If You Are Wrongly Accused
AI detection tools are unreliable and misjudge second-language writers most often. What to know, and what to do.
This chapter exists because the problem is real and under-discussed: students are accused of using AI on work they wrote themselves, on the evidence of tools that do not work reliably.
If it happens to you, understanding why it happens is most of your defence.
How detectors work, and why they fail
Detectors measure statistical properties of text — roughly, how predictable each word is given the previous ones. Very smooth, very predictable text scores as machine-written.
The problem is that plenty of human writing is smooth and predictable: formal academic prose, technical writing, and writing by people composing carefully in a language they learned rather than grew up with.
Why the arms race cannot be won
Anyone deliberately evading detection does so trivially: rephrase, add variation, pass it through another tool. Detection catches the careless, not the deceptive.
So the false positives land on honest students while the intended targets pass. That is the opposite of what the tool is for, and it is why a detector score should never be treated as evidence.
Protect yourself in advance
- Write in a document with version history, so your revision timeline exists automatically
- Keep your outline, notes and research separately
- Save drafts at milestones rather than overwriting one file
- Keep the sources you consulted, with your own annotations
- If you used AI legitimately, note what for — a one-line record costs nothing
If you are accused
- Stay calm and do not admit to something you did not do to make the situation end
- Ask what specific evidence the accusation is based on. If it is a detector score, ask what its false-positive rate is
- Offer your drafts, version history, notes and sources
- Offer to discuss the work in person — explaining your argument and your choices is far stronger evidence than any document
- If English is not your first language, say so explicitly and point out that detectors are documented to misjudge second-language writing
- Ask about the appeals process, in writing, early
For teachers reading this
Do not use a detector score as evidence. Use it at most as a prompt to have a conversation — and never confront a student with a number as proof.
The durable answer is assessment design: drafts, process, in-person discussion, and tasks anchored in personal or local context that a general model cannot supply. That is more work than running a detector, and it is the only approach that is both fair and effective.
What to take from this chapter
- Detection tools are unreliable and disproportionately misjudge second-language writers
- Keep version history, drafts, notes and sources as routine protection
- If accused: ask what the evidence is, offer your process, and request an in-person discussion
- A detector score is not evidence and should never be treated as proof
Try it
Check whether the tool you write in keeps version history, and turn it on if not. Then save a dated copy of your current draft. This takes two minutes and is the single best protection available to you.