Two families. Two teachers. The same suspicion, handled in two completely different ways. One teacher has draft history and a clear record of the student's work. The other has a screenshot of a percentage and a bad feeling.

The school now owns two problems instead of one. The second family will find out how the first was handled, and the difference will be very hard to explain.

What makes a writing-evidence practice defensible

Most conversations about academic integrity in the AI era go straight to policy language. The more useful question is procedural: when a suspicion arises, what does the school actually look at, and can it show its reasoning afterwards?

Three properties do most of the work.

1. The same record format everywhere

If the evidence depends on which teacher was suspicious, the school has no practice, only a set of individual instincts. A single, consistent record of how writing was produced means two classrooms produce comparable material for the same question.

2. The student can see it

A record the student cannot read is not fairness, it is disclosure control. When students can open their own record and see the replay of their own writing, the conversation starts from shared facts instead of accusation.

3. Staff access is logged

Every time a teacher or administrator opens a student's writing record, that access is written into the student's own log. This is a small feature with an outsized effect: it makes staff behaviour accountable by default, and it removes the "were you looking at my work?" question from the room.

"

[!IMPORTANT] The distinction that matters most A writing record documents a process. It is not a detection tool, it does not produce a suspicion score, and it does not return a verdict. Nothing about it asserts that a student did or did not use AI.

What this changes in an actual meeting

Imagine a parent evening where a grade is in dispute.

Without a record, the conversation is about whether the teacher "felt" the writing was not the student's. The parent asks for proof, and there is none to give, because a percentage is a guess about text.

With a record, the same conversation is concrete. Here is the assignment. Here is the empty document at the start of the period. Here are the nine snapshots in between. Here is the twenty-minute pause. Here is the block of text that arrived as a paste. Here is the question the student asked about the membrane, and the answer they got. Then the question becomes answerable by everyone in the room, including the student.

That is a materially different meeting, and it is one a school can stand behind.

The equity argument, which is the strongest one

Detection scores do not fail evenly. Research on false positives has consistently found higher flagging rates for multilingual writers, neurodiverse students and students writing in non-standard dialects. A practice built on those scores concentrates its errors on students who already face the most barriers.

A process record does something different. It does not infer anything from style. It reports what happened, which means it cannot penalise a student for writing in a way that a model finds unusual.

If you need the evidence for that argument in writing, why AI detectors harm students collects the research and the cases.

What to tell teachers, in plain terms

The staff-room version of this is short.

• The record is context, not a conclusion. It helps you understand the work. It never accuses on your behalf.
• Read the record before the conversation, not during it, and tell the student you read it.
• Ask about the gaps rather than asserting the worst reading of them. Long silences and pasted notes usually have ordinary explanations.
• Never present a detector percentage as the finding. If it prompted you to look, say that, then show what you found.

"

[!TIP] One sentence for a staff meeting "We are not looking for a better way to catch students. We are building a shared way to read the work."

Where the limits are, stated clearly

Superintendents and school leaders will ask what the record cannot do, and the answer should be ready.

• It does not resolve authorship. Session records show an account, not an identity. Shared passwords remain a human problem, not a data problem.
• It does not capture preparation. Reading, discussion and paper planning happen outside the workspace and leave nothing behind.
• It does not replace professional judgement. A person decides, after reading.
• It is deliberately narrow. No keystroke profiling, no clipboard reads, no behavioural scoring. Data minimisation is part of the design, not a limitation to apologise for.

The full list of what is kept and what is refused is on the Fair Writing Record Assignment page, including the privacy details behind the access log.

A short checklist to take to a department meeting

1. Agree one record format for written assessment, and use it in every course that requires extended writing.
2. Make the student view of the record non-negotiable, and show students how to open it.
3. Tell staff that access is logged, and that logging is the point.
4. Replace "the detector said" with "here is what the record shows" in every integrity conversation.
5. Keep the final judgement human, documented, and reviewable by the family.

For the teacher-facing version of this argument, read What a Fair Writing Record Can and Cannot Show. For the student-facing version, read When the Process Is the Only Evidence You Have.

To see how the record informs judgement without replacing it, see the human in the loop.