How Teachers Use AI Detection on Year 13 Coursework Submissions

Year 13 coursework lands in your tray at the worst possible moment. UCAS predictions are submitted, exam leave is days away, and a final piece of NEA writing turns up that reads two grade boundaries above anything the student produced in lesson. AI detection for Year 13 coursework is the most uncomfortable use of the tool, because the consequences of getting it wrong cut both ways - a wrongly accused student loses a university place, and a wrongly trusted submission undermines the integrity of the whole cohort's results.

This post is for the teacher staring at that final draft, deciding whether to run a likelihood check and what to do with the number that comes back. It walks through what Year 13 polish actually looks like at the wire, how to read a score on a final submission, the conversation to run before you mark, and what to do when the student is already on study leave and you cannot get them in the room.

What Year 13 Coursework Actually Looks Like at the Wire

By the time a Year 13 hands in a final NEA, they have spent months on it. They have been redrafting since October. They have had detailed teacher feedback at multiple checkpoints. They have probably shown it to a parent, a tutor, or a sibling who studied the same subject at university. The polish you are seeing is not, by itself, suspicious. A student who has worked seriously on a 4,000-word History NEA for eight months should sound more sophisticated than they did in their first 600-word essay back in September.

What you are looking for is a step-change that does not match the trajectory. If every checkpoint draft built incrementally on the last one, and then the final draft reads like a different person wrote it, that is the signal worth investigating. If you have early-stage drafts that already showed strong control of source material and argument, a polished final draft is exactly what the process should produce.

Trust the trajectory of the work, not the polish of the final paragraph. Run detection when there is a genuine discontinuity, not when a student finally sounds like they have learned what you spent a year teaching them.

Reading a Likelihood Score on a Final Submission

GradeOrbit's AI detection, part of its school accounts, returns a score between 0 and 100 with a Low, Medium or High confidence label, and there's no separate deep-analysis tier to weigh up for Year 13 coursework. What matters more for a final NEA submission is what the score is actually based on: GradeOrbit compares the writing against that student's own earlier work already on file and measures how far the new piece departs from it, which is a stronger basis for a decision this consequential than a generic likelihood score would be.

A high score on Year 13 coursework still is not proof. On generic detectors, NEA writing tends to push scores upward because the form itself rewards the patterns AI is good at producing - careful transitions, balanced argument, varied sentence rhythm, sophisticated vocabulary. Even against a student's own earlier work, eight months of redrafting and coaching can produce a genuine leap that looks like a departure, on writing that is entirely the student's own. Treat the score as one piece of statistical evidence alongside the drafting history, the in-lesson observations, and what you know about the student's voice.

A low score is also not a clean bill of health. A student who used AI early in the process to generate a structure and then rewrote the prose in their own voice can return a low score on a piece that started with a machine. Detection scores reflect the submitted text, not the process behind it.

The Conversation Before You Mark

If you have time before the marking deadline, get the student in the room. Frame it as a standard checkpoint, not an accusation. "I wanted to talk through your final draft before I mark it - can you walk me through how you put the closing argument together?" Most students who have written their own work will engage in detail. They will tell you which source pushed them toward a particular conclusion, which paragraph they rewrote three times, which sentence they are most proud of.

Specific prompts that work well on coursework: "Which secondary source changed your mind during drafting?", "What was the hardest part of the argument to land?", and "If I asked you to add a 200-word footnote on this point right now, by hand, what would it say?" A student who has genuinely written the piece can answer these. A student who has not will tend to deflect, generalise, or contradict things they appeared to argue in the text.

Document what you discussed and what the student said. Coursework decisions get audited - by your moderator, by the exam board, occasionally by an appeal panel - and the contemporaneous note you wrote on the day is what those processes rely on. For the wider framework, see our post on using AI detection as professional evidence.

When You Can't Get the Student in the Room Before the Deadline

Sometimes the work lands when the student is already on study leave or has finished school for the year. You cannot run the conversation, and the moderation deadline is non-negotiable. This is the situation where the trajectory evidence matters most.

Go back through the drafting record. Compare the final piece to the checkpoint drafts you have on file. If the final builds visibly on what came before - same argument, same sources, same voice, just tightened - submit it with confidence. If there is a discontinuity, escalate to your head of department before the moderation submission goes in. Most exam boards have a procedure for flagging suspected malpractice on a piece you cannot conscience-mark; the procedure exists precisely so you do not have to make the call alone.

Whatever you decide, write down what you weighed and why. A short paragraph on the moderation cover sheet is enough - what the AI likelihood score was, what evidence you cross-referenced, what you concluded. The exam board does not expect you to be a forensic analyst; it expects you to have used your professional judgment and to have documented it. For more on the conversation side of this, see how to talk to a student about a high AI detection score.

What Detection Cannot Tell You

A likelihood score will not tell you whether the AI use was permitted. Many Year 13 specifications allow students to use AI for structure suggestions, for source summaries, or for proofreading - provided the final argument and prose are the student's own. A 40% score on a piece where the student used AI as permitted is not malpractice; it is the specification working as designed.

Before you run a check, make sure you know what your specification permits and what your school's AI policy says. If you do not have a written school policy yet, our post on writing a school AI academic integrity policy covers what one needs to address.

Try GradeOrbit's AI Detection on Real Coursework

GradeOrbit's AI detection tool was built for exactly this kind of high-stakes teacher decision. The 0-100 score is an indicator to review, never proof, and it works the same whether it's a routine screen or a piece of NEA you need stronger evidence on. It works by comparing the submission against the student's own earlier writing already on file; a student with no earlier work on file is skipped rather than given a generic score. Detection is included with unlimited checks in every school teacher seat, and pupils' work is held under the school's signed Data Processing Agreement for the school's chosen retention window, wiped on schedule, and never used to train AI models.

Start a free school trial and run detection against coursework you have already moderated, then compare the scores to what you already know about your cohort. That comparison - your professional judgment against the tool's number - is the right way to learn how much weight to give the score on the next piece of work.