Using generative AI in academic work should generally be considered collaboration rather than cheating when its use is disclosed

What's this about?

People disagree about whether using AI that makes text counts as teamwork or cheating when students tell the truth about it.

The answer depends on the task, the amount of AI help, and who did the main thinking.

What supporters say

• AI can act like a tutor, explain ideas, help plan work, and give useful comments.
• Telling teachers about AI use helps show what the student did and what the AI did.
• Openly sharing AI use is better than hiding it or trusting weak AI-check tools.

What critics say

• AI can make false claims, wrong facts, and other errors that students cannot check.
• Telling the teacher does not break a class rule that bans AI or limits its use.
• If AI does most of the work, it may replace the skill that the task should test.

How to read this

The number of points on each side does not show who is right; the strength of the proof matters more.

The bottom line

AI use can count as teamwork when it gives limited help and the student stays in charge.

But telling the truth alone does not make AI use fair, so this broad claim does not fit every task.

The claim that disclosed generative-AI use should generally count as collaboration rather than cheating has gained support, but the evidence does not justify such a broad rule. The key questions are what the assignment is meant to measure, how much AI contributed, and whether the student can stand behind the final work.

The case for

The strongest case is for limited, transparent assistance that leaves the student in control of the important thinking. Human-centred guidance on AI use supports applications in which the tool helps but does not replace the student’s judgment. In that setting, disclosure allows teachers to see what the system did and what the student did, making the collaboration more accountable.1

Research also supports AI’s use as a tutor, explainer, brainstorming partner or feedback tool. A review of educational studies found possible improvements in feedback, engagement, explanations and performance, although results varied depending on the task, the student’s preparation and how the AI was used. A controlled study of a structured AI tutor similarly found that such systems can support learning in a defined teaching environment.2 That finding, however, does not show that AI should be allowed to write assessed work.

Disclosure is also preferable to relying on secrecy or automated detection. AI-text detectors can be unreliable, particularly across languages and different writing styles, and should not be treated as proof of misconduct. Reviewing declarations alongside drafts, prompts, revisions or explanations could make the process more open and reduce dependence on opaque accusations.3 But disclosure alone cannot prove that the use was allowed or that the work remains the student’s own achievement.

The case against

Assignment rules remain decisive. Disclosure does not override a course policy that bans AI, limits it to certain tasks or requires independent work. Institutional guidance, including Oxford’s, distinguishes permitted from prohibited uses by assessment. Education researchers likewise argue that acceptable help depends on what the assignment is designed to measure.4

If an assignment tests independent writing, reasoning, coding or source analysis, substantial AI-generated material may replace the skill being assessed. Research on take-home assessments warns that chatbot-written sections can make submitted work a poor measure of individual learning. Reviews also connect generative AI not only with learning support, but with unauthorised assistance and efforts to bypass assessment.5 Declaring that use does not turn substituted work into genuine collaboration.

There is a further problem of reliability. AI systems can introduce invented facts, biased claims or false references. Testing of chatbot bibliographies found uneven performance in retrieving sources, while broader reviews identify hallucinations, overreliance and continuing problems with authorship and assessment validity.6 Students remain responsible for checking the material, and disclosure does not remove that responsibility.

The bottom line

The evidence does not support a general presumption that disclosed AI use is not cheating. It more strongly supports a conditional approach: tutoring, language help, brainstorming and formative feedback can count as collaboration when they leave the assessed reasoning and answer under the student’s control.

By contrast, generating the core argument, solution or analysis may defeat the assignment’s purpose even when fully declared. Disclosure is best treated as an important accountability requirement, not a complete integrity safeguard. Research shows that declarations can be incomplete, inaccurate or hard to interpret, so institutions also need clear rules, meaningful attribution, verification and consistent enforcement.

Confidence is high that classification must depend on the assignment’s rules and learning objective. The remaining uncertainty is how schools and universities should draw the boundary between assistance and substitution across different subjects and tasks.