Two students submit essays that involved a generative AI assistant. The first wrote everything herself and asked the assistant to flag unclear sentences and awkward word choices. The second pasted in the prompt, received a draft and adjusted a few lines to sound like himself. Both “used AI”, but only the first essay belongs to its author. The right question is not whether you used AI but which part of the work it did.
Institutional and journal policies are changing fast and vary considerably. Most, however, turn on one principle: the thinking being assessed must be yours, and AI use must be transparent where the rules require it.
A five-level scale
| Level | What the AI does | Usually treated as |
|---|---|---|
| 1. Surface checks | Spelling, grammar, punctuation | Like a spellchecker; almost always allowed |
| 2. Language feedback | Flags unclear sentences and suggests clearer phrasing of text you wrote | Often allowed; some settings require disclosure |
| 3. Structure and argument feedback | Comments on your outline, spots gaps, asks critical questions | Depends on the rules; disclosure advisable |
| 4. Drafting sections | Writes paragraphs, background sections or transitions for you | Often restricted in assessed work; must be disclosed in publications, with full author responsibility |
| 5. Drafting nearly everything | Generates the piece from a prompt; you edit | Typically misconduct in assessed work; incompatible with authorship |
This describes general tendencies, not your specific rules. One module may ban even level 2 because its purpose is to build writing skills; another may permit level 4 and grade how well you critique the AI's output.
Why levels 4 and 5 are riskier than they look
- Fabricated references: AI-drafted passages often come with plausible citations that do not exist, or that exist but do not say what is claimed.
- Generic reasoning: fluent prose that is not anchored in your data and does not reflect what you actually found.
- You cannot defend it: when an examiner asks why you chose this theoretical framework, you have no answer because you did not choose it.
- Indirect plagiarism: generated text sometimes sits very close to existing sources, and you have no way of knowing which source to cite.
Getting value from levels 2 and 3
The principle: you write first, the AI responds. Some useful requests:
- “Here is a paragraph I wrote. Point out sentences that are hard to follow and explain why, but don't rewrite the paragraph.”
- “Act as a demanding reviewer: list the three biggest weaknesses in this section's argument.”
- “Here is my outline. Is there repetition or a missing logical step?”
- “I use the word ‘effect’ here. Given a correlational design, is that too strong?”
This keeps your voice intact, and the AI's response becomes something you weigh, accept or reject, much like comments from a friend reading your draft.
If English is not your first language
You stand to gain the most from level 2, and you are also most at risk of sliding into level 4. A safe routine:
- Write the main ideas in your own English, even if it is rough. If you are stuck, write in your first language and translate it yourself first.
- Ask the AI to identify errors and suggest alternatives, with explanations.
- Choose which suggestions to accept, and note new sentence patterns you have learned.
- Read the final version aloud: any sentence you could not have produced yourself deserves another look.
If you paste in a paragraph in your first language and receive a full English version, the ideas are still yours, but many policies treat this as machine translation requiring disclosure. Check the specific rules.
Keep evidence of your process
If your work is ever questioned, the best evidence is the history of how it was made:
- Write on a platform with version history (an online document editor, Overleaf, or dated file saves).
- Keep reading notes, outlines and handwritten drafts.
- Save AI conversations related to the piece, or at least a short log: date, request, and what you used from the response.
The log also lets you write an accurate AI disclosure instead of reconstructing it from memory at the last minute.
Theses and journal articles: higher stakes
A thesis is assessed as your independent intellectual work, and many universities have thesis-specific rules that are stricter than coursework rules. For journal articles, COPE and many publishers agree that AI cannot be an author because it cannot take responsibility; AI use in preparing a manuscript must be disclosed, and authors are accountable for every word. Discuss your intended AI use with your supervisor early and keep a record of what you agreed.
A pre-submission test
- Could I explain and defend every paragraph if asked?
- Have I personally opened and checked every reference?
- Does my level of AI use fall within the module, institution or journal rules?
- Have I disclosed it in the required form?
- Do I have version history and an AI use log?
Something to do now: find the AI policy for the module or journal you are submitting to next, identify the highest level on this scale it allows, and paste that at the top of your draft. Every time you open an AI assistant, you will know how far you can go.
Câu hỏi thường gặp
Is using AI to fix grammar considered cheating?
Usually not; grammar and spelling correction is treated like any standard checking tool. Some writing-focused courses do restrict it, so read the specific rules.
Can I use AI to write my literature review?
In assessed work this is often restricted or banned. In publications it must be disclosed and you remain fully responsible, including checking every reference, since AI frequently invents citations.
Do I need to disclose using AI to translate my paper into English?
Many policies treat AI translation as language assistance that must be disclosed. The ideas are still yours, but check your institution's or journal's requirements and name the tool used.
Can an AI tool be listed as a co-author?
No. COPE and many publishers state that AI tools do not meet authorship criteria because they cannot take responsibility for the work.
How can I prove I wrote my essay if I'm accused of using AI?
Version history, notes, outlines, drafts and an AI use log are the strongest evidence. Keep them from the start rather than waiting until questions arise.
