How to detect AI-generated work in BTEC assignments
The question has changed. It is no longer "is this AI-written?" — it is "can I show this is the learner's own work?" JCQ guidance is explicit that teachers must pay close attention to authentication when marking coursework, and BTEC assessors are the front line. Here is a workable process that treats AI detection as evidence gathering, not guesswork.
What the rules actually say
Two documents govern you: JCQ's AI use in assessments guidance and Pearson's own AI centre guidance for BTEC centres. The shared core: learners' work must be their own; where AI use is permitted at all, centres decide and communicate what is allowed and what must be acknowledged; assessors are responsible for authenticating evidence before it counts. Where AI use is not allowed and not acknowledged, that is malpractice — handled through your centre's malpractice policy, not by quietly marking work down.
The signs, in BTEC coursework specifically
Generic detectors produce a number that is hard to defend. What travels better is the BTEC-specific picture:
- Scenario blindness. AI-written evidence tends to float above the brief — the business, the data tables, the specific context get summarised but never used. Since Merit and Distinction criteria demand scenario-anchored analysis, AI filler often coincides with criterion gaps.
- Vocabulary mismatch. A learner whose classroom talk is basic producing polished board-report prose — the writing quality jumps well past anything in their drafts.
- Perfect structure, empty content. Immaculate headings mirroring the task list, but each section describes the topic generally — no figures from the brief, no calculations, nothing only this learner could know.
- Interrogation fails. Ask the learner to explain a sentence, justify a figure, or continue the argument. Inability to discuss their own "work" is the strongest single signal.
- Reference unreality. Cited businesses or data that don't exist, numbers that don't reconcile with the brief's tables.
A defensible process
- Set expectations up front. Tell learners in writing what AI use is permitted (if any) and what must be acknowledged. Most "cases" collapse at this step — no rule was ever stated.
- Collect process evidence as routine. Drafts, version history, in-class writing time, observation records — the positive authentication trail. The more routine, the less adversarial.
- Run a criteria-based check on the submission. Not a "detector score" — a read against the brief: does the evidence use the scenario's data, satisfy the command verbs, and hang together as one voice? CheckB's free authenticity auditor is built exactly for this pass — vocational depth, case reality and AI-generation indicators, plus generated viva-style questions for the conversation.
- Have the conversation before the allegation. Sit down with the learner and the work. Ask them to walk through it. Most situations resolve here honestly.
- Escalate on evidence, through policy. If it stands, follow your centre's malpractice procedure with the documentation you've gathered. That file — expectations, process evidence, check results, conversation notes — protects everyone.
What not to do
- Don't lead with a detector percentage. No detector score is proof, and an over-claimed number is easy for a parent to demolish. Use BTEC-specific indicators and the learner's own inability to discuss the work.
- Don't grade punitively. Marking work down "because AI" without process is not assessment — the criteria still decide grades. If the work isn't the learner's, that is the malpractice route; if it is, it gets assessed normally.
- Don't assume good prose means AI. Some learners write beautifully; some finally engaged with a topic. Evidence, not surprise.
FAQ
Should I run every submission through AI detection?
Screen every submission for authenticity — yes. That means the criteria and scenario checks above, not a magic number. A tool pass plus your knowledge of the learner is faster and more defensible than any single detector.
What if the learner used AI to plan, but wrote the evidence?
If your centre's stated rules permit that use and it was acknowledged, the evidence can still be the learner's own — assess it normally. If it wasn't permitted, that is the conversation.