How to Use AI Detection in GCSE Coursework Marking

With ChatGPT and Claude freely available to students, a teacher marking GCSE extended writing, non-exam assessment, or controlled assessment now has a second question alongside "is this good work?": "is this the student's own work?" AI detection tools can help answer it, but only when you are clear about what they can and cannot tell you.

The aim below is a detection routine that is fair to students, defensible to parents and colleagues, and sustainable across a full class set.

What a Likelihood Score Actually Tells You

GradeOrbit's AI detection tool returns a likelihood score between 0% and 100%. A low score suggests the text is strongly consistent with human-written work. A high score suggests it shares significant statistical characteristics with AI-generated output. What the score does not tell you - and what no AI detection tool can tell you - is whether a specific student opened ChatGPT, typed a prompt, and submitted the result.

Detection works by analysing linguistic patterns: sentence structure, vocabulary distribution, syntactic consistency, and the particular regularities that tend to distinguish AI-generated prose from human writing. It is probabilistic reasoning applied to text, not forensic evidence. A likelihood score is best understood as a starting point for professional judgment, not a conclusion that replaces it.

AI detection is included with unlimited checks in GradeOrbit's School teacher seats, so there is no separate standard-versus-in-depth tier to choose between and no per-check credit charge. Every run returns the likelihood score alongside a breakdown of the linguistic signals that contributed to it and a reasoning paragraph. On School accounts operating under a signed Data Processing Agreement (DPA), submitted coursework can be compared against earlier writing samples on file within your school's agreed retention window, providing a grounded comparison rather than an isolated score. For borderline cases - scores in the 40–70% range where the evidence is genuinely mixed - that baseline comparison is more useful than a bigger model would be on its own.

Which Subjects and Assessment Types Benefit Most

AI detection is most useful for assessments that involve extended, unsupervised writing. In the GCSE context, that means coursework components, non-exam assessments (NEAs), and any extended homework tasks that contribute to a final grade or inform teacher assessment.

English Literature and Language coursework are the most obvious candidates: AI tools are well-suited to generating the kind of analytical prose these assessments require, and the consequences of AI-generated content in a graded component are significant. History NEAs, Geography fieldwork write-ups, Religious Studies extended essays, and Psychology coursework all share the same risk profile - long-form analytical writing completed outside supervised conditions.

Detection is less meaningful for short answers, bullet-pointed responses, or structured tasks with very limited scope for AI involvement. Running detection on a ten-word definition is technically possible but tells you very little. Focus your detection workflow on the submissions where AI assistance would actually give a student a meaningful advantage.

How to Use Detection Without Causing False Alarms

The most common misuse of AI detection tools is treating a high score as confirmation of wrongdoing. This risks causing serious distress to students who did nothing wrong - and it exposes teachers to challenge from parents and senior colleagues who rightly ask what evidence supports the accusation.

A reliable process combines the detection score with everything else you know about the student. Before acting on any score above 60%, ask yourself: does this piece read like the student? Is the quality, register, and sophistication consistent with their previous work? Did they submit it in a way that fits their usual pattern? If the answers are broadly yes, a detection score alone is not sufficient grounds for escalation.

Conversely, if a student who has consistently produced brief, grammatically insecure responses submits something that reads like a polished Claude output, the detection score is one part of a picture that already raises questions. The score confirms what the reading suggested - it does not create the concern on its own.

For guidance on what to do when a score lands in genuinely uncertain territory, our post on how to handle ambiguous AI detection results covers the process step by step, including how to approach a conversation with the student.

Getting Coursework Into the Check

AI detection runs on school accounts, where coursework is stored against the pupil it belongs to under the school's Data Processing Agreement. Upload a set of work to a class and GradeOrbit reads each cover page and matches the name to a pupil; confident matches link automatically, and the rest wait for you to pick the right pupil from a list. You can also upload directly to a single pupil.

Linking work to the right pupil is what makes the result meaningful, because each piece is compared with that pupil's own earlier writing. It also means the record is there later if a moderator, exams officer or parent asks how a concern was reached.

Privacy protections depend on your account tier. On individual teacher accounts used for marking, student work is redacted in the browser and never stored on any server. For school accounts using AI detection under a signed Data Processing Agreement (DPA), submitted work is retained securely for your school's agreed retention window (typically 3 years) so moderation records remain accessible. Crucially, across all tiers, student submissions are never used to train AI models.

Making Detection Part of Your Marking Routine

Detection is easier to sustain when it is part of marking rather than a separate task. For a GCSE class set, one approach is to run a check on every submission as you mark it, then give closer attention to anything above a threshold - typically 65–70% - by reading the baseline comparison and the linguistic signals before deciding whether it needs a conversation.

This means you are not making decisions on the basis of a single number. You are pairing a routine screen with a closer read of the result - the comparison against the pupil's own earlier writing, the specific signals, and everything else you already know about them. With unlimited checks included in School teacher seats, departments can screen entire cohorts without worrying about per-check credit deductions.

Where a school or department has introduced a consistent AI detection policy, this kind of structured approach is much easier to sustain. For departments looking to standardise their process, our guide on how schools can implement AI detection consistently covers the policy and workflow considerations in detail.

The Point of Detection

Detection in GCSE coursework is not about catching students out. It is about keeping assessment honest now that AI help is easy to get. A detection result is one piece of information for a fair, evidence-based process; the judgment remains the teacher's.

In GradeOrbit, detection runs from the same dashboard as marking and accepts pasted text, uploaded documents, and scanned images of handwritten work. Unlimited checks are included in School teacher seats (£10 per teacher per month, billed annually), and the school plan has a 30-day free trial if your department wants to test it against its current coursework process.