A Taxonomy of Students’ Excuses for Secretly Using AI

Students are using ChatGPT to justify almost any level of AI assistance in their coursework, with a new study identifying six categories of excuses that help them blur the line between legitimate usage and outright cheating.

A new research collaboration from the US has found that many students no longer see AI-assisted cheating as cheating at all. Based on interviews with college students across the country, pursuing a diversity of topics, the study identified 23 different ways that AI use in coursework may be justified, from claiming that ‘everyone does it’ and ‘AI has no victim’, to arguing that using AI saves time, produces better writing – or still counts as original work, if the output is edited by the student afterward.

Some students openly admitted breaking course rules, while continuing to view their behavior as reasonable.

The study also shows students as either conflicted (when there is a choice to contend with) or confused (where the choices around AI use are not clear) in many scenarios, and fearing a severe competitive disadvantage if they perceive that others are making gains with AI (whether that is perceived as permissible or not).

The 23 examples were distilled down from a far greater number identified from the interviews conducted, and each sit in one of six ultimate categories decided on by the researchers:

Victimless Behavior (nobody is harmed); Minimal AI Contribution (AI only helped a little); Ex Ante Contribution (the student’s ideas came first); Post Hoc Contribution (later editing makes the work theirs); Responsibility Denial (someone or something else is responsible); and Perceived Benefit (the results justify the use).

The authors of the new work (titled “It’s OK Because…”: The Wild West of Student Rationalization of AI Use in Academic Writing, and coming from four contributors across Pennsylvania State University, the University of Michigan, and the University of Miami) note that the students’ arguments and accounts wander illogically between the various categories, even when alternate categories are not complementary or compatible.

They note further that there is often no consistent logic in the arguments – as if the students were panicking in their response (despite a scenario of total confidentiality); or alternately, as if this was the first time they had really been tasked with thinking about the moral dimensions of AI use in their education.

The authors state*:

‘Any formal learning environment arguably involves an implicit social contract in which the instructor helps the student learn, and the student is honest about what they know and do not know.

‘However, students’ AI-use rationalizations suggest that they are unaware of this latter expectation. Significantly, AI use blurs the boundary between student and AI work and it makes it difficult for instructor to assess their work and help them improve their learning.

‘A key intervention for higher education, then, is to help students understand the pedagogical rationale behind AI policies, including why honest representation of one’s own knowledge is important for both learning and assessment.’

The paper indicates that expectations around use of AI in an academic context are distributed across five segments: faculty intention; formal policy; student interpretation of these; student self-policy; and student actual practice.

Taking a look at the six ultimate defined categories, which we’ll do in a moment, shows up various ways in which these five domains are not in agreement with each other; are ill-defined; self-conflicting; or just plain ignored.

The data was obtained by interviewing twenty undergraduate students across 12 different U.S. universities, at a 15/5 female-male ratio, across a wide diversity of majors, from English and art through to statistics and molecular biology.

1: Victimless Behavior

The first category reflects perhaps the simplest justification of all: that nobody is really being harmed.

Students in this group argued that traditional concerns about plagiarism depend on the existence of a human victim, and that AI-generated text breaks that connection:

Since ChatGPT is not a person, they reasoned, using its output cannot be equivalent to stealing from another author (notwithstanding pending legal cases that would indicate otherwise).

Others suggested that AI merely combines and rewrites information from many sources rather than copying any one source directly, bringing issues of ownership into dispute.

In some cases, the only harm acknowledged was self-inflicted, with students arguing that the real risk lay in weakening their own learning, rather than wronging a teacher, classmate, or writer. One participant opined:

‘AI doesn’t have a soul… giving AI credit [or] respect as if AI had ownership or legality to certain information… that’s simply not true.’

Another:

‘It’s not hurting anybody …I think you are hurting yourself more than you’re hurting like a teacher or other people.’

2: Minimal AI Contribution

The second category contends that AI’s role is too small to matter. Rather than denying that AI had contributed to their output, students argued that the contribution was minor, routine, or comparable to forms of assistance that are already widely accepted:

Many drew a distinction between important academic work and what they regarded as administrative ‘busywork’, contending that low-value assignments were not worth the effort they demanded.

Others viewed AI in a similar light to grammar-checkers, editors, or search engines. Across these arguments, the common theme was that AI had helped, but not enough to undermine authorship, or to consider that the final submission was someone else’s work, or attributable to anyone but the student.

One of the participants illustrated the extent to which they considered ChatGPT an equally valid contributor to more traditional mentors in the academic space:

‘If I don’t have that answer for myself, then she [the coach] would say, you could do this. That’s like, the same as what ChatGPT is doing for me.’

3: Ex Ante Contribution

For many students, the strongest defense of AI use was not that the technology contributed very little, but that the important intellectual work had already been completed before ChatGPT became involved.

In these cases, AI was presented as a tool for expressing, organizing, expanding, or refining material that already existed in the student’s head:

Students argued that the real value lay in having the idea, as well as subsequently choosing the direction of the assignment, or gathering the relevant sources; while the AI merely helped turn those ingredients into finished prose.

This definition of authorship contends that ownership comes not from writing every word, but from providing the original intent and steering the process, allowing students to see themselves as the true creators, even when substantial portions of the final text were generated by AI.

4: Post Hoc Contribution

The fourth category focused on what students did after receiving AI-generated text. Students argued that paraphrasing AI output, selecting only certain passages, checking facts, verifying sources, or revising the wording was enough to make the final work their own:

Some also suggested that if ChatGPT had learned to write in a way that matched their usual voice, through habitual usage, the resulting text could reasonably be treated as their own work:

‘It has recognized a pattern of my way of writing and how I express things in, like projects or emails or anything like that. So now, it knows what kind of person I am like in the way of writing.

‘So now, it gives the prompts similar to like, what my thinking process is.’

5: Responsibility Denial

The fifth category involved shifting responsibility away from the student. Some argued that AI use had become so common that avoiding it placed them at a disadvantage, particularly if classmates were using ChatGPT to complete work more quickly or achieve higher grades:

‘Some of my classmates who just throw it [into AI], generate it, don’t even check it, [and] they get higher grades than I do.’

Other students indicated unclear policies, vague assignment instructions, or a perceived lack of concern from instructors:

Some students justified AI use on the grounds that it was unlikely to be detected or punished, while others suggested that responsibility lay with the technology itself rather than with the person using it.

Across these arguments, AI use seemed to be treated as something driven by circumstances, norms, or external factors, rather than by personal choice.

6: Perceived benefit

The final category was the most straightforward, wherein AI use was justified because it produces desirable results.

For some students, the main benefit was saving time, particularly on assignments they considered routine or low-value:

‘[I would rather] spend time with my girlfriend…’

Others argued that they still learned from the material by reviewing, rewriting, or working through AI-generated content:

Some believed ChatGPT produced stronger writing than they could manage on their own, while others focused on outcomes such as better grades, positive feedback, or simply completing the assignment successfully. Across these arguments, the value of the result was treated as more important than concerns about how that result had been achieved.

A Kingdom Divided

The tension between institutional rules and everyday AI use is highlighted in the paper through one participant, identified as P6.

This particular student acknowledged that using AI against course rules was wrong; agreed that undisclosed AI use could be considered plagiarism; and expressed concern that heavy reliance on AI might reduce learning and weaken communication skills:

‘Honestly, it’s probably not very ethical. I do a lot of my homework with AI, and it doesn’t always mean I’m learning […]’

‘[…] AI helps me a lot, because I can’t think about what I want to say. But if I have AI, write stuff and then I change it to say more what I want to say, I still think that’s ethical, because it’s still words I agree with and I believe in.’

At the same time, the participant argued that AI-generated text became more acceptable when it was edited, rewritten, or brought into line with personal views. AI use was also viewed differently depending on the assignment, the perceived value of the course, and whether the goal was to save time or learn something from the exercise.

Similar patterns appeared elsewhere in the interviews: rather than relying on a single rationale, many students combined several, weighing factors such as disclosure, effort, learning, convenience, and assignment value, when evaluating their own use of AI.

Conclusion

It would be a mistake to revel in students’ confusion and culpability around AI usage, given the current lack of reasonable and consistent guidelines in the workplace, or in society in general.

At the moment, the zeitgeist around AI is reactionary at best, with any usage whatsoever of AI considered in various communities and sectors to ‘tarnish’ output as entirely AI-generated. At the moment, there’s a lack of nuance, standards, or forgiveness.

It remains to be seen whether AI’s contribution to prose output will be perceived in the future as an interpretive layer, little more significant than a spell-check, or a concession of human agency, creativity, and interpretive faculty, ceding to machine-made patterns.

* My conversion of the authors’ inline citations to hyperlinks. However, due to the unusually limited formatting of the citations in this paper, I will not be able to provide the usual number of supporting links, due to lack of the time necessary to locate the missing links in the work.

First published Friday, May 29, 2026