AI Detector for Teachers
A detector percentage is the easiest evidence to reach for and the weakest evidence to rely on. This page is about what the number actually measures, what it cannot establish, and how to keep a flagged submission from turning into an accusation.
In short: AI detectors estimate how statistically predictable a piece of text is. They return probabilities, not proof of authorship, and they shift when text is edited. Used well, a detector is a prompt to look more closely at a draft and talk to the student about process. Used as a verdict, it produces false accusations. DeAIze shows which sentences carry the signal, so a conversation can start from specific lines rather than a single number.
Sentence-level, so you can point at something
Instead of one figure, you see which sentences read machine-made and which read like the student.
A teaching tool, not a tribunal
The same view a student sees: flat, uniform prose, and the specific lines that produced the score.
Built to be shown, not hidden
Detection is free and needs no account, so a class can look at the same text together.
The rewrite still has to be the student's
Rewrites are gated on semantic similarity, so a passage cannot be swapped for one that says something else.
| Signal | What it can suggest | What it cannot prove |
|---|---|---|
| A high detector score | The prose is unusually uniform, or was heavily polished | Who wrote it |
| A sudden change of voice between drafts | A different process, or heavy editing | That AI produced the new version |
| Fluency far above a student's earlier work | The process is worth asking about | Misconduct |
| A low detector score | The text did not trip that model today | That the work is the student's own |
What the percentage measures
A detector does not recognise authorship. It measures how predictable text is: how surprising each word is given what came before it, and how much sentence length and structure vary across a passage. Human writing tends to be lumpy — a long sentence, then a short one, then a fragment. Machine writing tends to be even.
That definition has consequences worth telling students about. Carefully proofread prose, boilerplate essay structures, and writing produced under a "no short sentences, no contractions" rule all look flatter than average, and flatter text scores higher. A high score is a statement about texture. It is not a statement about who typed it.
Handling a flagged submission
Start with the work, not the score. Ask the student to talk through the argument, to explain a source, to say what they would change. Ask for drafts, notes and reading lists if the assignment allowed process evidence. Someone who wrote the piece can usually say where a paragraph came from; someone who did not often cannot.
Then follow your institution's policy, and be explicit about which part of the policy is in question — using a tool to check grammar and having a tool write the argument are different matters, and only one of them is usually prohibited. A detector result on its own should not be the finding. It starts the process; it does not conclude it.
Designing assignments that make authorship visible
The most reliable defence against an unanswerable "did you write this?" is an assignment where authorship shows. A short in-class writing sample in the same register, a proposal and an annotated bibliography before the final draft, a paragraph of reflection on which source changed the student's mind, a marked-up draft with comments.
None of these require surveillance software. They produce a trail of thinking that a detector cannot fake and that a plagiarised or machine-written submission struggles to reproduce, because the trail is about decisions rather than prose.
Where a rewrite tool fits
When a passage is genuinely the student's but reads flat — long, even sentences, stock transitions, no specifics — rewriting it is normal editing. DeAIze highlights those sentences and rewrites them, keeping a rewrite only when semantic similarity confirms the meaning is intact, then re-scores the document so the change is visible.
Two ground rules keep that honest. The ideas, data and conclusions must be the student's own, and the policy on AI assistance still applies while they use the tool. Both are worth saying out loud in class rather than assuming.
What DeAIze is not
It is not a grading tool, not a plagiarism checker and not a misconduct finding. It does not know who wrote a text, and it will not produce an accuracy figure, because no detector has one worth quoting — scores move when the text moves.
What it does give a class is a shared object to look at: a document with the flat sentences marked and a score attached. That is enough to teach the difference between prose that sounds like a person and prose that sounds like an average. It is not enough to accuse anyone, and it should never be used that way.
Look at a draft together, sentence by sentence.
Detection is free and needs no account. Paste text, see which sentences carry the signal, and start the conversation there.
Get started freeRelated guides
Does Turnitin Detect ChatGPT? What Students Need to Know
Yes, Turnitin's AI writing indicator can flag ChatGPT text — but it's a probability, not proof. Here's how it works, why scores vary, and how to lower yours.
AI Detector Accuracy: How Reliable Are GPTZero, Turnitin and Others?
AI detectors are probabilistic tools with real false-positive rates. Here's what research says about accuracy and how to protect yourself from a wrong flag.
Undetectable AI for Students: Work That Passes on Its Own Merit
Students want writing that reads human, not a trick. Here's how to make AI-assisted work genuinely undetectable — remove the tell, don't hide it.
Frequently Asked Questions
Should a detector score be used as evidence of misconduct?
No. Detector output is a probability estimate about writing style, not proof of authorship. Treat a flag as a reason to look at the draft and talk to the student, and follow your institution's policy for anything formal.
Why does careful, well-edited student writing score higher?
Because detectors measure predictability and variance. Prose that avoids contractions, keeps sentences similar in length and uses standard academic transitions looks more uniform, and uniform text scores higher regardless of who wrote it.
Is DeAIze free for a class to use?
Detection is free and needs no account: paste text and read the sentence-level result. Humanizing and re-scoring use credits — 200 words are free on signup, and paid credits start at $20.99 for 1,000 words.
Does DeAIze tell students how to avoid detection?
It shows which sentences read machine-made and rewrites them. It does not help anyone pass off work they did not do, and the ideas, data and conclusions in a submission still have to be the student's own.