Short answerFor most product content teams, the EU AI Act creates three practical duties: AI literacy for staff using AI tools since 2 February 2025, transparency obligations under Article 50 since 2 August 2026, and structured due diligence on AI vendors. Typical uses such as enrichment, translation and tagging fall outside the high-risk category.
Key takeaways
- The AI Act, Regulation (EU) 2024/1689, sorts AI systems into four tiers: prohibited, high-risk, transparency-obligated and minimal risk, and most product content uses sit in the bottom two.
- Article 4 has required AI literacy measures for staff since 2 February 2025; the Digital Omnibus of July 2026 softened the wording but kept the duty for teams using AI tools.
- Article 50 transparency obligations apply from 2 August 2026, and the Digital Omnibus, Regulation (EU) 2026/1744, moved high-risk deadlines to December 2027 and August 2028.
Most product content AI sits in the lower risk tiers
The AI Act regulates by use case, and typical product content uses carry limited or minimal risk. Regulation (EU) 2024/1689 defines four tiers:
| Tier | Examples | Relevance for product content |
|---|---|---|
| Prohibited (from 2 Feb 2025) | Manipulative techniques exploiting vulnerabilities, social scoring | Avoid dark patterns built on AI-driven persuasion |
| High-risk (Annex III and product safety legislation) | Recruitment, credit scoring, safety components of regulated products | Rarely applies to content work |
| Transparency (Article 50) | Chatbots, synthetic images, deepfakes, certain AI-generated text | Shopping assistants and synthetic product imagery |
| Minimal risk | Classification, tagging, translation, quality checks | Most content use cases, voluntary codes only |
Attribute extraction, auto-tagging, translation and copy drafting do not appear in Annex III. A team should still document why each use case sits in its tier, because classification is the first question an auditor or customer will ask.
AI literacy has applied since February 2025
Article 4 requires providers and deployers to take measures ensuring a sufficient level of AI literacy among staff who operate AI systems. This duty has applied since 2 February 2025 and covers product content teams that use AI for enrichment, translation or tagging. The July 2026 Digital Omnibus reworded it towards supporting staff literacy, which lowers the bar without removing it. The Act sets no fixed curriculum, so a proportionate programme is the reasonable reading.
A practical programme covers:
- How the tools in use work, including their known failure modes such as invented facts.
- Which attributes and claims require human confirmation.
- How to record AI involvement and review decisions.
- When to escalate output that looks wrong, biased or legally risky.
Keeping attendance records and role-based material gives evidence of compliance. Depth should vary by role: content editors need practical rules for review and escalation, data stewards need an understanding of confidence scores and error patterns, and managers need enough grounding to approve use cases and set thresholds. A short refresher whenever a tool or model changes keeps the programme current at low cost.
Article 50 sets transparency duties from August 2026
Article 50 obligations have applied since 2 August 2026. Generative systems placed on the market before that date have until 2 December 2026 to add machine-readable marking. Three provisions matter for content teams. Providers of systems that interact with people, such as a shopping assistant or support chatbot, must make clear that the user is dealing with AI. Providers of generative systems must mark synthetic audio, image, video and text output in a machine-readable format. Deployers of deepfakes must disclose that content is artificially generated or manipulated.
For product imagery, photorealistic synthetic scenes showing a product in a setting that never existed deserve careful review against the deepfake definition. Provenance standards such as C2PA content credentials, discussed in rights management and content provenance, help meet marking expectations. AI-generated text published to inform the public on matters of public interest requires disclosure unless it underwent human review and editorial responsibility, which reviewed product copy normally does. Governance practices for that review are covered in generative product description governance.
Vendor due diligence protects the deployer
Most content teams act as deployers of AI built into PIM, DAM or translation platforms, which makes vendor assessment the main control. A due diligence checklist for vendors under /vendors/ should cover:
- Which general-purpose AI models underpin the features, and whether those providers follow GPAI obligations applicable since 2 August 2025.
- Whether synthetic output is marked in a machine-readable way.
- Data handling: training use of customer data, EU hosting, retention and sub-processors.
- Logging of prompts, outputs and model versions available to the customer.
- Configurable human review steps and the ability to disable AI features per category.
Contract clauses should require notice of model changes and cooperation with regulator requests.
Timelines may shift, so track them as amended
The Digital Omnibus on AI, proposed in November 2025, was adopted as Regulation (EU) 2026/1744 and entered into force on 27 July 2026. It moves stand-alone Annex III high-risk obligations to 2 December 2027 and AI embedded in regulated products to 2 August 2028. Article 50 kept its 2 August 2026 date, with the marking grace period noted above. Prohibitions, AI literacy and GPAI obligations already apply. Teams should monitor official publications and record which version of the timeline each internal decision assumed. A broader view of AI use cases appears in AI in product content.
Frequently asked questions
Is AI-generated product copy high-risk under the AI Act?
No. Generating product descriptions, translating content and tagging images are not listed in Annex III and are not safety components of regulated products. These uses typically fall into minimal risk, with possible Article 50 transparency duties for synthetic media and public-facing chatbots.
What counts as sufficient AI literacy under Article 4?
The Act sets no fixed standard. A proportionate approach covers how the tools work, their limitations, which outputs need human confirmation and how to escalate problems. Material should be tailored to roles and technical knowledge, and organisations should keep records of training delivered.
Does the Digital Omnibus postpone Article 50?
No. The Digital Omnibus, adopted as Regulation (EU) 2026/1744, postponed high-risk obligations and left Article 50 applying from 2 August 2026. The only relief is a grace period until 2 December 2026 for machine-readable marking by generative systems placed on the market before 2 August 2026.