Students edition
ChatGPT invisibly marks text in the EU. GPT-6 also gives free users a screen they can click on.
A watermark does not show how much work you did yourself
OpenAI is adding a pattern to the word choices in some AI-written text in the EU. The pattern is easy to disrupt. If it is missing, that does not prove a person wrote the text.
On Monday 5 October, OpenAI announced that ChatGPT and Codex text will get an invisible watermark in the EU. This supports the transparency rules in Article 50 of the AI Act. Those rules have applied since 2 August 2026. The technology is called textGrain. OpenAI will introduce it over the coming weeks. The watermark applies to all subscription plans, for users who are eligible. It applies only in the EU. Companies can also use ChatGPT through the API. This is a connection for developers. They can switch on the watermark anywhere in the world. It is off by default for them. A further deadline, 2 December 2026, applies to systems that were already on the market.
The watermark stays in place when you copy and paste text. A detector is a program that can recognise the watermark. For now, only approved researchers and organisations can use that detector. The textGrain technology makes a small change to word choice. Often, several words would fit in the same place in a sentence. The model (the program behind ChatGPT) then picks words that form a recognisable pattern a little more often. It is hard to spot the pattern in a single sentence. In a whole text, a detector can check whether the pattern appears more often than chance would explain.
That is also where this technology is weak. If you change the words, you can break the pattern. OpenAI says the chance of detection then falls from almost 100 per cent to less than 20 per cent. You only need to change a quarter of the words. So if you rewrite a ChatGPT text heavily, you remove most of the watermark.
The watermark proves little, whether it is found or not. If it is missing, that does not mean a person wrote the text. If it is there, it tells you something about where the text came from. It does not tell you how much you did or thought up yourself. It also applies only to ChatGPT and Codex in the EU. So a teacher cannot use it as the only proof. So it can help to keep versions of your work, your notes and your prompts. Prompts are the instructions you give the AI. They let you show how you worked, whatever detector or watermark your institution uses.
Background for subscribersThese stories show that a watermark or detection score does not prove how much you did yourself. Your versions, notes and prompts (the instructions you gave the AI) show that better, so it can help to keep them. A chart in ChatGPT can be wrong, just like plain text. Would you like to use Copilot Notebooks and the Study Guide? Your course can tell you whether they are available to you.
GPT-6 in ChatGPT: free users also get a new model and a screen to work with
OpenAI began rolling out GPT-6 in ChatGPT on 7 October 2026. Paying users with Plus, Pro, Business or Enterprise got GPT-6 Sol. Free users and Go users got the smaller model, GPT-6 Luna, on 8 October. A model is the program that makes the AI work. The models come with Intelligent UI. This is a new way to show answers. ChatGPT can now build answers using charts, buttons, forms and diagrams. It can also show small tools, such as a calculator or a cost splitter. OpenAI says the model starts answering while it is still thinking. Sometimes it searches the web for an answer. For these questions, the wait is 44 per cent shorter than with GPT-5.6 Instant. Do you not want to see these images and tools? You can turn them off on the web under 'Layout and visuals'. For Enterprise, administrators decide whether the feature is on. Instead of seeing only an explanatory paragraph, you may see a screen you can click on. For example, you can view a chart, fill in a form or use a calculator in the conversation. ChatGPT builds that screen itself, as part of the answer. The answer comes sooner because the model starts writing while it is still thinking. A polished diagram can seem more reliable than plain text. That may make you less likely to check whether it is correct. But a chart can be based on a wrong answer, just like text. So it can be worth checking a figure or definition from a chart against your course materials.
Background for subscribersThis is a taster.
Subscribers read the whole Students edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Students edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026