Detecting AI in End-of-Year Coursework Submissions

By the time end-of-year coursework lands on a teacher's desk in May or June, the pressure is at its highest point of the academic year. Final drafts need to be marked, internal moderation deadlines are tight, and any concerns about authenticity have to be resolved before the work is signed off. Detecting AI in end-of-year coursework submissions is now a routine part of that final review - and the way it is done matters as much as whether it is done at all.

This guide is written for UK secondary school teachers handling final coursework drafts in the summer term. It covers how a likelihood score actually works on coursework-length writing, what the score does and does not tell you, how to use GradeOrbit's detection tool to triage a stack of submissions, and how to bring professional judgment back into the centre of the decision.

Why End-of-Year Submissions Need a Different Approach

A coursework draft submitted in October can be revised, redrafted, and re-checked. A coursework submission landing in late May is final. There is no second draft, no follow-up homework, and very little time before the work goes to internal moderation. That changes how a teacher weighs a high detection score. The stakes are higher, and the cost of getting it wrong - in either direction - is significant.

Students also know this. The temptation to lean on ChatGPT or Claude is at its peak when other final deadlines are stacking up: A-Level exams, GCSE papers, EPQ submissions, and university applications all collide in the same window. A Year 11 student finishing their last piece of coursework on a Sunday night, with three other deadlines that week, is the most likely profile for late-stage AI use we see in the product.

What a Likelihood Score Actually Tells You

GradeOrbit's detection, part of its school accounts, returns a score from 0 to 100. The number measures how far the submission departs from that pupil's own earlier writing on file - not a verdict, not a confession, and not evidence on its own. A low score means the work is consistent with how the pupil usually writes. A high score means it differs from their writing in ways worth a closer look. The middle band is where teacher judgment matters most.

The score is not a court verdict. It is a triage signal. A high score tells you that a piece of work deserves a closer look - at the redrafts the student submitted along the way, at the in-class writing they have done, at the kinds of mistakes a human student in that subject usually makes and an AI usually does not. None of that lives inside the detection tool. It lives in your knowledge of the student.

Using GradeOrbit's Detection Tool on a Stack of Submissions

On a school account, a teacher can run detection on a whole class at once. Scan the stack, or export typed drafts to PDF, and bulk upload it with the number of pages per student; GradeOrbit reads the name on each cover page and matches it to the class list, and you pick the pupil if the match isn't clear. Detection is included in every school teacher seat, with no per-check charge, whatever the length or stakes of the piece. It works by comparing each submission against that pupil's own earlier writing on file, which is a stronger signal for a final coursework draft than a generic score would be. A pupil with no earlier work on file is skipped, and you are told.

Each result appears on the pupil's profile: a score, a Low, Medium or High confidence label, the signals it noticed, and a plain-English reasoning paragraph. There is nothing more - no flagged sentences highlighted on the page, no fingerprint claiming to identify which model produced the text. The restraint is deliberate. A teacher who sees a sentence-by-sentence highlight starts treating the highlights as evidence, when in fact they are just the model's best guess about which parts contributed most to the score. We have written more about that design choice in our guide to AI detection for teachers.

When the Score Is High - Acting on Professional Judgment

A likelihood score above 80% on an end-of-year submission is a strong signal. It is not a final answer. The next step is always the same: compare the submitted piece against the student's earlier drafts, their classwork from across the year, and your own knowledge of how they write.

Specific things to look for. Does the vocabulary in the final draft jump significantly above the level of their classwork? Are there ideas, citations, or framings that the student has never raised in lessons? Does the structure follow a pattern you have seen from other AI-written work - a tidy three-point introduction, evenly weighted body paragraphs, a clean summary conclusion, with no rough edges?

If those signals line up with the detection score, you have grounds to open a conversation. If they do not, the score on its own is not enough. A confident, high-attaining student who writes cleanly may well score 60 or 70%. That is the limit of what a probabilistic tool can do, and pretending otherwise damages the teacher-student relationship.

Talking to Students Without Accusing Them

The conversation with the student is the part of this process that the detection tool cannot help with. It has to be done in person, and it has to be opened in a way that protects the student's dignity if the score turns out to be a false positive.

The framing that works is curiosity, not accusation. "I have read your final draft and there are a few things I want to talk through with you - can you walk me through how you wrote the introduction?" gets you further than "the AI detector says this is AI". A student who has written the piece themselves can usually talk you through their thinking - the books they read, the points they wanted to make, the bits they found hard. A student who has not will struggle to reconstruct that process. That conversation is the actual evidence, not the score.

The same guide also covers talking to students about high AI detection scores, including what to do when a parent gets involved.

Documenting the Decision

For end-of-year coursework, the detection step needs to be documented even when the score is low. Internal moderators and exam boards increasingly ask schools to show their authenticity-checking process - not the result on any one student, but the process that was applied consistently across the cohort. A short note in the student's record showing that the detection check was run, what score came back, and what next step was taken (no action, follow-up conversation, redraft requested) is enough to satisfy most moderation processes.

AI detection runs on school accounts, where pupils' work is held under the school's Data Processing Agreement and compared with their own earlier writing, then wiped at the end of the school's retention window or at offboarding. Your note of the decision lives in your own records, on your school's systems. That arrangement is part of the data protection picture we cover in our guide to whether AI detection tools are safe to use in schools.

Try GradeOrbit's AI Detection on Your End-of-Year Submissions

If you are working through a stack of end-of-year coursework drafts and want to add a structured detection step to your authenticity check, GradeOrbit's school detection is built for exactly this kind of triage. It is included in every school teacher seat and weighs each submission against that pupil's own earlier writing on file, so you get a considered result on every draft without a per-check charge.

See how detection works on a school account and start a free school trial.