AI detection tools and even ChatGPT confirm that AI wrote my student’s text, but the student denies it!

AI detection products also cannot reliably distinguish between an AI generated text and text written by a human, and they reflect biases (for example more often flagging non-native speaker texts). A really useful article on the topic is Bassett et al. (2026), where they summarize why AI detection products cannot work in the first place:

•  Unverifiable probabilistic estimates. Usually, one should be able to distinguish detector outputs between “hit (true positive), miss (false negative), false alarm (false positive), or correct rejection (true negative)”, and from that understand the detector’s sensitivity. But for AI detection, there is no ground truth to verify the model, since the only texts that are definitely not written using AI are old and it is not a valid assumption that texts written today still follow the same linguistic and stylistic pattern (since students might have learned from AI generated texts, and language also just evolves over time).
• Mutually exclusive linguistic markers. “There is no principled reason to believe that a human cannot produce writing that contains linguistic features commonly found in AI-generated text” (especially since AI is trained on human writing, and as humans are exposed to AI texts and might adapt to them)

They also describe common methods that teachers use to validate AI detector outputs and why they don’t work, either:

• Linguistic markers: it sounds too perfect, or uses the type of lists or words that people associate with AI (“thus”): “staff reinforce their assumptions using reasoning that is entirely dependent on the AI detector’s potentially flawed output”
• Multiple AI detectors: which only “amplifies the shared flaws of these tools, creating a misleading appearance of consensus”
• Falsified references: “Fortunately, in such cases, the question of AI use is irrelevant. Submitting fabricated references constitutes academic misconduct in its own right, regardless of how they were generated”
• Student confessions: Might be “confession under duress. The reasoning mistakenly assumes that because the confession follows the AI detection flag, the flag must have been correct. Correlation, however, does not imply causation. Academic integrity must rise above believing in the equivalent of horoscopes, tarot cards, or Ouija boards, simply because their predictions occasionally seem to align with real events. The confession does not retrospectively validate the AI detector’s result, just as a seemingly accurate horoscope does not prove astrology’s legitimacy” (love this quote!)
• Using generative AI, for example by generating a response by AI themselves and comparing (not proof of anything, also easy to fall into confirmation bias) or asking AI if the text was AI generated (not proof of anything, since AI cannot detect AI output).
• Past writing styles: Writing styles evolve over time, and we want students to experiment with styles to find their own voice! So this does not work either.
• Hidden adversarial prompts: You know, the white, tiny text that is supposed to catch students that copy task texts into AI and then don’t pick up on the frequent mention of “tomato” or similar: “Setting traps for students in this way relies on deception, undermines trust between students and staff, and contradicts the principles of fair assessment and academic integrity. […] The issue at the heart of this method is that it assumes dishonesty by default, treating students as inherently deceptive rather than active participants in a learning environment built on mutual respect. Universities should lead by example, upholding the same ethical standards they expect from students” 

What makes things even more complicated is that text does not need to be 100% AI generated OR 100% human generated, it can exist anywhere on that spectrum, where it is generated WITH AI, and this makes AI detection tools meaningless from the start. Bassett et al. (2026) also discuss other facets of using AI detection products, like the burden of proof, the difficulty in defining what exactly is part of an assessment, security risks of uploading student texts to AI detection products, and the students’ right to silence if under investigation, which are all important considerations.

They end with a strong conclusion: “AI detection in education is not merely flawed; it is conceptually unsound. […] Institutions must accept that AI detection is an unworkable solution to a problem that cannot be solved through surveillance and punishment. The focus must move from detection and enforcement to assessment design that recognises AI’s role in learning and the reality that unsupervised assessments cannot be secured. The continued use of AI detectors exposes students to procedural injustices and signals a fundamental misunderstanding of education’s purpose. AI detection does not safeguard academic integrity; it undermines it.”

Read the original publication:

Bassett, M. A., Bradshaw, W., Bornsztejn, H., Hogg, A., Murdoch, K., Pearce, B., & Webber, C. (2026). Heads we win, tails you lose: AI detectors in education. Journal of Higher Education Policy and Management, 1-16.

Let me start by giving your student the benefit of the doubt: how might such 'AI characteristics' have arisen?

Words such as 'thus' and 'leverage' are not exactly common in formal and academic writing - which is often far from 'plain English'. A combination of formal phrasing (including passive voice) and 'jargon' may be what students understand to be 'appropriate academic style'. The student may simply be trying to achieve a 'proper academic style' (or 'voice'), perhaps based on (outdated) formal style guidelines or information about the proper connectives in arguments from school (this could well explain 'thus'). Mimicking what one read last is typical for inexperienced writers 'finding their voice' (see, e.g., King, 2010), so it seems reasonable that a student still learning to write 'academically' will mimic one or more of the papers they refer to most.

I also know from my own writing that I go through phases where I use a few words or phrases a lot more than I do at other times, so words used unusually often in one text is not necessarily about generative AI use. It takes more than a few 'characteristic words' to be able to suspect 'this was not written by a human'.

Indeed, that students can try too hard, and produce worse work as a result is not new: Kitching (2008) reported students trying too hard to be properly academic and making their work worse as a result — in this case it was about including a confused theoretical discussion, instead of presenting projects as straightforward empirical work, but I think the point of trying too hard to mimic what is perceived as 'correct' and ending up with something that doesn't 'ring true' as it were remains.

But there are questions you, as a teacher, have to ask yourself. It's clear now that at least some students will use generative AI for written assignments, but there are a number of ways in which they may use it, all with different implications. I saw a presentation (by André Mathe of the University of Bergen) a few months ago where some lecturers in Norway were interviewed about appropriate uses of generative AI. If I remember correctly, it was typically considered entirely acceptable to use genAI for text editing, and even for idea generation, though the lecturers were pretty much in agreement the the first full draft of the assignment should be the students' own work. So for those lecturers, a student drafting an assignment in everyday, informal language, and then uploading to genAI with a prompt along the lines of 'make more academic' would have been acceptable — and an over abundance of words such as 'thus' and 'leverage' (I'm guessing more common in academic texts than anywhere else) could appear.

There are also many things to consider when it comes to exactly why you're inclined to raise the issue to what you suspect may be an AI generated text: a) whether it matters, b) why it matters, c) how it matters, and d) whether (and why and how) it should matter.

But the damage you have done your student and their trust in you and perhaps even other academics, however accurate or inaccurate your questioning of them may have been, should not be underestimated. After all, you have the power.

• Kitching, G. (2008). The trouble with theory: the educational costs of postmodernism. Pennsylvania state university press.
• King, S. (2010). On writing, a memoir of the craft. 10th anniversary edition. Scribner.
• André Mathe's talk announcement: https://www4.uib.no/en/research/research-groups/teaching-and-learning-in-higher-education-teled-research-group/events/this-is-not-mine-exploring-the-challenges-metaphors-and-affordances-of-generative-ai-in-design-education; a publication with the same title is somewhere in preparation, so hopefully his observations will soon be available to all.