How Teachers Detect AI in A-Level Reflective Commentaries

Reflective commentaries are quietly one of the hardest places to detect AI in A-Level coursework. The evaluative writing that sits alongside an art portfolio, a design and technology NEA, a photography project or an extended piece of independent study is meant to be the student's own voice describing their own decisions — which is exactly the kind of fluent, first-person prose that generative tools produce convincingly. Learning to detect AI in reflective commentaries fairly is now part of the job for any teacher marking this component.

This post is for the teacher opening a folder of evaluations the week before an internal deadline, reading a paragraph that sounds a little too polished, and wondering what to do next. The honest answer is that you cannot prove authorship from the text alone — but you can run a fair, repeatable process that turns a suspicion into a defensible professional judgment rather than a guess.

Why Reflective Commentaries Are an AI Blind Spot

Most coursework detection advice focuses on the analytical or factual parts of a submission — the essay, the source analysis, the technical write-up. Reflective commentaries slip through because they are supposed to read differently. They are personal, evaluative and process-focused: "I chose this composition because...", "On reflection, my second prototype failed because...". When the genuine article is meant to sound natural and a little informal, a smooth AI-generated paragraph does not stand out the way a perfectly structured exam answer might.

There is a second reason the commentary is vulnerable. By the time a student writes it, the hard creative or practical work is finished and the deadline is close. A tired Year 13 student who has built the artefact but left the 800-word evaluation to the last night is precisely the person most tempted to paste their bullet points into a chatbot and ask it to "write this up properly". The work behind the commentary may be entirely their own, while the words describing it are not — which makes a calm, evidence-led check more important, not less.

Reading a Likelihood Score on a Reflective Commentary

The first habit to build is the right mental model for what an AI detection tool actually gives you. A likelihood score is a signal, not a verdict. GradeOrbit's detection tool returns a probability between 0% and 100% that the text under review was AI-generated, alongside a short, plain-English explanation. It does not return a finding, and it certainly does not return a confession. It tells you where to look more closely.

Treat the number like a temperature reading rather than a result. A commentary at 9% likelihood does not need a second glance unless something else has already caught your eye. A commentary at 88% likelihood does not warrant an accusation — it warrants a closer read, a comparison against the student's known work, and, where your school's process requires it, a second marker. GradeOrbit lets you choose between a faster 1-credit model for a routine first pass and a more thorough 3-credit model for the cases where the cost of a wrong call is higher. For a close-call evaluation that could affect a coursework grade, the more thorough check is the one to reach for.

Because reflective commentaries are often short, treat very brief samples with extra caution. The less text a tool has to work with, the noisier any probabilistic estimate becomes — a single paragraph is a far weaker basis for a judgment than a full 800-word evaluation. If our wider guidance on reading these numbers would help, see how to interpret AI detection likelihood scores.

How to Handle a High Score Fairly

When a commentary returns a high likelihood score, the single most useful move is to put it next to writing you already know is the student's. Reflective commentaries are unusually good for this, because you almost always have a paper trail: lesson-time annotations on their sketchbook, handwritten development notes, earlier drafts of the same evaluation, or a mid-project review they wrote in class. Compare the rhythm, the vocabulary range and the kinds of mistakes the student tends to make. Genuine student writing has a fingerprint — recognisable stock phrases, a particular way of explaining a decision — and generative AI rarely reproduces it.

The mismatch to look for is content, not just style. A real reflective commentary references specific, idiosyncratic details of the student's own project: the exact material that warped, the lesson where their idea changed, the peer feedback they acted on. AI-generated reflection tends to be confidently generic — plausible-sounding evaluation that could attach to almost any project of that type. If the writing is fluent but strangely free of the messy specifics you watched the student wrestle with, that gap is more telling than any single sentence.

If, after that comparison, you are still unsure, the right move is to log the case as ambiguous and follow your school's process rather than forcing a conclusion. A detection score should never be the sole basis for a decision that affects a grade or a student's record. Our walkthrough on how teachers handle a high AI likelihood score fairly sets out that conversation step by step.

Keeping the Process Fair and Documented

Coursework carries a moderation and appeal trail, so anything you do around AI detection needs to be the kind of thing you would be comfortable explaining to a Head of Department, a parent, or an exam board. That means consistency: the same threshold for when a commentary gets a closer look, the same next step when one is flagged, and a short written note of the score, your professional read, and the decision made. The goal is not a file of accusations — it is a record that shows you applied judgment fairly and the same way for every student.

It also means protecting the student's data while you do it. With GradeOrbit's solo tool, you redact any names or personal details with on-screen black boxes before anything is processed, the work is never stored, and the file is gone the moment the check is done. That keeps the data side of your academic integrity policy clean when someone asks where flagged scripts are held. For the bigger picture of using a score as part of a professional record, see how to use AI detection as professional evidence.

Add a Calmer AI Check to Your Coursework Marking

Reflective commentaries deserve the same calm, evidence-led approach you already bring to blind marking and moderation. A likelihood score is a prompt to look again, a comparison against known work is the safety net, and a documented decision is the deliverable. Detection supports your professional judgment — it never replaces it, and it never decides on its own.

Try GradeOrbit free today and add a fair, private AI detection step to your next round of A-Level coursework marking. Start at gradeorbit.co.uk — your first checks are on us, and there is no card required to try the tool on real student work.