A student emails you at 11pm. Their essay came back flagged — "82% likely AI-generated" — and they didn't use AI. They wrote it over a week, at the library, with their own sources. Now they have to prove a negative to a machine that won't explain itself. What do you tell them?
This happens constantly, and it happens because AI detectors are answering an impossible question. They look at the finished text and try to reverse-engineer its origin from word choice and sentence rhythm. But careful academic prose and machine prose look alike — both are fluent, hedged, and structured. So detectors flag honest students who write cleanly, and wave through dishonest ones who paraphrased the output a little. False positives and false negatives, at the same time, from the same tool.
You can't fix this by tuning the threshold. The information isn't in the finished text. Here's the same document, seen two ways:
“Retrieval practice reliably outperforms re-reading for long-term retention, an effect that holds across ages and materials…”
The detector never saw the writing happen. It’s guessing from the words on the page — and it guesses wrong in both directions: flagging careful human prose, missing text that was lightly paraphrased. A number, dressed up as proof.
The detector on the left is guessing from the words. The record on the right isn't guessing at all — it's what actually happened while the document was written.
The question a detector can't answer
"Is this text AI?" is unanswerable from the text alone, and it's also the wrong question. What a teacher actually wants to know is: did this student do the work? That question does have evidence — it's just not in the final paragraph. It's in the process. Did the draft grow over time or arrive in one paste? Were the sources real, and did the student engage with them? Is there a revision history that looks like thinking?
None of that is visible to a detector, because a detector only ever meets the document at the finish line. But it's all visible if you were watching the whole race.
Positive evidence, not a probability score
That's the flip we made in Folio. Instead of scoring the finished text, we keep a record of how it was made — and hand that record to the student as something they own, not something used against them.
- A revision timeline. Steady drafting across sessions looks nothing like a single burst. We don't accuse; we just show the shape, and let a human read it.
- Per-citation verification. Every reference is checked against CrossRef and the retraction record. "The sources are real" becomes a claim anyone can check, not one you take on faith.
- An integrity certificate the student attaches to their submission — proof they can point to, instead of a flag they have to explain away.
The difference in posture matters. A detector treats every student as a suspect until a model clears them. A process record treats every student as an author who can show their work. One of these is how you'd want to be treated.
What this looks like in a class
We built Folio Classroom around exactly this. You share one assignment link. Every submission arrives with its process record attached — the drafting, the citations, the timeline. You spend your time reading the work and the evidence of the work, not adjudicating a percentage from a black box.
It won't tell you a student is "82% honest," because that number would be as meaningless as the detector's. It tells you something better: here is how this document came to exist. Read it, and decide like the expert you are.
The honest students are the ones being hurt
It's worth being blunt about who pays for detector-based grading. The students most likely to be falsely flagged are often the careful, fluent writers — and the ones being missed are the ones who put in the least effort to disguise a paste. The tool inverts the thing you're trying to reward.
Detectors ask a question that can't be answered and then answer it anyway, confidently, in a way that lands hardest on the people doing it right. We'd rather ask a question that has a real answer — how was this written? — and let the answer speak for itself.
If you teach, Folio Classroom is free. If you write, your integrity certificate is already there on every document. Either way, the point is the same: you shouldn't have to prove a negative to a machine. You should get to show your work.