The age of labelling AI content: watermarks and tags are spreading

When Anthropic, the maker of the Claude chatbot, began embedding a watermark — an invisible digital marker showing that the content was generated by AI — into text produced by its latest models, it set off a heated debate online. The move was made first and foremost in response to the European Union's transparency rules, but the company is rolling it out everywhere Claude is available. And Anthropic is far from alone: media and tech giants are racing to introduce their own rules for "labelling" AI content.

The trend is taking many forms. Streaming service Spotify announced that from mid-September it will attach an "AI Persona" tag to AI-generated "artists" — fully AI-created profiles with no real performer behind them — and automatically drop such profiles from editorial and algorithmic recommendations. Music-generation platform Suno is rolling out audio watermarking (a hidden digital marker inside the audio file), and YouTube has begun tagging realistic AI content even when the creator has not disclosed it, while on TikTok billions of clips are already flagged as "AI-generated." Google, by contrast, is moving the other way: the company said users will be able to remove the visible watermark from images, video and music created in several of its models — though the invisible marking still remains.

The shifts have sparked sharp arguments online, particularly on X (formerly Twitter). Blogger Erick Erickson wrote in a post: "I dropped Grammarly for Claude to edit my writing because it works better. Now everything I write gets flagged as made by Claude. It's ridiculous." In his view, the practice casts unfair suspicion on authors.

Researcher Federico Germani, in a paper published in July 2026, points to a deeper problem: a hidden watermark only encodes the model's origin, but once it is turned into a visible "AI-generated" label, it flattens a complex creative process into a misleading binary and says nothing about the content's authenticity. As a result, such labels can stigmatise legitimate uses of AI while reinforcing unwarranted trust in unlabelled content.

The situation is riddled with contradiction: even as media companies push audiences to lean ever more on artificial intelligence, platforms are training that same audience to treat AI output with suspicion. AI is being normalised on one hand and branded on the other. For brands and marketers, the lesson is that transparency in content strategy is no longer optional — it is becoming an inseparable part of keeping an audience's trust.