OpenAI Adds Text Watermarking to ChatGPT in EU
Left 50%
Center 50%
Right 0%
1 left · 1 center · 0 right
What happened
OpenAI announced it will begin testing visual display ads in ChatGPT in the United States later this month. The ads will initially appear alongside images that users generate, remain separate from those images, will not change ChatGPT's answers, and will not be shown to Plus, Pro, or Enterprise subscribers. Separately, OpenAI said it will introduce an invisible, machine-readable text watermark, called textGrain, for eligible ChatGPT and Codex users in the European Union over the coming weeks to comply with the EU AI Act's transparency rules. The watermark will not be a global default at launch; API customers worldwide can opt in immediately, and detector access will be limited to approved researchers and expert organizations. OpenAI published a technical report on the method and said it does not guarantee reliable detection, especially after editing, and that a missing watermark does not prove human authorship.
Sourcing check · Mostly on the record · 4 articles
3 of 4 articles cite a named source for the story. It started with unnamed sources; a named source appeared 2 hours later.
What these terms mean and how they're counted
- On the record · names a source 3 of 4
- On the record means the article cites a source by name for the story's main claim: a person, or an institution speaking officially, such as a company statement or a named agency's spokesperson. The opposite is an unnamed source. The count is how many of the counted articles cite at least one such source; the names are everyone named across them, each once however the articles spell it. Here: OpenAI
- Primary evidence · cites a document or data 1 of 4
- Primary evidence means the article cites a record you could check yourself: a court filing or ruling, a dataset or report, a recording or transcript, an official statement or press release, market data. A quote is not a document. The count is how many articles cite one. Here: blog post
- Credits another outlet 0 of 4
- The article says the story comes from another outlet's reporting (“first reported by Reuters”). Those articles relay the story; they are not independent confirmation. Here: Each article presents its own reporting.
- Unnamed sources 0 of 4
- The main claim is attributed to people who are not identified: “people familiar with the matter”, “officials said”. The count is how many articles do this.
- How it developed
- When the first report landed and whether it named anyone; when the first named source and the first document appeared. Here: First report Oct 5, 13:09 UTC by The Verge, with no named source. First named source 2 hours later, from TechCrunch. First document 7 hours later, from TechCrunch (blog post).
- Not counted 0
- Opinion, analysis and video pieces, and articles too short to assess. Every count above is out of the counted articles. An article republished under a new headline is one article.
This is about who is cited, not whether the story is true. Each source below says what it names. How we count
BLINDSPOT.
Only left-leaning outlets are covering this story
— the other side's media is silent.
Omitted — what each side leaves out
Unpacked
Right-leaning outlets had not covered this as of publication. Left-leaning coverage reports that OpenAI will test visual ads beside ChatGPT image results for U.S. users later this month and will add an invisible text watermark for ChatGPT and Codex in the European Union. The fact that changes what those marks actually do is also missing from that coverage: company tests found watermark detection fell to 66 percent after light synonym edits.
Left-leaning coverage opens the ads story by saying OpenAI is "sticking more ads" into ChatGPT and that the new creatives will not influence answers. It never states that advertisers would eventually reach 1.2 billion weekly users. Left-leaning coverage of the watermark instead stresses a regional launch rather than a worldwide switch.
Will a watermark that ordinary editing can weaken still meet the EU requirement that other systems be able to identify AI-generated text?
Bottom line
Free U.S. ChatGPT users will get picture ads next to generated images, while EU users get a hidden watermark that is not a global default. The Verge reported that split; right-leaning outlets have not.
The receipts — 5 facts one side's coverage lacks
- Replacing 10% of words with synonyms dropped watermark detection from about 92% to 66%.
- Ads will not appear for ChatGPT Plus, Pro, or Enterprise plans.
- OpenAI had built a text watermark before but held off, partly over concerns that users would switch to rivals.
- The EU AI Act’s transparency rules took effect on August 2.
- Detector access is limited to approved researchers and expert organizations and will not be publicly available at launch.
The Left View
The Verge framed the ad change as OpenAI sticking more ads into ChatGPT, focusing on a new visual format that places images of sponsored products on screen when users generate pictures, expanding beyond the name-logo-and-link ads introduced in February. It reported that ads stay labeled and separate from generated images, will not influence answers, are excluded from paid Plus, Pro, and Enterprise plans, and that OpenAI says guardrails will keep them out of emotionally vulnerable or sensitive conversations. On watermarking, The Verge reported an EU-only rollout of OpenAI's textGrain system for ChatGPT and Codex, compared it with Google DeepMind's SynthID and Anthropic's similar EU-driven move, and relayed OpenAI's caveats that detection is not guaranteed, the feature is not a global default, API customers can opt in worldwide, and watermarks do not prove accuracy, ownership, or how much a human contributed.
Our Take (balanced)
The story is substantive: it rests on OpenAI's own product announcements, a named method (textGrain), a published technical report, specific U.S. and EU rollouts, named subscription exclusions, and named advertising-measurement partners. What is established is that OpenAI has publicly committed to testing visual ads beside image generation in the U.S. later this month and to EU-only text watermarking for ChatGPT and Codex, with documented limits on detection after editing and no claim that absence of a watermark proves human writing. What is not yet independently verified is how disruptive the ads will be in live use, how often watermarks will fail or be stripped, and whether either feature expands beyond the announced regions and plans. No right-leaning outlet has reported this. Watch for the U.S. ad test going live, watermarks appearing in EU ChatGPT and Codex output, independent tests of textGrain's robustness, and any later decision to make watermarking a global default or to show visual ads outside image generation.
4 sources
More stories
The week's bottom lines, in your inbox
One email a week: the five stories that mattered and what they actually mean. Free.