How Teachers Document AI Detection Decisions Fairly

When an AI detection tool returns a high likelihood score on a piece of student work, the temptation is to act on it straight away. But a likelihood score is the start of a process, not the verdict. What protects you, the student, and your school is not the number itself — it is the record you keep around it. Good ai detection record-keeping for teachers turns a contested moment into a defensible, fair decision.

This guide walks through what to record, when to record it, and how to align those records with your school's academic integrity policy, so that if a decision is ever challenged, your professional judgment is backed by evidence rather than a single screenshot.

Why a Written Record Matters

A likelihood score is probabilistic. It expresses how closely a piece of writing resembles patterns associated with AI-generated text — it is not proof that a specific student used a specific tool. That uncertainty is exactly why documentation matters. A clear record shows that you treated the score as one signal among several, applied your own professional judgment, and gave the student a fair hearing.

Without a record, two things go wrong. First, you cannot reconstruct your reasoning weeks later when a parent or an appeal panel asks why you reached a particular conclusion. Second, inconsistent handling across a department becomes invisible — one teacher escalating at 60% while another waits until 90% looks arbitrary unless the thresholds and reasoning are written down. Documentation is what makes a moderation process consistent and reviewable.

What to Record When a Score Comes Back High

A useful record captures the decision, not just the number. Aim to note the following at the point you review the work, while the detail is fresh:

• The likelihood score and the model used. GradeOrbit returns a score from 0 to 100% and offers a faster 1-credit check or a more thorough 3-credit analysis. Record which you ran, because the depth of the check is part of the context.
• The specific evidence in the work. Note what made you look more closely — an abrupt shift in vocabulary, claims the student could not explain in class, or formatting inconsistent with their usual style. The score points; your reading confirms or softens it.
• Context you already hold. The student's prior work, their performance in lessons, any known circumstances. A high score on a student whose classwork matches the submission reads very differently from a high score on work that is out of character.
• Your provisional conclusion and next step. Whether you are escalating, opening a conversation, or treating the score as inconclusive and moving on.

The aim is a short, factual note — not an essay. A few lines per case is enough to make the decision traceable. For more on reading the number itself, see our guide on how teachers handle a high AI likelihood score fairly.

Documenting the Student Conversation

The most important record is the conversation. A high score should open a discussion, not close a case. Ask the student to talk you through their process, their sources, and their drafting. Record what they said, the date, and whether their account was consistent with the work in front of you.

This protects everyone. A student who can explain their work has a documented defence. A student whose account does not hold up has been given a fair chance to respond before any consequence is applied. And you have a written basis for whatever you decide next, rather than relying on memory. The conversation note is often the single piece of evidence that an appeal panel weighs most heavily.

Aligning Records With School Policy

Your records only carry weight if they map to your school's academic integrity policy. Before you rely on detection scores at all, the policy should answer: what score, combined with what other evidence, triggers a conversation? Who reviews escalations? What consequences follow a confirmed case, and what happens when a case is inconclusive?

If those answers are not written down, the records you keep will be inconsistent because the standard is inconsistent. A shared threshold and a shared escalation route mean every teacher's documentation looks the same shape, which is what makes the whole fair process ai detection approach hold up under scrutiny. Departments that standardise this early avoid the situation where two students with identical scores are treated completely differently.

How GradeOrbit's Likelihood Scores Fit the Record

GradeOrbit is built to support documentation rather than replace your judgment. The detection tool returns a clear 0 to 100% likelihood score, with a lightweight 1-credit check for triage and a deeper 3-credit analysis when a case warrants a closer look. Crucially, GradeOrbit never stores the student work you upload, and personal information is redacted client-side before anything is processed — so building a record does not mean building a privacy risk.

The score gives you a defensible starting point; your evidence, your conversation note, and your policy alignment give you the rest. Used together, they turn a probabilistic signal into a decision you can stand behind. GradeOrbit does not make the call for you — it gives you a consistent, reviewable input to a process you control.

Build a Fairer Detection Process With GradeOrbit

If your current approach to AI detection is a screenshot and a gut feeling, it will not survive a serious challenge. GradeOrbit gives UK secondary teachers a clear likelihood score, privacy-first processing, and a depth of check you choose per case — the inputs a defensible record needs. Pair it with a shared policy and a documented conversation, and you have a process that is fair to students and protects your professional judgment. Visit GradeOrbit to see how the detection tool works.