EU AI Act enters force — phased rollout starts with GPAI rules

The EU AI Act entered force on August 1, 2024, with a phased enforcement calendar. The first binding provisions — general-purpose AI (GPAI) model transparency obligations and the prohibited-use list — take effect February 2025. High-risk-system obligations follow in August 2026.

The scope is wide. Any AI system placed on the EU market must meet the Act's requirements, which includes models that serve EU users from non-EU infrastructure. Fines for the most serious breaches reach up to 7% of global annual turnover.

For Shopify merchants using AI to surface product recommendations or to generate marketing copy, the Act matters in two specific ways. First, if you sell into the EU and your store uses an AI system classified as 'high-risk' — recommender systems at scale, for example — you inherit obligations for transparency and human oversight. Second, the GPAI transparency rules mean the upstream model providers now document training data signals, which changes how citations are likely to behave.

The second point is the more interesting one for GEO. As GPAI models disclose their training composition, AI-answer surfaces will increasingly be auditable in terms of where citations come from. That rewards merchants with clean public retrievability — llms.txt, open product feeds, stable canonical URLs — and penalises merchants whose content is locked or schema-less.

The Act itself doesn't change what you publish on your PDPs, but it does raise the floor on what AI-answer surfaces can say about sources. Clean, parseable catalog content becomes more valuable, not less, under this regime.