Skip to main content
Back to Blog

How Teachers Use AI Detection on Controlled Assessment

Breanna Mitchell·Content Writer
7 min read

Controlled assessment is supposed to be the safe ground. The work is produced under supervision, in your classroom, within a fixed window — so the temptation is to assume AI use simply cannot happen. In practice the picture is messier. Research and planning often happen at home. Earlier drafts may be typed up elsewhere and brought in. Some specifications allow notes or preparatory materials into the supervised session. By the time a polished final piece lands on your desk, "controlled" rarely means "every word written in front of me". That is exactly why AI detection for controlled assessment still has a place in a fair, defensible marking process.

This guide is written for the teacher who has just run a likelihood score on a controlled piece and is now asking the only question that matters: what do I do with this number? It is not a guide to what AI writing looks like, and it is deliberately not a checklist of tells — that kind of content reads, in reverse, as an evasion manual. The focus here is process: how to act fairly once a score comes back, especially on work that was meant to be supervised.

Why detection still matters on supervised work

The phrase "controlled assessment" covers a spectrum. At one end, every stage is supervised and notes are forbidden. At the other, students prepare extensively at home and the supervised session is essentially a write-up of pre-existing material. Most coursework sits somewhere in between, and the in-between is where AI use creeps in. A student who drafted an entire response with an AI tool the night before, memorised the shape of it, and reproduced it under supervision has produced something that looks handwritten and original but is not their own thinking.

Detection does not "catch" anyone on its own. What it does is flag pieces where the language patterns are consistent with AI generation, so you know where to look more closely. On controlled work, a high score is a prompt to ask how the preparation stage was conducted — not an automatic accusation about the supervised session itself.

What a likelihood score actually means here

A likelihood score is probabilistic, not a verdict. GradeOrbit returns a percentage from 0 to 100 representing how consistent the writing is with AI-generated text. It is a signal to weigh alongside everything else you know — not a switch that proves guilt. This matters even more on controlled assessment, because you have additional evidence the score does not see: you watched the session, you know the student's classwork voice, and you may have their earlier drafts.

Treat the score as one input into your professional judgment. A 90% on a piece that bears no resemblance to the student's normal writing is worth a calm, careful conversation. A 90% on a piece that matches their classwork voice exactly, and which you watched them write, tells you the model is reacting to style — confident, well-structured prose can read as "AI-like" even when it is entirely a strong student's own work. The number narrows where you look; it never decides for you.

Handling a high score fairly

The fair response to a high likelihood score on controlled work is the same disciplined sequence you would use anywhere, with one advantage: you have more context. Work through it in order.

  • Gather your own evidence first. Compare the flagged piece against the student's classwork, earlier drafts, and any preparation notes. On controlled assessment you often have a planning stage on record — use it.
  • Have the conversation, not the accusation. Ask the student to talk you through their planning and choices. A student who genuinely produced the work can explain their reasoning; the conversation is informative whichever way it goes.
  • Align with your school's policy. A likelihood score is evidence to be weighed, not a finding in itself. Your academic integrity policy should set the threshold for escalation and the steps that follow.
  • Document the decision. Record the score, your supporting evidence, the conversation, and the outcome. On controlled work this paper trail protects both the student and your moderation process.

If you want the full sequence laid out, our guide to how teachers handle a high AI likelihood score fairly walks through each stage, and how to interpret AI detection likelihood scores covers the mid-range cases where the number sits stubbornly in the middle.

GradeOrbit's built-in detection tool

GradeOrbit's AI detection runs on the same upload-and-scan flow as the marking workflow, so checking controlled assessment fits the routine you already use. You upload the student's work — typed or handwritten — and the tool returns a likelihood score from 0 to 100%. There are two models to choose from: a 1-credit option for a quick first pass across a class set, and a 3-credit option for a more thorough analysis on the pieces you want to scrutinise. That split lets you screen broadly and then look closely only where it counts, rather than burning the same cost on every script.

Crucially, GradeOrbit treats the score as professional evidence for you, not a judgment delivered to the student. The output supports your decision; it never replaces it. For solo and team teachers, uploaded work is never stored — it is processed and discarded — which keeps the privacy contract intact while you check controlled pieces.

The bottom line

Controlled assessment is not immune to AI use, because the preparation around it rarely happens under the same supervision as the final write-up. Detection gives you a defensible way to flag pieces worth a second look, and a likelihood score gives you a starting point — not a conclusion. Used inside a fair process of evidence, conversation, policy and documentation, AI detection makes your moderation stronger without ever taking the professional judgment out of your hands.

Check Controlled Assessment with GradeOrbit

GradeOrbit gives UK teachers a built-in AI detection tool that returns a clear 0–100% likelihood score on typed or handwritten work, with a quick 1-credit pass and a thorough 3-credit analysis. It supports your professional judgment — it never replaces it. Try GradeOrbit today and bring fair, evidence-led AI detection into your controlled assessment marking.

More on this topic

30 June 20267 min read

How Teachers Detect AI in GCSE Film Studies Coursework

GCSE Film Studies coursework asks for analytical writing about real films — exactly the kind of task AI handles well. A guide for teachers on reading likelihood scores, fair process, and aligning with school policy.

Read more
8 June 20267 min read

How Teachers Handle a Parent Disputing an AI Score

A parent says the AI score is wrong. A UK teacher guide to the fair, evidence-led conversation — likelihood is not proof, policy comes first, and the relationship stays intact.

Read more

Ready to save time on marking?

Join UK teachers using AI to provide better feedback in less time.

Get Started Free