How to Use AI Detection in Your School Fairly

AI detection is now part of the landscape in UK secondary schools. ChatGPT, Claude, Gemini, and a growing list of similar tools are accessible to every student with a smartphone, and the writing they produce has become sophisticated enough that experienced teachers routinely struggle to identify it by eye alone. The tools that detect AI-generated content have improved in parallel - but a detector without a clear, fair process behind it is not just unhelpful. It can be actively harmful.

This guide is for secondary school teachers and leaders who want to use AI detection fairly: not as a gotcha mechanism, but as one evidence-based input into professional judgement. The goal is not to catch students. The goal is to uphold the integrity of assessment in a way that is defensible, consistent, and kind.

Why Fairness Has to Come First

AI detection tools produce a likelihood score: a probabilistic estimate, not a verdict. No tool currently available, including the most sophisticated commercial platforms, can tell you with certainty that a specific piece of writing was generated by AI. What they can tell you is that a piece of writing shares statistical features with AI-generated text to a greater or lesser degree.

That distinction matters enormously in a school context. A student who writes in a formal, well-structured register - perhaps because they have been explicitly taught to - may score higher on an AI detector than a student who writes more colloquially, even if neither used AI at all. Students who use AI as a drafting aid and then substantially rewrite the output may score lower than students who used no AI but happened to produce writing that resembles AI outputs. Treating a likelihood score as proof of wrongdoing, without further investigation, is both professionally indefensible and potentially very unfair to individual students.

A how to interpret AI detection scores uses the result as the beginning of an investigation, not the end of one.

What a Good AI Detection Result Actually Tells You

The most useful AI detection tools do more than return a single percentage, and the fairest ones measure work against something meaningful. GradeOrbit's detection, available on school accounts, compares a pupil's new work with that pupil's own earlier work already marked in GradeOrbit, and returns a score from 0 to 100, a confidence rating (Low, Medium, or High), a list of the specific signals it noticed, and a plain-English summary explaining the overall assessment. A pupil with no earlier work on file is skipped, and the teacher is told, rather than being given a generic score.

Understanding what each of these elements means in practice helps you make better use of the result.

The score indicates how far the new work departs from the pupil's own earlier writing. A low score suggests the writing is consistent with how that pupil usually writes. A high score suggests it differs in ways worth a closer look - but it is still not proof. Because the comparison is with the pupil's own work, a naturally formal writer is measured against their own formal writing, not against a generic idea of what AI sounds like.

The confidence rating tells you how reliable the score is likely to be given the length and nature of the text. A high-confidence result on a 600-word essay is more actionable than a low-confidence result on a 150-word structured response. Short texts inherently produce less reliable detection signals, and a responsible tool tells you when that is the case rather than projecting false certainty.

The signal list describes the specific linguistic features that contributed to the score - things like formulaic paragraph structure, even distribution of specification content, unusually low variation in sentence length, or absence of a personal voice. These are the details that allow you to build a professional case if you need one, and to have a grounded conversation with the student rather than simply citing a number.

Detection is unlimited and included in every school teacher seat, so cost never decides which pupils get checked. You can run it on a whole class at once for routine screening, or from a single pupil's profile for a piece of work where you are seriously considering raising a concern formally.

Building a Consistent School Policy

Individual teachers making individual decisions about when and how to use AI detection, based on their own threshold for suspicion, is a recipe for inconsistency - and inconsistency is unfair. A student in one teacher's class may face a formal conversation about a piece of work that would have passed without comment in another teacher's class, not because the work is different, but because the processes are different.

A fair school policy on AI detection addresses several questions in advance. Which assessments will be routinely screened? What score threshold triggers further investigation rather than automatic escalation? Who reviews flagged work, and who makes the final judgement? What happens during the investigation - is the student interviewed, and who is present? Is there a formal record kept regardless of outcome?

Involving Heads of Department in drafting this policy matters for two reasons. First, the norms around AI assistance vary considerably by subject. In Computer Science, using GitHub Copilot to autocomplete code may or may not constitute academic misconduct depending on the specific task. In History coursework, generating a draft argument and then editing it is categorically different from submitting an unmodified AI output. Second, HODs are the people who will need to support their teams in applying the policy consistently - and they will do that better if they had input into designing it.

Once a policy exists, it should be shared explicitly with students and parents. Transparency about what is being scanned, on what basis, and what the process is if something is flagged reduces the likelihood of disputes and signals that the school's approach is principled rather than arbitrary.

How GradeOrbit's Detection Tool Works

GradeOrbit's AI detection is part of its school accounts, built into the same platform as the marking workflow - which is also what gives it each pupil's earlier work to compare against. Teachers can screen submitted work without managing a separate subscription or workflow.

Work goes in as scans, photos or PDFs, including photographs of handwritten work taken with the phone camera link; typed work is uploaded as a PDF. This is important because not all suspicious work arrives as a typed document. A student who generates AI text on their phone and then copies it onto paper by hand, or who uses AI to produce a draft and handwrites a lightly edited version, may still produce work that differs from their usual writing. GradeOrbit reads the page images directly, so a handwritten script is checked just as a typed one is. For a class set, bulk upload splits the pages by pages per student and matches the name on each cover page to the class list.

Pupils' work is held under the school's Data Processing Agreement and is never used to train AI models. It is kept only for the school's chosen retention window and wiped on schedule or at offboarding. This is a deliberate design choice that keeps GradeOrbit's handling of student work consistent with UK GDPR obligations and the professional ethics of working with data about minors.

Having the Conversation With Students

If a detection result warrants a conversation with a student, the way that conversation is handled matters enormously - both for the student's experience and for the integrity of the outcome.

Start from the result, not from a conclusion. "I've noticed some features of this work that I'd like to ask you about" is very different from "I know you used AI for this." Show the student the specific signals the tool identified. Ask them to talk through their process - how they planned the piece, what sources they used, how they drafted it. A student who genuinely wrote the work will usually be able to provide a plausible and detailed account. A student who copied an AI output will often struggle to explain specific choices in the text.

Take notes during the conversation, and consider having a second adult present if you anticipate a formal outcome. The detection result and the conversation together constitute the evidence base for any decision - not the score alone.

Try GradeOrbit's AI Detection for Schools

GradeOrbit's detection is designed for exactly this context - checks against each pupil's own earlier work, explained results, confidence-rated scores, support for handwritten work, and pupils' work held only under your school's Data Processing Agreement. It works alongside GradeOrbit's AI marking workflow, so you can screen and mark from a single platform.

Start a free school trial and run your first AI detection check.