Imagine submitting work that you can explain and defend. Someone asks: “Did you do this with ChatGPT?” It is a hypothetical situation, but it raises a useful question: what would someone need to examine to assess your contribution? Using a tool says little, on its own, about how you researched the subject, what you rejected or why you chose a conclusion.
What OpenAI announced
On 5 October, OpenAI announced textGrain, an invisible statistical watermark in word choices. It plans to introduce it to eligible ChatGPT and Codex text in the European Union over the coming weeks. The announcement gives no general rollout date for the UK, Latin America or the US. Initial detector access will be restricted to approved researchers and expert organisations, rather than public at launch. OpenAI.
What a watermark can reveal
OpenAI’s documentation says detecting the watermark neither identifies the user nor measures their contribution. It does not verify accuracy. The signal is in word choices, without hidden characters. These features help explain why reviewing someone’s work requires more context than a detector result. OpenAI Help Center.
Detection can also fail
OpenAI acknowledges false positives and negatives. An undetected watermark does not prove human authorship either. REDES’ editorial recommendation is to ask for context and examine the process before using a technical signal to make a judgement about a person. OpenAI.
The task changes what you need to demonstrate
Consider two illustrative situations. On a course assessing your ability to write without assistance, delegating the writing could break the assignment rules. In a project that permits supporting tools, using one to organise notes may fit the objective. The quality of the result matters, alongside whether you did the work you were asked to do. Understanding a submission does not make a prohibited use permissible.
Make your decisions visible
Keep an initial outline, relevant drafts and the sources you consulted. Note which AI suggestions you accepted, which you rejected and why. You do not need to archive every interaction to document an important decision. If an idea changed after you checked a fact, that revision can help you explain how you reached the result.
Check before submitting
Open the cited sources and check that they support your claims. Review figures, dates and context. Try explaining the argument without relying on the tool’s response. If you cannot justify a section, work on it again. Fluent writing can conceal a weak conclusion; answering specific questions makes it easier to examine that conclusion.
Clarify the rules at the start
Check what assistance your course, client or team allows and how it should be disclosed. If the guidance is ambiguous, ask about a specific use: organising ideas, editing wording or generating a draft. Disclosing a tool does not replace permission to use it. For assessors, it helps to set out the criteria before receiving the work: what the person must resolve independently and which forms of support are acceptable.
A clarification for people using other applications
An API lets an application communicate with a service. OpenAI offers optional watermarking worldwide for compatible models; it is off by default. This developer option does not mean ChatGPT watermarking is generally enabled in every country. OpenAI Help Center.
Where would you draw the line?
Using AI is not enough to determine a student’s or professional’s value. The conversation becomes more useful when it focuses on the task: what they learnt, what they checked, which decisions they took responsibility for and whether they followed the agreed conditions. In which tasks would you accept that assistance, and what evidence would you ask for to assess the person’s contribution?