OpenAI has announced that it plans to apply an invisible watermark to text generated by ChatGPT and Codex in the EU. According to the report, OpenAI is also opening an opt-in option for API users. The company’s own testing, as described in the source, indicates that editing generated text can weaken the watermark.

For European businesses, this is not just a technical update. It has practical implications for AI content workflows, internal governance, client deliverables and how organisations document the use of generative AI in marketing, support, software and knowledge work.

What happened

According to Search Engine Journal, OpenAI said it will introduce invisible watermarking for ChatGPT and Codex text in the EU. The watermark is intended to mark text generated by these systems without changing the visible wording for readers.

The same report says OpenAI will make watermarking available as an opt-in feature for API users, rather than applying it automatically in every API use case. The source also notes that OpenAI’s testing found that edits to generated text can reduce the watermark’s reliability.

Based on the available information, this is an announced product and policy change rather than a new EU legal requirement in itself. The source does not suggest that all AI-generated text can be reliably detected in every case.

Why it matters for European businesses

Many European companies now use large language models for website copy, product descriptions, customer service drafts, internal documentation, coding support and marketing production. If a major AI provider starts watermarking text in the EU, businesses may need to review how AI outputs are created, edited, stored and shared.

The main business significance is not only detection. It is governance.

• Content operations: Marketing teams using ChatGPT for draft creation may need clear rules on whether and how AI-assisted content is reviewed and rewritten before publication.
• Client transparency: Agencies and service providers may need clearer contractual or operational language on the use of generative AI in deliverables.
• Internal compliance: Companies in regulated or risk-sensitive sectors may want better records of where AI-generated text is used in customer-facing or operational processes.
• Technical planning: Businesses building on the OpenAI API may now have to decide whether opt-in watermarking supports their governance, trust or audit objectives.

The fact that editing can weaken the watermark is also important. It suggests watermarking should not be treated as a guaranteed detection or compliance mechanism. In practice, businesses should see it as one signal among several, not as a definitive proof of origin.

Who may be affected

The impact is likely to be strongest for organisations that use OpenAI tools directly or indirectly in routine business processes.

• Marketing teams and e-commerce businesses using AI for articles, landing pages, product content, ad copy or email campaigns.
• Software and IT teams using Codex or related coding assistance in development workflows.
• Digital agencies producing client content or development work with AI-supported tools.
• Founders and SMEs that rely on ChatGPT for everyday drafting, customer communications or process automation.
• Regulated businesses that need stronger documentation, review and approval controls around externally published or operational text.

Companies using multiple AI vendors should also note a wider market issue: watermarking approaches may differ across providers, and not every AI tool will handle provenance in the same way.

What companies should consider

• Review AI usage policies: Define where generative AI is allowed, what level of human review is required and which content types need approval before publication.
• Map AI-generated content flows: Identify which teams use ChatGPT, Codex or API-based tools for customer-facing, internal or regulated outputs.
• Assess API settings carefully: If your business uses the OpenAI API, evaluate whether opt-in watermarking supports client trust, auditability or internal controls.
• Avoid overreliance on detection: Because edits may weaken the watermark, do not treat watermarking as a complete safeguard against misuse or mislabelling.
• Update client and stakeholder communication: Agencies and service providers may want to clarify when AI assistance is used and how outputs are reviewed.
• Monitor future guidance: Businesses operating in regulated environments should watch for further technical details from OpenAI and for any relevant European regulatory or contractual implications.

For European businesses, the immediate takeaway is practical: invisible watermarking may become part of AI content governance, but it does not remove the need for human review, documentation and clear internal rules on how generative AI is used.