure under Article 50(4).

The exception: certain health, safety and sustainability claims

The Commission specifically carves out claims relating to areas such as health, consumer safety and sustainability.

Where AI-generated product content makes such claims and falls within Article 50(4), substantive human review is important. The Guidelines explain that human review requires genuine examination of the substance by someone with relevant knowledge and professional judgment. Fact-checking is a minimum requirement.

Simple spell-checking, grammatical corrections, automated checks or cursory approval are not sufficient.

Icecat therefore expects health, safety and sustainability claims supplied by brands to have been appropriately validated by humans. Where this is not the case and disclosure is required, Icecat advises the brand to identify the content as AI-generated directly within the relevant asset so that the disclosure remains visible when the public encounters it.

The Guidelines also distinguish between a party controlling the use of AI and one merely receiving AI-produced content. For example, a company commissioning an advertisement without controlling whether or how an agency uses AI is not, for that reason alone, the deployer of the agency’s AI system.

Conclusion: ordinary AI-generated product descriptions generally do not require AI labeling. Additional assessment may be required for claims concerning matters such as health, consumer safety or sustainability.

4. Product translations: no AI flag required

The Guidelines explicitly include:

“AI-generated translations of text”

among examples of standard editing and minor alterations covered by the Article 50(2) exception.

Icecat translations preserve the substance of the original product content across more than 40 languages and therefore fall within this category.

Where the underlying content contains health, safety or sustainability claims, the Article 50(4) considerations described above may still be relevant. The Guidelines also give an AI-supported translation that has undergone human review as an example satisfying the human-review exception.

Conclusion: AI-generated product translations do not, in themselves, require AI labeling.

5. Product imagery: generally no AI flag if the product itself remains authentic

Article 50(4)’s deepfake provisions concern content that falsely appears authentic or truthful.

The Guidelines explain that AI processing used for aesthetic or technical purposes generally does not meet this threshold. Examples include colour correction, background extensions or replacements, arrangements of existing products and image resizing.

The Commission specifically gives the example of:

“A real product (e.g., a car) shown in an advertisement against an AI-generated background and surrounding environment”

provided that the advertisement is not likely to mislead people about the actual product, its characteristics or its use.

The position changes where AI alters the product itself in a misleading way. The Guidelines identify as a potential deepfake an AI-generated product image that makes the product appear different, more appealing or higher quality than it actually is.

Icecat uses AI-generated backgrounds, staging, cut-outs and technical enhancements while maintaining the authentic representation of the product supplied by the brand.

Conclusion: AI-assisted product imagery generally does not require deepfake disclosure where AI changes the presentation rather than materially misrepresenting the product itself.

6. AI agents and assistants: disclosure required

Article 50(1) applies to AI systems that interact directly with natural persons.

Article 50(5) requires the relevant information to be provided clearly and distinguishably, at the latest at the time of the first interaction or exposure.

Icecat’s chat widgets, support agents and other directly interactive AI assistants therefore disclose at the beginning of the interaction that the user is interacting with an AI system.

Conclusion: unlike ordinary AI-generated product content, directly interactive Icecat AI assistants must disclose that they are AI.

7. Why Icecat is identifying AI-generated content anyway

The AI Act generally does not require Icecat to add a visible AI label to ordinary AI-generated product content.

Icecat is nevertheless developing an AI-generated content layer because channel partners are asking for greater transparency and because this improves the usefulness of Icecat’s product data.

It will allow retailers, marketplaces and other channel partners to distinguish between brand-validated, human-reviewed and AI-generated content, and to use those distinctions in their own:

  • content quality and ranking rules;
  • marketplace publishing policies; and
  • AI shopping assistants and other automated applications.

The objective is therefore broader than regulatory compliance: to give channel partners greater transparency, choice and control over the content they consume and publish.

Sources

  • Regulation (EU) 2024/1689 (AI Act), Article 50
  • European Commission, Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act, C(2026) 5054 final, 20 July 2026
  • Regulation (EU) 2026/1744 (AI Omnibus), OJ 24 July 2026, in force 27 July 2026
  • European Commission, Code of Practice on Transparency of AI-generated Content

This article describes Icecat’s understanding of the AI Act as applied to its own services. It does not constitute legal advice. Brands and channel partners should assess their own use of AI under the AI Act and applicable national law.

Lucas van Rijen

Recent Posts

OrangeQS Welcomes IQM to MAX Partnership Program

Orange Quantum Systems (OrangeQS), a venture of Icecat Capital, has welcomed IQM Quantum Computers as…

11 hours ago

Icecat Studio Sprint 103 Release Notes: Accessibility Goes Live, Full Design Fidelity in the Sandbox

Sprint 103 was a release-heavy sprint. Our accessibility and interface quality program shipped to production…

3 days ago

AI Token Prices Hit Record Lows as Competition Drives Down the Cost of Intelligence

The cost of using artificial intelligence is falling rapidly, potentially changing the economics of building…

3 days ago

ERP Quoting: How Quote Pages Connect PIM and ERP

Building a quote in an ERP sounds straightforward until the information you need is somewhere…

4 days ago

Carrefour Belgium Brings Grocery Shopping Into ChatGPT

Carrefour Belgium is bringing grocery shopping directly into ChatGPT with a new AI-powered shopping assistant…

4 days ago

Amazon Expands Pan-EU FBA Listing Requirements to the Netherlands and Belgium

Amazon is expanding the marketplace requirements attached to its Pan-European Fulfillment by Amazon program, bringing…

5 days ago