OpenAI adds visual ads beside ChatGPT image results in the U.S. and watermarks ChatGPT text in the EU

OpenAI will begin showing visual display ads next to image-generation results in ChatGPT for U.S. users later this month, according to TechCrunch. The ads will be labeled, will not influence ChatGPT's answers, and will launch with an initial group of test advertisers. The format extends an ad business OpenAI started earlier this year, when it introduced ads to support its free and low-cost tiers, and expanded them to India in August.

Separately, OpenAI is adding an invisible, machine-readable watermark to text generated by ChatGPT and Codex for users in the European Union, across all plans, to comply with the EU AI Act's transparency requirements that took effect on August 2. The system, called textGrain, embeds a subtle statistical pattern in word choice. OpenAI says it found no meaningful difference in overall performance, citing a Terminal-Bench score of 53.90% without the watermark and 56.06% with it, and says textGrain matches or exceeds other watermarking approaches. API customers worldwide can opt in, and the feature is off by default for them.

The watermark has clear limits. OpenAI's own figures show detection falling from about 92% to 66% when 10% of words are replaced with synonyms, and to 17% at 25%; detector access is limited to approved researchers, and OpenAI notes that the absence of a watermark does not prove a human wrote the text.

What to watch: advertiser uptake and user response to image-adjacent ads, which matter for OpenAI's push to monetize its free users, and whether other AI providers adopt comparable marking for EU users as enforcement of the AI Act's transparency rules begins.