Where the line really is

A spectrum, not a switch

People argue about AI and cheating as though there were one line. There is a spectrum, and most of it has been settled for decades under other names.

Roughly from unproblematic to prohibited:

1. Spellcheck. Universal, permitted everywhere, nobody discusses it.
2. Grammar checking. Long-standing, permitted almost everywhere.
3. A thesaurus or dictionary. Ancient, permitted.
4. Translation of a source you are reading. Normal scholarly practice.
5. Explanation of a concept. Identical to asking a tutor, a friend, or reading a textbook.
6. Generating practice questions. Produces nothing in the submission. Safe everywhere.
7. Critique of your draft. Identical to a writing centre appointment, which institutions provide and encourage.
8. Editing your prose at sentence level. Where policies begin to differ, and where the correct an error versus rewrite it for me distinction starts to matter.
9. Translation of your own writing into the assessed language. Usually prohibited in language assessments, variable elsewhere, and frequently requires disclosure.
10. Generating text you submit. Prohibited in most current policies unless explicitly permitted.
11. Generating the whole submission. Prohibited everywhere it is not the point of the exercise.
12. Fabricating sources or data. Serious misconduct, and it always was.

Items 1 to 7 are, for almost every policy in the world, fine. That is a great deal of useful help, and students who believe the whole category is forbidden are denying themselves things their institution actively provides.

The question that resolves most cases

What is this assessment trying to measure, and does my use interfere with measuring it?

A translation exercise measures your ability to translate; using a translator defeats it entirely. An essay measures your ability to construct an argument; having grammar corrected does not touch that, and having the argument constructed destroys it. A programming assignment measures whether you can write the program; having an error message explained is fine, having the function written is not.

This single question resolves most cases correctly, and it generalises to tools that do not exist yet, which no list of rules does.

The two-part test

When the question above is still unclear, apply both parts:

Would I be comfortable writing exactly what I did in the disclosure statement? If not, that discomfort is data. It is usually the most reliable signal you have, and it arrives before the deadline.

Could I now do this unaided? If a use leaves you able to do it yourself next time, it taught you something. If it leaves you unable, it substituted for you. This is the test that protects your education rather than your record, and it is the one that matters in five years.

Three cases worth thinking through

A student asks a model to explain a concept, understands it, then writes the essay entirely themselves. Fine everywhere. Identical to reading a textbook. No disclosure required under most policies, and disclosing costs nothing.

A student writes an essay, asks a model to improve the prose, and accepts most of the changes. Depends on the policy and on how much changed. If the argument, structure and evidence are yours and the sentences have been smoothed, most policies permit it with disclosure. If entire sentences now say things you did not write, you have crossed into 10. The honest test is whether you can explain every change.

A student uses a model to translate their essay from their first language into English for submission. Genuinely contested. Some institutions treat it as language support and permit it with disclosure; others treat it as generating the submitted text. It also raises a fairness question, since a student writing in their first language gets a benefit their classmates do not need. Find out your institution’s position before submitting rather than after, and if there is not one, ask.

The part nobody has settled

Where the line should sit is a live argument, and it is being argued differently in different countries, disciplines and institutions. Some academics think most of this will be permitted within a few years and assessment will change to measure other things. Others think closed assessment will expand. Both are defensible; neither is settled.

What is settled, and has been for a very long time: submitting work as your own when it is not, and citing sources you have not read. Those two hold in every version of the future being argued about, and they are the two that generate almost all of the actual cases.