Accused of writing with AI: what you can actually show
A detector score on its own is not proof and cannot establish or refute anything. What can be shown is the process: the document's version history, notes and source material, references, working outlines, and the ability to talk about the content without preparation. The strongest defence is showing how the text came about, not arguing about a number.
What holds up as evidence
Version history is the strongest. A document in Google Docs, in Word with autosave, or in git shows that the text grew in pieces over time. Notes, handwritten material, e-mails with a supervisor and links to the sources you used all help.
What does not help: running the text through another detector and bringing back a lower number. It only shows that tools disagree, and the other side can produce yet another number, this time higher.
How to talk about it
Ask about specific passages. Which section is suspect, which tool flagged it, and from what threshold that tool calls a text machine-made. A general claim that the detector showed it cannot be refuted until you know what exactly is being discussed.
Stay with the content. Offer a conversation about the work or a short defence — for an honest author it is the fastest way out, and for the other side it is something they can actually work with.
What to do next time
Write in a versioned editor and do not delete your notes. The trail builds itself and costs you nothing until you need it.
If you used AI for proofreading or research, say so in the work itself. Most rules allow that use when it is declared; what causes trouble is leaving it unsaid.
FAQ
Do I have to prove that I wrote it?
Under most academic codes the burden is on whoever raised the suspicion. In practice, though, your own material ends the dispute faster than an argument about who has to prove what.
Can a score alone be grounds for failing someone?
It should not be. No AI text detector is a hundred percent accurate and false positives on honest texts are documented. A result is a reason to look closer, not a verdict.
What if we have no rules for AI at all?
Ask for them first. When it is not written down what is allowed and how suspicion is handled, no case can be judged fairly.
Try it on your own text
Paste a text or upload a file and look at the score and at the specific sentences that came out suspicious.
Run a detection