What do I do when a replay looks wrong but I cannot prove anything?

Open the session with the student and ask them to talk you through it. A replay gives you a specific, factual opening: the window lost focus eleven times, and 340 characters arrived in one edit at 10:22. Most conversations end there without an accusation ever being made, because the student is explaining events rather than defending a detector score. If the explanation is reasonable, the record supports them just as well as it implicates anyone else.

How do I handle students who need extra time, dictation, or a phone?

Set the time limit generously or leave it off; nothing about the recording depends on it. Dictation and other accessibility tools insert text in blocks, so they will be flagged as large insertions, which means you should know in advance which students use them and read those reports accordingly. Phone typing, including predictive text, arrives in small chunks and counts as typing. The point of the record is that you can see what happened and apply judgment, not that a flag decides anything.

The AI-proof in-class essay

Q: Is an in-class essay really AI-proof?

No, and it is worth being blunt about that. A student writing on a school Chromebook has a chatbot one tab away, and a phone in a pocket. What the in-class structure does is remove the easy path: the prompt is new that day, the work happens in a room with an adult in it, and the session is recorded, so pasting stops working and retyping from a second screen leaves a visible trace. That is a large reduction in cheating, not a wall.

The shape teachers keep converging on: reading and notes at home, the draft written in the room, the prompt shown that day, and the session recorded. This is what each part is for, and what it still does not fix.

The split

Homework becomes input, not output. Students read, annotate, gather quotations, and bring notes. None of that is graded for authorship, so a chatbot summarising the reading is not the crisis it used to be — a student who arrives without having read has notes that fall apart the moment they try to build an argument out of them.

The essay is written in one period, in the room. That removes most of the surface area: no overnight window, no shared doc a sibling can open, no "my internet went out". The cost is real — a period is not enough for a polished five-paragraph essay, so the assignment shrinks to one strong section or a timed response, with revision handled separately once authorship is established.

Why the prompt is revealed in class

A prompt a student has had for a week is a prompt a language model has had for a week. Announce the text and the skill in advance; hold back the actual question until the period starts. Prompts naming a specific passage, a line number, or something said in the room yesterday are much harder to pre-generate than "write about ambition in Macbeth". This costs nothing and is the highest-leverage change on the list.

Recording the session, and saying so

Blocking paste without telling anyone is a trap. Saying it at the start — "this page blocks paste and records typing, tab switches, and paste attempts, and I can watch it back" — is a deterrent, and it is the version that survives a conversation with a parent. It also lets the students who were going to do it honestly stop worrying about being suspected, which is the group most often hurt by detector-based approaches.

WrittenLive is the recording piece. You type a title, send one link, and students write with no login on whatever device they have. Paste, drop and drag-in are blocked and counted. Afterwards you get a scrubbable keystroke replay, a class table with words, active minutes, percent typed versus inserted, tab switches and idle gaps, and a printable one-page report per student. No AI judges the text anywhere in it — see how a session runs.

From a hunch to a conversation

The usual failure mode is an accusation built on a feeling: this does not sound like you. The student denies it, you have nothing further, and either you drop it or it escalates on the strength of a vibe.

A replay changes what you open with. Instead of a verdict you have events: the tab was left eleven times, 340 characters arrived in one edit at 10:22, the rest was typed at an even pace with almost no deletions. Put that on a screen and ask the student to walk you through it. Often that is the whole meeting. Sometimes the explanation is good and the record clears them — a clean replay is the strongest thing a wrongly-suspected student has ever been able to offer.

Honest FAQ

Is an in-class essay really AI-proof?

No. A Chromebook has a chatbot one tab away and a phone sits in a pocket. The structure removes the easy paths — new prompt, adult in the room, paste blocked, session recorded — so cheating gets slower and leaves traces. It is a large reduction, not a wall, and any tool that tells you otherwise is selling something.

What if a student retypes from their phone?

It counts as typed, because it was typed. What shows up is the shape: an unusually even pace, very few deletions, few long thinking pauses, often with tab-aways or hand-off-keyboard gaps between chunks. Worth asking about; not proof, and the report does not call it proof.

What do I do when the replay looks wrong but proves nothing?

Sit with the student and the timeline and ask them to explain the specific moments. Facts with times on them are a much better opening than an accusation, and they leave room for the answer to be innocent.

Students with extra time, dictation, or accommodations?

Leave the time limit off. Dictation tools insert text in blocks and will be flagged as large insertions, so know in advance who uses them and read those reports with that in mind. Flags are prompts to look, never decisions.

Do I have to stop assigning take-home essays?

No. Establish authorship once in class, then let take-home work be revision, research, and longer projects built on a draft you watched being written. Several teachers described exactly this: one recorded in-class draft per unit, everything else normal.

What does it cost?

Nothing for one class this school year. No card field, no trial clock. A department version with shared dashboards would eventually be paid; your free class keeps working either way.

Made by a developer after teachers asked for it. I have never taught a class. Everything above about the classroom comes from teachers describing what works for them; the only part I can speak to first-hand is the recording. If a piece of it is wrong, tell me and I will change it: lehrerenterprise@gmail.com.

Set one up for tomorrow's period

Prefer to look before you touch anything? The public demo dashboard has two sample sessions in it, one clean and one flagged.