How Teachers Flag AI in End-of-Year Extended Writing
End-of-year extended writing tasks sit in a uniquely vulnerable spot in the calendar. Students are exhausted, exam revision has already started for the year above, deadlines stack up across subjects, and the temptation to outsource a 1,500-word essay to a generative tool peaks in late May and June. For the teacher running an AI detection pass across the cohort, the question is not whether some pieces will return high likelihood scores — they will — but how to handle those flags fairly when the pressure on everyone is at its highest.
This guide is for the Year 9, 10, or 12 teacher reading through a stack of end-of-year extended writing pieces, running them through a detection tool, and trying to decide what to do with the results. It covers why this particular deadline produces so many flags, how to read a likelihood score in context, the fair process for a flagged piece, and where the line sits between a conversation and a formal academic integrity case.
Why End-of-Year Writing Attracts AI Use More Than Other Deadlines
Three things converge in late May. Students have exam fatigue from the year above's GCSEs and A-Levels dominating the school timetable. Coursework deadlines compress as departments try to bank grades before reports go home. And the perceived stakes of an end-of-year extended writing task feel lower than a controlled assessment, but higher than a piece of regular homework — which is exactly the range where students are most likely to take a shortcut.
A piece submitted at 11pm on a Sunday in the final fortnight of term, by a student who has otherwise been a competent middle-of-the-pack writer, is the modal flag. It does not mean cheating. It means the piece warrants the same careful read you would give to any submission that surprised you.
Reading a Likelihood Score in Context
A likelihood score is a probability, not a verdict. GradeOrbit's detection tool returns a 0-100% score reflecting how closely the writing's linguistic patterns resemble patterns commonly produced by large language models. A score of 78% means the piece is statistically interesting. It does not mean the student definitely used AI.
Context that should pull your judgment downward includes: the student is a heavy reader who naturally writes in a polished register, the student has EAL background and leans on careful phrasing they have practised, the piece was written in class under your supervision, or the topic is one where the student has demonstrably deep prior knowledge from outside school. Context that should pull your judgment upward includes: the sentence structures look nothing like the student's previous work, ideas appear that you have not taught and would not expect the student to reference, the submission timestamp is consistent with last-minute generation, and the student is unable to discuss specifics of what they wrote when you ask.
For more on how detection scores behave with EAL students specifically — a common false-positive source — see our post on how teachers read AI likelihood scores on EAL student work.
The Fair Process for a Flagged Piece
Do not confront the student in the lesson the flag returns. Take an evening to gather context — prior writing samples, draft history, submission timing — and write a short timeline before speaking to anyone. Hold the conversation privately, frame it as a discussion of how the piece was put together rather than as an accusation, and listen carefully to what the student tells you.
Specific opening questions that tend to work: "Can you walk me through how you structured the opening paragraph?", "Which point did you find hardest to argue?", and "If I asked you to write another paragraph on this topic right now by hand, what would you say?" Students who genuinely wrote the piece engage with the questions. Students who did not often struggle to recall specifics, contradict themselves, or deflect.
Our guide to talking to students about high AI detection scores covers the conversation in more depth, including the handwritten control paragraph that often settles the question.
Where Conversation Ends and Formal Process Begins
Most flagged end-of-year pieces resolve at the conversation stage. Either the evidence supports the student's authorship and you close the case, or the student acknowledges the AI use and you handle it under your school's academic integrity policy without escalation. The minority of cases that need to go further are the ones where the evidence is strong, the student denies, and the consequence — a redacted mark, a re-sit, a formal note on file — needs a documented process behind it.
What that process looks like depends entirely on your school. If your school has an AI academic integrity policy, follow it. If it does not, our post on writing a school AI academic integrity policy covers the ground a workable policy needs to address before the next set of cases lands.
How GradeOrbit's Detection Tool Fits the Process
GradeOrbit's AI detection tool returns a 0-100% likelihood score with two model options. The single-credit fast check is the right choice for first-pass screening across a class set. The three-credit deep analysis runs a more thorough statistical check and is the model to use once a fast check has flagged a piece and you want stronger confidence before opening a conversation. Both models report a probability — neither claims certainty.
Student work is never stored. The page is processed in memory and discarded the moment the result is returned. Names and identifying information are redacted with a black box on the page before upload, burnt into the image on your device, so the AI model only ever sees the anonymised version. That redaction is part of the workflow, not a setting you can forget to enable.
Try GradeOrbit on This Term's Extended Writing
If you have a set of end-of-year extended writing pieces sitting on your desk, the fastest way to see how the likelihood score behaves on real student work is to run a sample through GradeOrbit. Head to the GradeOrbit homepage and create an account. The first batch of credits is free, so you can calibrate the tool against pieces you have already read carefully and compare the scores against what you already know about your students. That calibration is the best preparation for the moment a real flag comes back on a piece you have not read yet.