AI Detector for Teachers: Free Class Links & a Fair Policy

Most detector workflows put the teacher in the role of investigator: collect the essays, run them through a tool, confront whoever scores high. ScanForAI's Class links reverse that. Students scan their own work, see the result first, and choose to send it to you. This page explains how the feature works, how to read what comes back, and how to write that into a classroom policy that will survive an appeal.

What Class links are

You create a class in your ScanForAI dashboard and get two things: a link in the form scanforai.com/c/K7M3PX and the same six-character code to read out in class. A student opens the link, pastes their essay, and sees their own result first: a 0 to 100 score, a confidence band, and the sentences the detector doubted, with the reasons. Then they decide. Send to my teacher shares the full report; Send score only shares the number. Nothing is sent or stored until they press one of those buttons, and the student keeps the same report you receive.

On your side the class table fills with names, scores and the full results for each line, and you can export all of it to a spreadsheet. An optional roster accepts only the names you list. You can close the class at the deadline and delete anything with one click. There is no plugin to install: in ScanForAI's words, it works alongside every LMS because it touches none of them.

It is free. A free account runs up to five open classes at once with up to 100 submissions each, and the account itself needs no card and has no trial that runs out. Pro removes the limit on open classes and raises each class to 1,000 submissions. Schools with real volume can talk to them about the Team plan.

Why students-scan-first is the fairer design

Every AI detector has the same failure modes, and ScanForAI lists its own openly: formal or non-native writing, heavily edited or newer AI output, and short texts, where under about 300 words most signals abstain or wobble. The 2023 Stanford study by Liang and colleagues found seven widely used detectors flagged 61% of TOEFL essays by non-native English writers on average. If you run essays through a detector in bulk and act on the top scores, the students you will confront first are disproportionately the careful, formal, non-native writers.

Letting students scan first changes what the score is for. It stops being evidence collected against them and becomes feedback they can act on before the deadline: a flagged sentence is usually a generic one that could appear in any essay on the topic, and replacing it with the student's own example or opinion makes the essay better regardless of who wrote it. It also means a student accused later has the same report you have, which is the single most useful thing in an integrity meeting. ScanForAI publishes no accuracy percentage and says why: the number would only be true for its own test set. Its teacher guidance is blunt: never grade or punish a student on a score alone. Read our longer explainer on why detectors flag human writing if you want the mechanics.

Set up a class in five minutes

1. Create a free account at app.scanforai.com/register. No card. New accounts get five deep scans until the email is confirmed, so confirm it.
2. Open Classes in the dashboard and create a class. Name it after the course and the assignment, since you will have several open at once.
3. Add a roster if you want only listed names accepted. Skip it for a low-stakes practice round.
4. Tell students to set the Text type to Student essay and to tick English isn't my first language when that applies. Both change how the signals are weighed and both are ScanForAI's own advice.
5. Share the link or code wherever you already talk to the class. Tell them explicitly that nothing is sent until they press Send, and that they may choose Send score only.
6. Close the class at the deadline and export the table to a spreadsheet alongside the submissions themselves.

Reading the class table

The score is a 0 to 100 signal, not a probability and not a share of the text, and it arrives with a confidence of high, moderate or low. The published bands are Low under 25, Moderate 25 to 49, Elevated 50 to 71 and High 72 and over. ScanForAI's own reading guide calls Moderate normal for edited or formal writing, and says Elevated and High are a reason to ask for drafts, notes or version history, never a verdict. Short texts, and scores in the unclear middle around 30 to 55, get low confidence, and low confidence means not enough to say, never "human". ScanForAI weighs dozens of language signals in five families, which it calls rhythm, word choice, phrasing, easy-to-guess and structure, and it does not publish the weights. So treat the table as a sorting aid, not a verdict list:

• Low confidence means the detector is unsure. A score of 70 at low confidence tells you less than a 55 at high confidence.
• Score-only rows carry less information. The student chose not to share the line-level view. That is their right under this design; it is not an admission.
• Open the full report before forming an opinion. The line view shows which sentences were doubted and why. Generic topic sentences and textbook transitions are the usual culprits.
• Compare with the student's earlier work. A sudden change in rhythm and vocabulary is a far stronger signal than any single score.

A classroom policy you can paste

AI assistance. You may use AI tools for brainstorming, outlining and language checks unless an assignment says otherwise. Say so in a one-line note at the end of the work. Submitting AI-generated text as your own writing is not allowed.

Detectors. We use an AI-writing detector as one signal among several, never as proof. You will be able to scan your own work first through a class link and decide whether to share the result.

If your work is flagged. You will be asked about your process and invited to show drafts, version history, notes and sources. You may be asked to discuss the work or write a short passage under supervision. No mark is changed on a detector score alone.

Keep your drafts. Write in a tool with version history and keep your research notes until the grade is final.

What to do with a high score

1. Open the full report and read the doubted lines. Ask yourself whether they are generic or genuinely out of character for this student.
2. Pull up the student's previous submissions and compare vocabulary, sentence rhythm and argument style.
3. Talk to the student. Ask how the piece was written and ask to see version history, notes and sources. Students who used AI for brainstorming often have the chat log; our chat export cleaner turns it into a readable PDF.
4. If doubt remains, ask for a short supervised writing sample or an oral discussion of the argument.
5. Decide on the whole picture. If the only evidence is the score, the policy above says the mark stands.

Limits and honest caveats

• Scoring covers English, Spanish and Turkish only.
• There is no Google Classroom, Canvas or Moodle integration by design; you share a link or code.
• Deep scans are metered: a free account gets 30 an hour and 300 a month. Fast scans are unlimited. A class of 100 submissions sits comfortably inside the monthly limit, but confirm before a whole department starts using one account.
• Five open classes at a time on the free plan. Close finished classes to free a slot.
• No detector, this one included, can prove authorship. The design above is built around that fact rather than against it.

See Class links on ScanForAI

Related guides: Why AI detectors flag human writing · Essay flagged as AI? What students should do · Image prompt injection: how to check an image.

Disclosure: ScanForAI is a partner site. Links on this page carry tracking tags so they can see visitors arriving from SimpleToolsHub; we don't earn a commission, and Class links are a free feature. Feature descriptions and limits are as published on scanforai.com in November 2026. We make no accuracy claims for any detector.