How to use AI responsibly when writing an essay.
You can usually use AI to help you understand a topic and find leads. You usually can't use it to write, rewrite or plan the essay for you. This is where the line tends to fall, why it is there, and how to use AI without the work stopping being yours.
By Sean Stevens, Chair of Judges
The short answer
Can you use AI to write your essay? If the essay is being assessed, almost certainly not. Handing in text that a tool wrote, or that a tool rewrote from your draft, is treated as cheating by schools, exam boards and competitions, in the same way as handing in someone else's work.
Using AI around the essay is a different question. Many schools and competitions allow it for things like explaining a concept, suggesting questions to investigate or pointing you towards sources, as long as you check what it tells you and say that you used it. The rules for your essay decide, and they vary, so read them first.
Whose rules apply
There is no single rule. The one that counts is the one set by whoever is assessing the work:
- Exam coursework. In the UK, the Joint Council for Qualifications treats misuse of AI in assessed work for GCSEs and A levels as malpractice. Other exam systems, such as the IB, publish their own guidance.
- Your school or teacher. Many schools now have an AI policy, and individual teachers may set tighter rules for a particular task.
- Competitions. Each sets its own rules, usually in a short AI-use policy or in the terms.
If the rules say nothing about AI, don't take that as permission. Ask the person setting the task, and keep their answer.
Where the line usually falls
Policies differ in the details, but most draw the line in a similar place. Use this as a starting point, then check it against the rules you are under.
| Usually allowed, if declared | Check the rules first | Usually not allowed |
|---|---|---|
| Explaining a concept you are trying to understand | Asking for feedback on a draft | Writing any part of the essay |
| Suggesting topics or questions to investigate | Translating your own writing | Rewriting, paraphrasing or “improving” your sentences |
| Pointing you to sources, which you then read | Summarising a long source you have also read | Planning or structuring your argument |
| Checking spelling and grammar | Suggesting objections to your argument | Choosing your position for you |
| Producing references you haven't read |
A quick test for anything not on the list: after using the tool, could you explain and defend every sentence, every source and every step of the argument without it? If the honest answer is no, the tool did work the assessor thinks you did.
Why the line is there
An essay is assessed as evidence of your thinking: what you understood, which evidence you chose, how you weighed it and what you concluded. If a tool chose the structure or wrote the sentences, the marks would be for the tool.
There is a second reason, which matters even where AI is allowed. Writing is where most of the thinking happens. Deciding what your main claim is, which point comes first and which objection is strongest is hard because it forces you to work out what you actually think. Skip that and you end up with an essay you can't defend and a skill you haven't built.
Five uses that keep the thinking yours
These uses are widely allowed and genuinely save time. The example prompts show the difference between asking for help and asking for answers.
- Explain something you are stuck on.“Explain what a carbon price is, as if to someone who has studied GCSE economics. Then give me two questions I could ask to check I've understood.”
- Widen the field before you narrow it.“List ten debates about how cities might adapt to hotter summers.” Then choose and narrow the question yourself.
- Find leads, not evidence.“Which organisations publish data on youth employment in Europe?” Then go to those organisations and find the figures yourself. The guide to finding evidence shows how to trace a figure to its source.
- Test your understanding, not your essay.“Ask me five questions about how vaccines are approved, one at a time, and tell me what I got wrong.”
- Check spelling and grammar at the end.Accept corrections to spelling and punctuation, but not suggestions that rewrite whole sentences.
Notice what the prompts don't do: none asks for a thesis, a plan, a paragraph or a better version of your sentence.
Check everything it tells you
AI chatbots produce fluent, confident text whether or not it is true. Two problems come up again and again in essays.
Invented sources
Chatbots can produce references with real-sounding authors, journals and links for papers that don't exist. In one study published in Scientific Reports in 2023, 55% of the references generated by GPT-3.5 and 18% of those generated by GPT-4 were fabricated. Newer tools may do better, but the rule doesn't change: open and read every source before you cite it. The guide to referencing covers this in more detail.
Wrong or out-of-date facts
A chatbot may give a figure without a source, mix up two studies, or describe a situation that has since changed. Treat anything it tells you as a lead to check, not as evidence. If you can't find a reliable source for a claim, leave it out.
Keep a record and declare it
Where AI use is allowed, it almost always has to be declared. Make that easy for yourself:
- Save your conversations as you go, with the date. UK exam-board guidance asks students to keep a non-editable copy, such as a screenshot, of the questions they asked and the answers they got.
- Keep your drafts. Version history in Google Docs or Word shows how the essay developed, which is useful if anyone asks how you wrote it.
- Declare specifically. Name the tool and say what you used it for.
The stronger version tells the reader exactly what the tool did and what you did. Declaring allowed use is normally not penalised. Leaving it out, if it comes to light later, usually is.
How AI use is spotted
Many schools and competitions run entries through AI-detection software. These tools are not reliable on their own. A 2023 study by researchers at Stanford found that several popular detectors wrongly flagged more than half of a set of essays by non-native English speakers as AI-generated, which is why careful assessors don't decide on a detector score alone.
What usually gives AI-written work away is reading it. Typical signs are a smooth but general argument with no specific evidence, sources that can't be found, a voice that doesn't match the student's other work, and a student who can't explain their own essay when asked. If you did the work yourself, that last test is easy to pass: you can talk through how you chose your question, where your evidence came from and why you rejected the strongest objection.
Common questions
Does Grammarly count as AI?
Basic spelling and grammar checks usually don't. Features that rewrite sentences, change your tone or generate text do, and most policies treat them like any other AI tool.
Can I write in my first language and translate it?
It depends on the rules, so ask. Automatic translation turns your words into someone else's phrasing, and some assessors treat that as rewriting.
What if my friend used AI and didn't get caught?
Work can be reviewed after results are published, and the risk sits with the person whose name is on it. More to the point, your friend has an essay they can't defend and a skill they didn't practise.
Further reading
- AI use in assessments: protecting the integrity of qualifications (PDF), Joint Council for Qualifications
- Generative AI in academic writing, The Writing Center, University of North Carolina at Chapel Hill
- Fabrication and errors in the bibliographic citations generated by ChatGPT, Walters and Wilder, Scientific Reports (2023)
- GPT detectors are biased against non-native English writers, Liang and others, Patterns (2023)
Try it on this year's questions
Ten Years From Now is a free essay competition for students aged 14–18, anywhere in the world. Choose one of three questions and make an evidence-based case in up to 1,000 words, judged against six published criteria. Its AI-use policy allows AI for brainstorming, finding sources you then check, explaining concepts and checking spelling and grammar, as long as you declare it. The argument, the structure and the words must be your own. Round 1 closes on 28 February 2027 and Round 2 on 15 June 2027.
More guides
The next ten years deserve your best thinking.
Pick a question, investigate it properly and argue for what you think should happen. Entry is free.