What you’ll be able to do

Right, so this is about using AI properly for an essay, not getting it to write one for you and hoping nobody notices. There’s a difference, and it matters. By the end of this you’ll be able to use AI to brainstorm ideas, build a structure, stress test your argument, and clean up your writing at the end. What you won’t be doing is handing in a paragraph you didn’t actually think through yourself. That’s the whole point.

Used well, AI is a bit like a study partner who never gets tired of your questions. It can poke holes in your argument, suggest counterpoints you hadn’t thought of, and tell you if your structure makes sense before your tutor does. Used badly, it writes the essay for you, invents some references, and you get flagged in your viva or worse. So let’s do the first thing.

Before you start

Check your institution’s actual policy on AI. Not what your mate said, not what you assume, the real policy. Most universities have published something on this now, and they’re often stricter than students expect, sometimes more relaxed too. Some ask you to declare AI use, so keep a note of what you used and for what, just in case.

If you’re still working out where the line sits, there’s a related problem sitting just underneath this one. The same sort of rules apply to everyday coursework, and it’s easy to cross the line without feeling like you have. The guide on using AI for homework shows how to use it to understand the work: worked examples, hints and checks, so you do the thinking and AI just helps you get there.

The general line most places draw is this: using AI to brainstorm, structure, or check your work is fine. Using it to generate the actual essay text, the bit you hand in as your own words, is plagiarism at most institutions. And it’s detectable, not always by software (that’s patchy), but by a tutor who knows your writing and suddenly reads something that doesn’t sound like you at all. Vivas exist partly for this reason.

Keep your drafts. Seriously, all of them. If anyone ever asks how you got from a blank page to a finished essay, being able to show your working, notes, outlines, revisions, is the best evidence you have that the work is yours.

Drafts are only half of the paper trail, though, because the revising you do afterwards matters just as much. When exams roll round you’ll want the same material in a form you can test yourself on, and I’d rather you didn’t spend an evening copying out cards by hand. Turning your notes into flashcards covers how to get cards from notes, slides or a textbook chapter that you’ll really revise from.

Get set up

You don’t need anything fancy here. A free account with ChatGPT, Claude, or Gemini will do most of what this guide covers. If your university provides access to Microsoft Copilot through your student account, that works too, and it may already meet whatever data protections your institution requires, worth checking.

Some tools worth knowing about beyond the general chatbots. ChatGPT has a study mode designed to walk you through problems rather than just handing you answers, worth trying if you want something that pushes back a bit. NotebookLM works differently: you feed it your own reading (PDFs, notes, articles) and it answers based on that material, rather than the internet at large, which is genuinely useful for a literature-heavy essay. For citations, Zotero is free and does the boring reference-formatting work so you don’t have to. And once your essay’s drafted, Grammarly is worth a pass for the final polish, catching typos and clunky phrasing, not for writing content.

None of these need much setup. Sign up, verify your email, you’re in. The bit that actually matters is knowing what not to paste into them, which I’ll get to.

Try it yourself

Here are three prompts that do different jobs. Copy them, adjust the brackets, use them as a starting point rather than a script.

First, brainstorming. This is for when you’ve got a topic but no idea where to start.

I'm writing a university essay on [topic]. I have to argue [thesis or question]. 
Can you suggest five different angles I could take, along with the main 
argument each one would need? Don't write any essay content, just the angles 
and a one-line summary of each.

Second, testing your argument once you’ve got a rough position. This one’s the most useful thing AI does for essay writing, in my experience.

My essay argues that [your argument]. What's the strongest objection someone 
could raise against this? Then tell me what evidence or reasoning I'd need 
to answer that objection convincingly.

Third, checking structure once you’ve got a draft outline or a rough draft written.

Here's my essay outline: [paste your outline, not full essay text]. 
Does the argument flow logically from one point to the next? Are there any 
gaps where I've made a claim without backing it up? Don't rewrite it, just 
point out the weak spots.

Notice what’s happening in all three. You’re asking for feedback, angles, and objections, never for paragraphs you can drop straight into your essay. That’s the line.

Check the result

Once you’ve drafted your essay yourself, using whatever the AI helped you think through, run a few checks.

First, read it out loud. Sounds daft, but if a paragraph sounds like it was written by someone else, even if you wrote it, that’s usually a sign you leaned too hard on AI phrasing while drafting. Rewrite it in your own words.

Second, check every single citation. This is the big one. AI tools sometimes invent references that look completely real, a plausible author, a plausible journal, a title that sounds right, and none of it exists. This is the classic failure mode and it will sink you faster than almost anything else. Search for the actual source. If you can’t find it, don’t cite it, and don’t trust the AI’s answer when you ask it to confirm the source is real either, ask it and it might just double down on the fabrication.

Third, ask the AI itself to be your sounding board on the finished piece. Something like: does this paragraph actually support my thesis, or does it wander? Useful for a gut check, not a rewrite.

Fourth, keep your drafts, as mentioned. Version one, version two, whatever you end up with. If your process is ever questioned, this is your proof.

If it doesn’t work

Sometimes the AI gives you rubbish. Vague suggestions, generic objections that don’t actually apply to your argument, citations that don’t check out. Usually this means your prompt was too vague. Go back and be specific: what’s your actual thesis, what’s the actual assignment brief, what stage are you at. The more context you give, the better the response, generally.

Now the bit that matters most: what not to hand an AI tool at all.

Don’t paste in unpublished data, other people’s unpublished work, or anything covered by a confidentiality agreement, dissertation data you haven’t published yet, for instance. Once it’s in a chat window, you’ve lost control of where it goes, depending on the tool’s data policy, and those policies vary and change, so don’t rely on your memory of what a company said last year.

Don’t paste in your full essay if you’re not sure about your institution’s stance on AI-assisted checking, some flag AI submission and detection tools might treat pasted drafts differently than you’d expect. Better to paste small sections, an outline, a paragraph, rather than the whole thing, if you’re unsure.

Don’t rely on AI for anything requiring current fact-checking without verifying separately. It can be confidently wrong, especially with dates, statistics, and specific studies. Treat anything factual it gives you as a lead to verify, not a fact to cite.

And if privacy is a genuine worry, check whether your account settings let you opt out of your chats training the model, most major tools now offer this, though the setting and its exact effect varies by provider, so check their current documentation rather than trusting what a forum post told you.

Questions people ask

Is using AI to write a university essay plagiarism?

The line most institutions draw is that using AI to brainstorm, structure or check your work is fine, but using it to generate the actual essay text you hand in as your own words is plagiarism at most institutions. Check your institution's actual policy, not what your mate said, and note that some ask you to declare AI use.

Which AI tool is best for essay writing help?

A free account with ChatGPT, Claude or Gemini will do most of this, and Microsoft Copilot works too if your university provides it. ChatGPT's study mode pushes back rather than handing you answers, NotebookLM answers from your own reading, Zotero handles references and Grammarly is for final polish.

How do I check AI citations in my essay?

Check every single citation. AI tools sometimes invent references that look completely real, so search for the actual source, and if you can't find it, don't cite it. Don't trust the AI when you ask it to confirm a source is real either, as it might just double down on the fabrication.

What should I not paste into an AI tool for my essay?

Don't paste unpublished data, other people's unpublished work, or anything covered by a confidentiality agreement. Also avoid pasting your full essay if you're unsure of your institution's stance, paste small sections, an outline or a paragraph instead.

Why is the AI giving vague or useless essay feedback?

Usually it means your prompt was too vague. Go back and be specific about your actual thesis, the actual assignment brief and what stage you're at, because the more context you give, the better the response.

Keep exploring

If this has been useful, there’s more worth reading. How to get AI to help with homework covers the same honesty line but for smaller, more frequent tasks. How to get AI to help me research a topic goes deeper into finding and evaluating sources, which pairs well with the brainstorming stage here. And How to get AI to edit my writing picks up exactly where this guide leaves off, once your draft exists and needs polishing rather than building.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.