What Teachers Do When a Student Denies Using AI

The hardest moment in AI detection is not the high score. It is the moment after — the conversation where the student looks you in the eye and says they wrote every word themselves. The AI likelihood score on your screen says 84%. The student says zero. Now what?

This is the situation almost no detection guidance covers honestly. Most posts stop at "interpret the score carefully" and leave you to figure out the human bit yourself. This one walks through what UK secondary teachers can actually do when a student denies using AI, how to gather evidence beyond a single number, and how to keep the process fair to a student who might be telling the truth.

The First 24 Hours After a High Likelihood Score

Before anything else, slow down. A likelihood score above 70% is a strong statistical signal that the text resembles AI-generated writing, but it is not proof and it is not a confession. Acting on it within the same lesson, or sending a snap email to parents the night the score comes back, is how good teachers end up apologising publicly a week later.

Use the first day to do three things quietly. Re-read the work yourself, without the score in front of you, and note what feels off in your own words — sudden vocabulary shifts, perfectly balanced paragraphs, examples that do not match the texts you taught. Pull any earlier work from the same student so you have a baseline to compare against. And check whether the academic integrity policy at your school requires a specific procedure before any conversation happens at all.

For more on reading the score itself, see our guide to how AI detection likelihood scores work.

Running the Conversation Without Accusing

Open the conversation with curiosity, not certainty. "I want to understand your essay better — can you walk me through how you wrote it?" gets more honest information than "the detector says you cheated." A student who used AI will usually struggle to explain choices in their own text; a student who wrote it themselves will explain it easily, including the bits that are clumsy or unfinished.

Ask process questions: which paragraph did you write first, what made you change your conclusion, where did you find the quote on page two. Ask them to talk you through a specific paragraph and explain why they used a particular word. None of this requires you to mention the detector at all in the first pass. The goal is to gather information, not to extract a confession.

If the student denies AI use after this conversation, take that seriously. Note exactly what they said, when, and what evidence they offered. You are now in evidence-gathering mode, not enforcement mode.

Building Evidence Beyond the Score

One number on its own cannot carry a misconduct case. UK secondary schools that handle this well build a small evidence pack before any formal step. The pack typically includes:

• The likelihood score itself, with the model used and the date.
• Two or three earlier pieces of work from the same student, ideally produced in class under supervision, to establish a writing baseline.
• The drafting history if the work was done in a school-managed system that records edit timestamps (Google Docs version history is the most common).
• Your own annotated read of the flagged piece, noting specific passages that feel inconsistent with the student's usual voice.
• A short note from the conversation: what the student said, what they could and could not explain.

If the evidence pack still points one way, you are on firmer ground. If it points in different directions — high score but consistent voice, plausible drafting history, confident verbal explanation — the right call is often to give the benefit of the doubt and add lighter-touch monitoring to the next piece of work. False positives happen, and an unfair accusation costs more than a missed cheat.

When to Escalate vs When to Coach

Escalation is for cases where the evidence pack is consistent across multiple sources and the student's account does not hold up under gentle questioning. At that point, it is no longer a classroom conversation — it is a safeguarding-shaped process that needs your head of department, the academic integrity lead, and your school's documented procedure.

Coaching is the right call for everything else. Many students experiment with AI without understanding it counts as misconduct, especially in Year 9 and Year 10 when the rules feel abstract. A clear, calm conversation about why the work has to be theirs — and a fresh piece of work done under supervised conditions — resolves more of these cases than any disciplinary route ever will.

For schools building a consistent approach across departments, see how schools can implement AI detection consistently.

Aligning With Your School's Academic Integrity Policy

Whatever route you take, the policy has to lead. If your school's policy says a single detection score cannot trigger a misconduct allegation on its own, do not let one trigger it. If the policy requires two members of staff to review the evidence before a parent is contacted, get the second pair of eyes. Procedures exist because they protect students from teacher mistakes and protect teachers from challenge later — and a denial case is exactly when both protections matter most.

If your school does not yet have a written policy, this is the case that proves you need one. The denial scenario is where ad-hoc decisions go wrong. Use what you have learned to push for a documented procedure before the next high score lands.

The Outcome Is Not Always a Verdict

Some of these cases end with a student admitting they used AI, a clear consequence, and an honest re-do. Some end with the teacher concluding the work was genuine after all, and the relationship with the student is stronger for the process. Some end inconclusively — the evidence is genuinely mixed and the right call is to monitor closely next time. All three are legitimate outcomes. What is not legitimate is treating one number as a verdict and skipping every step in between.

Use GradeOrbit's Likelihood Score Alongside Your Professional Judgement

GradeOrbit's AI detection tool gives you a likelihood score between 0% and 100%, with a choice between a faster 1-credit check and a more thorough 3-credit check for cases that need extra confidence. The score is designed to inform your judgement, not replace it — exactly the kind of input you need when a student denies and you are deciding what to do next.

Visit the GradeOrbit homepage to try it on a piece of student work and see how the score fits into the fair process your school already runs.