# AI Detector for Teachers — Evidence, Not Accusations

An AI checker built for discussion, not accusation: per-line evidence you can show a student, honest accuracy limits, and a free deep scan.

- Source: https://scanforai.com/for-teachers
- ScanForAI — AI-writing detection. A score is a signal, not a verdict.

---

# An AI checker for teachers who don't want to accuse anyone

You already know the uncomfortable truth: detectors get it wrong, students get hurt, and a bare percentage can't be discussed — only believed or disputed. ScanForAI is built for the conversation instead: every score comes with the specific writing patterns behind it, in plain English, free.

## Why a breakdown beats a percentage

When a submission scores high, ScanForAI shows *why*: sentences with unusually even rhythm, stock AI phrasing, structure that follows the default machine template, wording that's statistically too predictable. That gives you discussion points — "walk me through how you wrote this paragraph" — instead of a courtroom number. Students who wrote their work honestly can usually explain it; that conversation is more informative than any score.

## What to be careful about (we mean it)

- **Non-native English writers score higher on every detector.** Independent research has shown other tools flagging non-native writing at alarming rates. We test against non-native writing specifically and relax formality signals by register — but the risk never fully disappears, ours included. Weight it accordingly.
- **Formal writing looks "AI-ish" by nature.** Tell the scanner it's an essay (the "Text type" selector) so it doesn't punish a student for writing formally.
- **Never grade or discipline on a score alone.** Our [interpreting guide](https://scanforai.com/docs/interpreting-scores) is written to be shareable with colleagues and students.

## A workflow that holds up

1. Scan the submission (free, up to 20,000 characters — more with a free account). Set the text type.
2. Read the *signals*, not just the score. Look at the sentence heat-map.
3. If it reads AI: have the conversation, with the specific passages as prompts. Ask about drafts, sources, process.
4. Keep a [free account](https://app.scanforai.com/register) to save scans — history and revision comparison document the process if questions come later.

## Class links: results your students choose to share

Create a class in your [dashboard](https://app.scanforai.com/classes) and you get a link and a six-character code — post it in your LMS, put it on the projector, print it on the assignment sheet. Each student scans their *own* work, sees their *own* result first, and presses **Send to my teacher** when they're ready. Your class table fills with names, scores and the full per-line analysis; export the lot to a spreadsheet.

- **Consensual by construction.** Nothing reaches you the student didn't choose to send — and they hold the same report you see. Most detectors hide the score from students; we lead with it.
- **Your conditions.** Optional roster (only listed names accepted), close the class at the deadline, delete anything with one click.
- **No IT ticket.** Works alongside every LMS because it touches none of them. A free account runs one open class; Pro removes the limits.

## Why free?

ScanForAI is free — the whole scanner, accounts, history and reports. Our engine runs on our own hardware (no per-scan AI-API costs), which is also why anonymous scans are never stored: your students' text isn't our data.

Per-line evidence you can put in front of a student — a conversation starter, not a verdict.

## Detection in the classroom.

Not reliable enough to act on alone — ours included, and we say so on every result. Independent testing shows all detectors miss edited AI and flag some honest writing. Treat a score like a smoke alarm: a reason to look closely, never the verdict itself.

Gather evidence beyond one score: run a second detector, compare against the student's earlier writing, look at draft history if your LMS keeps it, and talk to them about the content — someone who wrote an essay can discuss it. Our per-line marks give that conversation something concrete to point at.

Learned formal English is careful, even and template-shaped — exactly the surface habits of machine text. Every detector shows this bias; we document it openly and mark the specific lines so you can judge whether they read 'machine' or just 'careful'.

Paraphrasing tools lower scores without making text human — a decent detector still reads elevated, and the writing gets worse, which your own eyes catch. The honest arms-race answer: detection plus judgment beats detection alone, which is why we give you reasons, not just numbers.

You create a class and share a link; students scan their own writing and choose to send you the result — score, analysis and text, or score only. You get the table and the export; they keep an identical copy. Nothing is stored until a student presses send, which keeps our no-storage promise intact and makes the whole thing consensual by design.

Yes — the scanner is free anonymously, and a free account adds batch scanning (a whole class set at once), history, PDF reports and the full pattern breakdown. No card, no trial clock.