AI Detection for Criminology Coursework: A Teacher's Guide

Criminology is now a common option in UK sixth forms and colleges, and its assessment structure leaves it unusually exposed to AI-generated writing. The WJEC Level 3 Diploma is built around controlled assessments that ask learners to analyse crime data, evaluate campaigns, and write extended reasoned arguments about the criminal justice system. Those are tasks that ChatGPT and Claude handle fluently, so anyone delivering Criminology needs a fair way to assess work when a learner may have leaned on them.

Doing that well means understanding what a probabilistic likelihood score measures, having a plan for when a controlled assessment scores high, and weighing detection results against your professional judgment as the assessor.

Why Criminology Coursework Is Especially Exposed

Unit 2 and Unit 4 of the WJEC Diploma are internally assessed under controlled conditions, but the preparation and drafting that surrounds them often happens away from your classroom. Learners research campaigns, gather evidence, and plan responses before the formal write-up. That preparation window is where AI tools get used - a learner might ask a chatbot to summarise the arguments for and against a particular sentencing policy, then reproduce that structure in their assessment.

The writing that results tends to be fluent, well-organised, and confident. It hits the assessment criteria for analysis and evaluation on the surface. But when you ask the learner to explain how they reached a conclusion, or to defend a claim about recidivism rates, the understanding is not there. That gap between polished writing and shallow comprehension is the classic signature of over-reliance on AI.

The Evaluation Problem

Criminology rewards evaluation - weighing theories, judging the effectiveness of policies, reaching justified conclusions. Large language models are good at producing balanced-sounding evaluation because they have absorbed vast quantities of argumentative text. The result reads like genuine critical thinking but is often a smooth restatement of common positions with no original reasoning. Detecting that difference by eye alone, across a full cohort, is slow and inconsistent.

How Likelihood Scores Actually Work

Modern AI detection does not give you a yes/no verdict. It gives you a likelihood score: an indicator to review, never proof. GradeOrbit's detection, on school accounts, returns a score from 0 to 100 that measures how far the new work departs from that learner's own earlier writing on file: a low score means it is consistent with their own writing, a high score means it differs from it in ways worth a closer look.

That probabilistic framing matters enormously. A score of 84 does not prove a learner cheated. It tells you the writing - sentence rhythm, vocabulary, the predictability of word choices - departs sharply from what this learner has produced before, in ways consistent with AI assistance. You then bring your knowledge of the learner, their previous work, and the assessment context to interpret what the score means. The number narrows your attention; it does not make the decision for you.

Baseline-Aware Detection on School Accounts

On a GradeOrbit school account, unlimited AI detection is included in every teacher seat, with no per-check charge. Instead of a generic "does this read like AI" score, GradeOrbit compares the submitted work against that learner's own earlier work already marked in GradeOrbit, which is a more useful signal for extended evaluative writing than a one-off score would be. Each result comes with a Low, Medium or High confidence label, the signals it noticed and a plain-English reasoning paragraph, and a learner with no earlier work on file is skipped, with you told, rather than given a generic score. Individual teacher accounts do not include AI detection.

For Criminology, a sensible pattern is still to run detection across every learner's controlled assessment at the deadline, and then give the small number of pieces that flag highest the closer look - checking against the learner's track record, having the conversation, and applying your professional judgment. The same approach works well for BTEC coursework, which shares Criminology's extended-writing structure.

What to Do When Criminology Work Scores High

A high likelihood score is the start of a conversation, not the end of one. Once a learner's controlled assessment scores above your centre's threshold, do not fill in a malpractice form before doing three things.

First, look at the learner's track record. Does this writing match the style and standard of what they produce in class discussion and timed tasks? Criminology teachers usually have months of evidence to compare against - seminar contributions, earlier drafts, exam responses. A high detection score on work that is wildly out of line with everything else a learner has produced is far more concerning than a high score on work that matches their usual standard.

Second, talk to the learner. Ask them to explain a specific paragraph in their own words. Ask why they chose a particular campaign, theory, or piece of evidence. Genuine writers can always account for their choices. Learners who have leaned heavily on AI usually cannot - and that conversation, recorded in your assessor notes, is often stronger evidence than the score itself.

Third, apply your professional judgment. You are the assessor. The detection score is one data point. Your knowledge of the learner, the assignment brief, and the standards of the qualification matter more than any single number.

Building a Department Policy Around Likelihood Scores

If you lead Criminology in your school or college, agree a shared threshold with your team before the next controlled assessment window - for example, a score of 70 as the trigger for a verifier conversation, or a stricter 85. The exact figure matters less than the consistency - every learner is treated the same way, and every flagged piece goes through the same evidence-gathering process.

Document the policy and tell learners at the start of the unit that AI detection is part of how you check submitted work. That transparency tends to reduce AI use far more than punishing learners after the fact.

How GradeOrbit Handles Student Data

Because AI detection runs on a school account covered by a signed Data Processing Agreement (DPA), learners' work is held under that agreement with their real names, so each check can compare against that learner's earlier work on file. Uploaded scans are removed when a run archives (after 90 days by default, and the school can set anything from 7 to 365 days), learner records follow your school's retention window, and everything is wiped at offboarding. Individual teacher accounts do not include detection; they cover marking, where uploaded work is redacted client-side, never stored, and students are referred to only as "Student 1", "Student 2", and so on. Across every tier, student work is never used to train AI models.

Before the Next Assessment Window

Agree the threshold with your team, tell learners that detection is part of the process, and set aside time for conversations with the few learners whose work flags highest. In Criminology the evaluation sections are usually where a learner's own reasoning, or its absence, shows most clearly, so that is where the conversation should start; the decision on authenticity stays with you as the assessor.

If you want to try baseline-aware detection on a real deadline, the GradeOrbit for Schools page explains the school plan and how to start a free school trial.