AI-generated content disclosure for ecommerce: EU AI Act, FTC rules and best practices

Thousands of ecommerce stores now use AI to write product descriptions, generate lifestyle images and synthesize customer reviews. The tools are fast, cheap and effective. The legal obligations around them are no longer theoretical. They are enforceable.

Two regulatory frameworks drive disclosure requirements: the EU AI Act, which applies to any store selling to EU customers, and FTC guidance on endorsements and testimonials, which applies to US stores. This guide covers what each requires, how they overlap, what to label, how to label it and how to build a compliant AI content workflow without hurting your conversion rates.

What counts as AI-generated content in ecommerce

The scope of "AI-generated content" is broader than most merchants assume. It is not just chatbot responses:

Disclosure requirements at a glance

EU AI Act Article 50: what it actually requires

Article 50 of the EU AI Act covers transparency obligations for AI systems. Paragraph 4 is the most relevant for ecommerce: deployers of AI systems that generate synthetic text, images, audio or video must ensure the output is "marked as artificially generated or manipulated."

The obligation applies when the content could "appreciably resemble existing persons, objects, places, or other entities or events and would falsely appear to a person to be authentic." For ecommerce, this covers AI-generated product photographs, synthetic model images, AI-written product descriptions and AI-generated testimonials.

There are exceptions. The obligation does not apply when the AI-generated nature is "obvious from context" (an obviously stylized illustration) or when the content is "necessary for the exercise of the right to freedom of arts and science." Standard ecommerce product content does not qualify for these exceptions.

Who is the "deployer"?The store using the AI tool to generate content. If you use GPT-4 to write product descriptions, you are the deployer and you are responsible for labeling. OpenAI is the "provider" of the AI system, not the deployer. The disclosure obligation falls on you.

Timeline. Article 50 obligations for general-purpose AI systems took effect in August 2026. Full obligations apply from August 2027. Stores selling to EU customers should treat August 2026 as the operative deadline.

For a broader view of how the EU AI Act affects ecommerce, see our guides on EU AI Act transparency requirements and AI Act risk categories for ecommerce.

FTC rules on AI-generated reviews and endorsements

The FTC updated its Endorsement Guides in 2023 and finalized its Rule on the Use of Consumer Reviews and Testimonials in October 2024. The rules specifically address AI:

• Fake reviews are prohibited regardless of source. An AI-written five-star review, even if based on real customer data, is a fake review if it does not reflect the honest opinion of a real customer. Civil penalties up to $50,120 per violation.
• AI-assisted review summaries need clear attribution. Using AI to synthesize a "customers love this for..." summary from real reviews is permissible but requires clear labeling that the summary is AI-generated, not a direct customer quote.
• Incentivized AI reviews are not a loophole. Paying a real customer to use an AI tool to write a review does not make it genuine. The endorsement must reflect the person's actual experience in their own expression.
• AI-generated testimonials in marketing. Using AI to create fictional customer testimonials or case studies violates FTC endorsement rules. All testimonials must be from real customers reflecting real experiences.

In 2025, the FTC fined three companies a combined $4.2 million for AI-generated fake review practices. The precedent is established and enforcement is active.

AI-generated product images: a specific risk category

AI-generated product images present a distinct compliance challenge because they can be materially misleading:

• EU consumer protection enforcement. The German Wettbewerbszentrale issued a warning in 2025 to a fashion retailer whose AI-generated clothing images showed a fit and texture that differed significantly from the physical garment. Other EU consumer protection bodies have signaled similar priorities.
• AI model images. Using synthetic AI models instead of real people to display clothing or accessories is increasingly common. Several major retailers (Levi's, Mango) have disclosed their use. Disclosure is required under Article 50.
• Lifestyle image generation. Using AI to place products in lifestyle settings (a sofa in a staged room, a watch on a wrist) requires disclosure if the image could be mistaken for a real photograph.
• Minor vs material AI editing. Basic AI edits (background removal, brightness) are generally not "AI-generated" for disclosure purposes. Material alterations (changing product color, modifying texture, adding features) likely require disclosure.

The practical rule: if the AI changed what the product looks like in a way that could affect purchase decisions, disclose it and ensure the image still accurately represents what the customer will receive.

AI product descriptions at scale: quality and accuracy risks

Many mid-market stores use AI to generate product descriptions at scale. A catalog of 10,000 products that would take months to write manually can be described in days. The compliance challenge is accuracy:

• Hallucinated specifications. AI models generate technically plausible but incorrect specs: wrong dimensions, incorrect materials, inaccurate certifications. In regulated categories (supplements, electronics, medical devices), this creates product liability exposure.
• EU AI Act accuracy requirements. If AI writes that a supplement contains 500mg of vitamin C when the actual product contains 250mg, that is both a product liability issue and an AI Act violation if the system was used without adequate human oversight.
• Duplicate content. AI descriptions across a large catalog tend to be structurally similar. Without differentiation, product pages compete with each other and with competitors using the same tools.
• Unsubstantiated claims. AI may generate claims ("clinically proven," "best in class") that are not substantiated. These trigger FTC advertising substantiation requirements regardless of who or what wrote them.

Labeling approaches that work in practice

The EU AI Act does not prescribe a specific label format. It requires content be "marked" in a "suitable" way. Approaches that work:

Product image labels

A small "AI-generated image" label in the corner of AI-generated product photographs. Some stores use an icon; others use text. Position it in the image caption or overlay, not buried in the footer.

Description footnotes

• "This description was written with AI assistance and verified by our product team."
• "AI-assisted description. Product specifications verified against manufacturer data."

These labels are clear, specific and signal quality control. They satisfy disclosure while maintaining trust.

Review section labels

• "AI-generated summary of customer reviews"
• "This summary was created by AI from verified customer reviews"

Page-level disclosure

For stores where most content is AI-generated, a sitewide disclosure on a content policy page can complement individual labels. This should not replace per-content labels.

What to avoid

• Vague language. "Computer-assisted" or "technology-enhanced" are likely insufficient under EU AI Act requirements. The disclosure must clearly communicate AI generation.
• Hidden labels. Disclosure buried in page source code, invisible to users, does not satisfy the requirement.
• Overly prominent labels. A full-screen warning is unnecessary. A small, clear label is sufficient.

Does disclosure hurt conversion rates?

The data is more reassuring than most merchants expect:

• Proactive disclosure protects trust. A 2025 Reuters Institute study found 62% of consumers want to know when content is AI-generated. Purchase intention was only negatively affected when AI content was perceived as deceptive, not when disclosed proactively.
• "Verified by our team" is key. Labels combining AI disclosure with human verification perform significantly better than bare "AI-generated" labels. The human verification signal offsets the AI disclosure.
• Discovery is worse than disclosure. Hidden AI content that is later discovered signals deception. The reputational damage from a "brand uses undisclosed AI" headline far exceeds any conversion impact from a small label.
• Category matters. AI disclosure has minimal impact on commodity products (phone cases, office supplies). More impact on premium or trust-dependent categories (luxury, supplements, baby products). Adjust by category if needed.

SEO implications of AI content disclosure

Google's position has stabilized: AI content is not penalized per se, but content lacking E-E-A-T signals will rank poorly regardless of creation method.

• The disclosure label has no SEO impact. Adding "AI-assisted" labels does not trigger any ranking penalty. Google has explicitly stated this.
• Content quality matters. Generic AI descriptions that are similar to competitor output rank poorly due to lack of differentiation.
• Hybrid content ranks best. AI first drafts enhanced with human expertise (product specialist insights, real usage details) consistently outperform both pure AI and unstructured human content.
• Structured data still works. Product schema, review schema and FAQ schema work the same on AI-assisted pages. Continue implementing structured data.

Building a compliant AI content workflow

Step 1: Audit existing AI content

Inventory your catalog. Which descriptions, images and review summaries were AI-generated? You cannot label what you have not identified. If you adopted AI content generation over the past two years, you may have thousands of unlabeled AI descriptions.

Step 2: Establish a review process

• Check product specifications against manufacturer data sheets
• Verify dimensions, weights, materials and certifications
• Remove unsubstantiated marketing claims
• Add product-specific details AI cannot know (feel, fit, common customer questions)
• Ensure images accurately represent the physical product

Step 3: Implement disclosure labels

• Image-level labels for AI-generated product photographs
• Description-level footnotes for AI-written product text
• Section-level labels for AI-synthesized review summaries
• Interaction-level disclosure for AI chatbots (must identify as AI at start of conversation)

Step 4: Document your AI systems

The EU AI Act requires deployers to maintain documentation. Record which AI tools you use, what content they generate, your human oversight process and when AI content was published.

Step 5: Train your content team

Create a one-page guide: "If you used AI to create this content, it needs a label. Here is how to add one." Anyone publishing product content should know the requirements.

Step 6: Monitor regulatory developments

The EU AI Act is still being interpreted by national regulators. Monitor guidance from the European AI Office and the FTC. For a broader compliance view, see our ecommerce compliance audit checklist.

Penalties and enforcement

The EU AI Office has been established and national AI authorities are being designated in each member state. The FTC has made AI-generated content a stated enforcement priority for 2026-2027. Early compliance avoids the scramble when enforcement actions begin targeting ecommerce specifically.

Frequently asked questions

Do AI-generated product descriptions need to be labeled?

Under the EU AI Act Article 50(4), AI-generated text that could be mistaken for human-created content must be marked as artificially generated. A visible label like "Written with AI assistance, verified by our team" satisfies the requirement.

Can you use AI to write product descriptions legally?

Yes. Using AI for product descriptions is legal. The EU AI Act requires disclosure, the FTC prohibits fake reviews regardless of source, and AI content must be factually accurate. The key is transparency, not prohibition.

Does disclosing AI content hurt conversion rates?

Research shows proactive disclosure has minimal negative impact. Purchase intention is only hurt when AI content is perceived as deceptive, not when disclosed proactively. Labels with human verification signals ("verified by our team") perform best.

Are AI-generated product images regulated?

Yes. The EU AI Act requires AI-generated images to be labeled. EU consumer protection rules also prohibit AI images that misrepresent the actual product. Enforcement has already begun in Germany.

What are the penalties for not disclosing AI content?

EU AI Act: up to 15M euros or 3% of global turnover. FTC: up to $50,120 per violation for fake reviews. EU consumer protection fines vary by member state.

When does the EU AI Act disclosure requirement take effect?

Article 50 transparency obligations took effect August 2026. Full obligations apply from August 2027. Stores selling to EU customers should treat August 2026 as the deadline.

Does Google penalize AI-generated product descriptions?

No. Google does not penalize AI content per se. Content quality and E-E-A-T signals matter, not creation method. The disclosure label itself has no SEO impact.

Do AI review summaries need disclosure?

Yes. AI-synthesized review summaries must be labeled. The FTC requires disclosure of content not reflecting honest customer opinions. "AI-generated summary of customer reviews" is a compliant label.

How should I label AI-generated content?

Use clear labels. Descriptions: "Written with AI assistance, verified by our team." Images: "AI-generated image." Review summaries: "AI-generated summary of customer reviews." Avoid vague terms like "computer-assisted."

Do I need to disclose AI if a human edited the output?

If AI generated the base content and a human edited it, disclosure is still recommended. The EU AI Act covers content "generated or manipulated" by AI. Light editing does not remove the AI-generated nature. When in doubt, disclose.