Green Claims and AI-Generated Marketing Copy: The Other EU Deadline

Back to blog
Table of contents

Updated September 2026. Reading time: about 4 minutes. General information, not legal advice.

While fashion has spent September absorbing the AI Act's image rules, a second European deadline has arrived almost unnoticed. From 27 September 2026, the Empowering Consumers for the Green Transition directive applies across the EU: the set of consumer-law amendments written to end vague environmental marketing. On its own, it is a sustainability story. Combined with how fashion now produces its marketing text, it is an AI story, and that intersection is the one nobody is writing about.

What changes for green claims

The directive amends EU consumer law to prohibit, among other practices, generic environmental claims that cannot be backed by recognised excellent environmental performance. Words that have carried fashion marketing for a decade become regulated statements: eco-friendly, green, sustainable, climate neutral, biodegradable, and their cousins. Sustainability labels not based on approved certification schemes are restricted. Claims built on emissions offsetting face specific limits. The direction is uniform: every environmental statement must be specific, substantiated and verifiable, or absent.

Member states have been transposing the directive into national law through 2026, and its provisions apply to practices from 27 September. 

Where AI turns a wording problem into a volume problem

Here is what makes this a different risk in 2026 than it would have been in 2020. Fashion product copy is increasingly generated: thousands of descriptions produced from templates, prompts and product data, at a speed no copywriter ever matched. That efficiency has a sharp edge. A human writing one product page makes one mistake. A generation template that reaches for "eco-friendly" as filler makes that mistake across every product it touches, in every language you generate, dated and archived.

Generative models are enthusiastic about exactly the vocabulary this directive restricts. Sustainability adjectives are statistically everywhere in fashion marketing text, so they are everywhere in what models produce unless you explicitly forbid them. Nobody decided to make a green claim. The template made it a thousand times.

The second intersection: when text meets the AI Act

There is a quieter overlap on the AI Act side. Its text-disclosure rules are narrow for marketing: ordinary product copy is generally outside them. But the scope covers AI-generated text published to inform the public on matters of public interest, and the Guidelines treat sustainability and health information as exactly that kind of matter. A product description that stops describing a garment and starts making claims about environmental impact is drifting from commercial speech toward public-interest information. The safe reading: the more your AI-generated copy leans on sustainability claims, the more of the AI Act's text rules you may be pulling onto yourself, on top of the consumer-law exposure. We will publish a full analysis of this boundary in October.

Three checks worth doing this week

  • 1. Audit your templates and prompts, not just your published pages. Search generation templates, system prompts and product-data feeds for the restricted vocabulary. The published pages are symptoms; the template is the cause.
  • 2. Match every surviving environmental claim to its evidence. Specific and substantiated survives; generic and decorative does not. If the evidence file does not exist, the claim should not either.
  • 3. Put the prohibition into the prompt. The same generation systems that create the exposure can enforce the fix: an explicit instruction listing the terms your copy may never use, and requiring specificity where environmental facts appear. It is the rare compliance control that costs one paragraph.

Fashion's regulatory season did not end in August. The image rules and the deepfake classification now share the calendar with consumer law that reads your product pages as closely as your pictures. The teams that treat generated text with the same discipline as generated images will spend this autumn calmly.

How would you rate this article