# Should I use AI to write my essay? Why a tool never should

**Should I use AI to write my essay, or is that cheating?** No. In EdCitation's view, AI should never write your essay. A tool should find sources, check your work and explain what you got wrong, while the writing stays yours, because writing is how you learn and what you are assessed on. At Cambridge, Stanford and Toronto, handing in AI-written work without permission or acknowledgement can be academic misconduct.

Published 2026-10-02 by EdCitation. https://edcitation.com/newsletter/why-a-tool-should-never-write-your-paper

Should you use AI to write your essay? This is EdCitation's view, and every tool we make is built on it: no. A tool should find sources, check your work and explain what you got wrong. It should never write your paper, because the writing is how you learn the subject and it is the thing your marker is assessing. Hand the sentences to a model and you lose both.

That is not a case against AI. Our line is narrower: a tool may help you think, and the words you hand in stay yours. The evidence follows, from university policies, sector guidance and studies of learning, each read on 2 October 2026.

EdCitation never writes any part of anyone's work, and we are proud of that. Its free tools do the jobs a tool should: [Find sources](https://edcitation.com/) searches about 300 million published works, [Verify references](https://edcitation.com/verify-references) checks that every reference exists, and [Check your paper](https://edcitation.com/check) reads your assignment's instructions into a checklist.

## Should I use AI to write my essay?

No, in our view, and the reasons have nothing to do with getting caught. An AI-written essay is poor evidence of your learning, it skips the work that does the learning, and it can carry claims and sources that are not true.

### The essay is the evidence

A degree is awarded on evidence that you met its learning outcomes, and coursework is much of that evidence. The UK's Quality Assurance Agency warned in 2023 that generative AI means some assessments can no longer be confidently credited to the student who handed them in (QAA, 2023). The paper on assessment reform commissioned by TEQSA, Australia's higher education regulator, names the same worry first: AI use in a task calls the student's own learning into question (Lodge et al., 2023). An essay a model wrote tells the marker about the model.

### Writing is how the learning happens

Graham et al. (2020) pooled 56 experiments in which pupils wrote about what they were studying in science, social studies and mathematics, and found that the writing reliably improved learning, with an average effect size of 0.30. The studies were in schools, years 1 to 12, so they make the case for writing to learn rather than measure university essays.

### The paper has to be true

A language model writes what sounds right. Cabezas-Clavijo and Sidorenko-Bautista (2026) collected 400 references from eight free chatbots across five fields: only 26.5% were entirely accurate, and nearly 40% were flawed or fabricated. [A reference is a promise](https://edcitation.com/newsletter/a-reference-is-a-promise) explains why we never write one, and [what studies say about AI-fabricated references](https://edcitation.com/newsletter/what-studies-say-about-ai-fabricated-references) sets out the tests model by model.

## What does the research say about AI and learning?

It finds that AI which hands over the answer improves the work in front of you and leaves less in your head.

### When the tool gives answers

Bastani et al. (2025) ran a field experiment with nearly 1,000 students at a high school in Turkey. Some practised maths with a GPT-4 tutor that behaved like ordinary ChatGPT, some with a version told to give teacher-designed hints instead of answers, and some with no AI. With the tool, practice grades rose by 48% for the plain version and 127% for the hinting one. Then came an exam without AI: students who had used the plain version scored 17% lower than those who never had access. The hinting version largely removed that harm, though it brought no gain either. Many in the plain group had copied solutions, and they did not notice any loss in their learning. (A correction in August 2025 changed only an author's affiliation.)

### When the tool improves the essay

Fan et al. (2025) gave 117 university students a writing task with help from ChatGPT, a human expert, writing analytics tools or nothing extra. The ChatGPT group improved their essay scores the most, yet their gain in knowledge, and their ability to apply it to a new task, were no different from the other groups'. The authors warn of what they call metacognitive laziness: leaning on the tool instead of monitoring your own learning.

Kosmyna et al. (2025) asked 54 people to write essays with a language model, a search engine or neither. In the first session, 15 of the 18 who used the model could not correctly quote a sentence from the essay they had just written; in each of the other two groups, 2 of 18 could not. It is a small preprint: a warning, not a verdict.

### What the studies do not show

None follows students through a degree, and Bastani's tasks were maths, not essays. Still, the pattern is clear: when the tool produced the answer, the work improved and the learning did not; when it explained, the harm shrank. That is where we draw our line.

## Is it cheating to use AI to write an essay?

It depends on your course's rules. In every policy we read that addresses it, handing in AI-written text as your own without permission or acknowledgement is a breach; they differ on what is allowed by default.

| University | What its published guidance says |
| --- | --- |
| University of Cambridge, UK | AI may support study, research and formative work; unacknowledged AI content passed off as your own in a summative assessment is misconduct, unless the brief says otherwise (University of Cambridge, n.d.) |
| LSE, UK | Each course takes one of three positions for 2026/27: no authorised use, limited authorised use or full authorised use (London School of Economics and Political Science, n.d.) |
| Stanford University, US | Without a clear statement from the instructor, AI is treated like help from another person, and using it to "substantially complete" an assignment is not permitted (Stanford University Board on Conduct Affairs, 2023) |
| Harvard University, US | Every instructor must state a course policy; the samples run from forbidding AI at every stage to encouraging it, and each leaves the student responsible for what is submitted (Harvard University, n.d.) |
| University of Toronto, Canada | Where AI is prohibited, using it to produce even part of a marked assignment is use of an unauthorised aid (University of Toronto, n.d.) |
| University of Sydney, Australia | AI may be used in open, unsupervised assessments, with acknowledgement, and generally not in secure, supervised ones (University of Sydney, 2026) |

### Where the policies disagree

Sydney encourages AI in open assessments; Stanford starts from restriction; LSE and Harvard leave it to each course. All 24 Russell Group universities have reviewed their conduct policies to show where AI use is inappropriate (Russell Group, 2023), while the TEQSA paper sees little value in blanket bans (Lodge et al., 2023). For you, your own course outline wins. Our view goes one step further: even where AI writing is allowed, the studies above suggest you learn less from it.

The Higher Education Policy Institute's 2026 survey of 1,054 UK undergraduates found 94% using generative AI to help with assessed work, and 12% saying they had put AI-generated text directly into it, up from 3% in 2024 (Stephenson & Armstrong, 2026). The commonest uses were explaining concepts, summarising and structuring ideas. For the grey area of AI editing, see [is using AI to edit my essay plagiarism?](https://edcitation.com/newsletter/is-using-ai-to-edit-my-essay-plagiarism)

## What may an AI tool do for a student, and what should it never do?

A tool may find, check and explain; it should never write. The table is our view, with the evidence behind each line.

| Job | A tool may | A tool should never | Source |
| --- | --- | --- | --- |
| Finding sources | Search published records and show what exists | Invent a source, or summarise one you have not read | Cabezas-Clavijo and Sidorenko-Bautista (2026); [Find sources](https://edcitation.com/) |
| References | Look each one up in a record and flag retractions | Write a reference from memory | [Verify references](https://edcitation.com/verify-references); [Cite a source](https://edcitation.com/cite) |
| The brief | Turn the instructions into a checklist | Decide what your answer argues | [Check your paper](https://edcitation.com/check) |
| Understanding | Give hints, explain a concept, ask you questions | Hand you the finished answer | Bastani et al. (2025) |
| Feedback | Show where a draft falls short, quoting it | Rewrite your sentences | Fan et al. (2025) |
| Spelling and grammar | Correct slips where your course allows it | Rephrase your argument | LSE asks courses that bar AI to say whether grammar tools count |
| The writing | Nothing | Draft, paraphrase or disguise any part | Cambridge; Stanford; Kosmyna et al. (2025) |

## How do I use AI for an essay without letting it write?

Keep the tool on the jobs that leave the writing to you. In order:

1. **Read your course's AI statement.** If there is none, ask your tutor before you start, as Toronto advises.
2. **Turn the brief into a checklist** with [Check your paper](https://edcitation.com/check), free, so you know the limits, sources and style first.
3. **Find and read your sources.** [Find sources](https://edcitation.com/) searches by topic or by the claim a sentence needs to support.
4. **Write the first draft yourself**, from your own notes, before any tool sees the question.
5. **Ask a tool to explain, not to answer**, if your course allows it: a hint, a definition, a question back.
6. **Check every reference** in [Verify references](https://edcitation.com/verify-references), free, so nothing invented reaches your marker.
7. **Declare what you used** as your course asks ([how to declare AI use in an assignment](https://edcitation.com/newsletter/how-to-declare-ai-use-in-an-assignment)), and keep your drafts.

## Why does EdCitation check and never write?

Because finding, checking and explaining are the jobs a tool can do without taking your learning, and they are the only jobs EdCitation does. It works from the publisher's record: a reference is looked up, never written, and no part of a paper is written for you. For the commonest factual failure of AI writing, the reference that does not exist, [Verify references](https://edcitation.com/verify-references) is the best tool for the job. It searches the record for each entry, which no chatbot can promise; in the chatbot test above, only about a quarter of references were entirely accurate.

### What our two runs returned

On 2 October 2026 we gave [Check your paper](https://edcitation.com/check) a short brief we wrote for this test. It is ours, not a course's:

> Essay: 1,500 words (plus or minus 10%), not counting the reference list. Answer the question: To what extent should universities allow students to use generative AI in assessed writing? Use at least eight sources, at least five of them peer-reviewed journal articles published since 2019. Reference in APA 7th edition. Use 12-point Arial, double line spacing and 2.5 cm margins. Include an introduction, two or more body sections with headings, and a conclusion. Generative AI may be used only to check spelling and grammar; declare any use in an appendix.

It read nine rules, each shown with the sentence it came from. Five of them:

| What it asks | Asked for, as returned |
| --- | --- |
| Word count | 1,350 to 1,650 words |
| Sources (peer-reviewed) | At least 8 sources |
| Source dates | At least 5 peer-reviewed sources: 2019 or later |
| Citation style | APA 7 |
| Margins | 0.98 in |

The eight-source row is labelled "peer-reviewed", though our brief asked that of only five; the sentence shown beside it makes the difference plain, so read each rule against its source. It listed three sentences as things no rule can test: the question, the introduction, body and conclusion, and the AI rule.

We then gave [Verify references](https://edcitation.com/verify-references) three entries from this guide's reading. Graham et al. (2020), with its DOI, came back verified: "The DOI resolves to this record and the title matches." So did Bastani et al. (2025), without one: "Title, year and first author match a published record." The third, Fan et al. (2025), we altered on purpose: its DOI, three of its nine authors, and a wrong year, journal and volume. It came back "check this": "The publisher's record also names Shen, Zhao and Shen, who are not in your entry. Check the authors." That reason named three of the six authors we cut, and not the wrong journal or year, so set a doubtful entry against its record in full before you fix it.

### What it costs

Check your paper and Verify references are free, with no account. Pro, $8 a month, adds [References from a file](https://edcitation.com/tools/references-from-a-file) and [Mechanics QA](https://edcitation.com/tools/mechanics-qa). Max, $24 a month, adds Theoretics QA, which judges your paper against your assignment's own instructions with quotations from the paper and never rewrites anything, and the [Library](https://edcitation.com/tools/library). [Pricing](https://edcitation.com/pricing) has the detail.

## Quick questions

### Can AI write my paper for me?

It can produce text, but in our view it should not, and many policies forbid handing it in as your own. In Bastani et al. (2025), students who leaned on AI for answers did better in practice and 17% worse in the exam without it.

### Is it cheating to use AI to write an essay?

At Cambridge, Stanford and Toronto, submitting AI-written work without permission or acknowledgement can be misconduct. Sydney allows AI in open assessments, so your own course's statement decides.

### What are the ethics of AI writing tools for students?

Using a tool to find, check or explain is ethical where your course allows it and you declare it. Handing in its sentences as yours misstates who wrote the work.

### Does EdCitation write essays or rephrase text?

No. EdCitation never writes any part of anyone's work. It finds sources, builds references from the publisher's record, checks that references exist, and reads your assignment into a checklist with [Check your paper](https://edcitation.com/check).

## References

- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. *Proceedings of the National Academy of Sciences, 122*(26), Article e2422633122. [https://doi.org/10.1073/pnas.2422633122](https://doi.org/10.1073/pnas.2422633122)
- Cabezas-Clavijo, Á., & Sidorenko-Bautista, P. (2026). Assessing the performance of 8 AI chatbots in bibliographic reference retrieval: Grok and DeepSeek outperform ChatGPT, but none are entirely accurate. *Journal of Data and Information Science, 11*(2), 102–116. [https://doi.org/10.1515/jdis-2025-0326](https://doi.org/10.1515/jdis-2025-0326)
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. *British Journal of Educational Technology, 56*(2), 489–530. [https://doi.org/10.1111/bjet.13544](https://doi.org/10.1111/bjet.13544)
- Graham, S., Kiuhara, S. A., & MacKay, M. (2020). The effects of writing on learning in science, social studies, and mathematics: A meta-analysis. *Review of Educational Research, 90*(2), 179–226. [https://doi.org/10.3102/0034654320914744](https://doi.org/10.3102/0034654320914744)
- Harvard University, Office of Undergraduate Education. (n.d.). *Generative AI guidance*. [https://oue.fas.harvard.edu/faculty-resources/generative-ai-guidance/](https://oue.fas.harvard.edu/faculty-resources/generative-ai-guidance/)
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). *Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task* [Preprint]. arXiv. [https://doi.org/10.48550/arXiv.2506.08872](https://doi.org/10.48550/arXiv.2506.08872)
- Lodge, J. M., Howard, S., Bearman, M., Dawson, P., & Associates. (2023). *Assessment reform for the age of artificial intelligence*. Tertiary Education Quality and Standards Agency. [https://web.archive.org/web/20260929222431/https://www.teqsa.gov.au/sites/default/files/2023-09/assessment-reform-age-artificial-intelligence-discussion-paper.pdf](https://web.archive.org/web/20260929222431/https://www.teqsa.gov.au/sites/default/files/2023-09/assessment-reform-age-artificial-intelligence-discussion-paper.pdf)
- London School of Economics and Political Science. (n.d.). *School position on generative AI*. Eden Centre. [https://info.lse.ac.uk/staff/divisions/Eden-Centre/Artificial-Intelligence-Education-and-Assessment/School-position-on-generative-AI](https://info.lse.ac.uk/staff/divisions/Eden-Centre/Artificial-Intelligence-Education-and-Assessment/School-position-on-generative-AI)
- Quality Assurance Agency for Higher Education. (2023). *Reconsidering assessment for the ChatGPT era: QAA advice on developing sustainable assessment strategies* [PDF]. [https://www.qaa.ac.uk/docs/qaa/members/reconsidering-assessment-for-the-chat-gpt-era.pdf](https://www.qaa.ac.uk/docs/qaa/members/reconsidering-assessment-for-the-chat-gpt-era.pdf)
- Russell Group. (2023). *Russell Group principles on the use of generative AI tools in education* [PDF]. [https://russellgroup.ac.uk/media/6137/rg_ai_principles-final.pdf](https://russellgroup.ac.uk/media/6137/rg_ai_principles-final.pdf)
- Stanford University Board on Conduct Affairs. (2023, February 16). *Generative AI policy guidance*. Office of Community Standards. [https://communitystandards.stanford.edu/generative-ai-policy-guidance](https://communitystandards.stanford.edu/generative-ai-policy-guidance)
- Stephenson, R., & Armstrong, C. (2026). *Student generative AI survey 2026* (HEPI Report 199). Higher Education Policy Institute. [https://web.archive.org/web/20260929005549/https://www.hepi.ac.uk/wp-content/uploads/2026/03/HEPI-Report-199-Gen-AI-Survey-2026.pdf](https://web.archive.org/web/20260929005549/https://www.hepi.ac.uk/wp-content/uploads/2026/03/HEPI-Report-199-Gen-AI-Survey-2026.pdf)
- University of Cambridge. (n.d.). *Artificial intelligence (AI)*. Plagiarism and academic misconduct. [https://www.educationalpolicy.admin.cam.ac.uk/plagiarism-and-academic-misconduct/artificial-intelligence-ai](https://www.educationalpolicy.admin.cam.ac.uk/plagiarism-and-academic-misconduct/artificial-intelligence-ai)
- University of Sydney. (2026, June 2). *Using artificial intelligence tools responsibly*. [https://www.sydney.edu.au/students/responsible-ai-use.html](https://www.sydney.edu.au/students/responsible-ai-use.html)
- University of Toronto. (n.d.). *Using ChatGPT or other generative AI tool on a marked assessment*. Academic Integrity. [https://www.academicintegrity.utoronto.ca/perils-and-pitfalls/using-chatgpt-or-other-ai-tool-on-a-marked-assessment/](https://www.academicintegrity.utoronto.ca/perils-and-pitfalls/using-chatgpt-or-other-ai-tool-on-a-marked-assessment/)