The UK Department for Education is testing AI to generate writing samples for assessing pupils’ literacy at the end of key stage 2. The approach aims to reduce annual costs by 95 per cent compared with using real children’s work. Around 2,000 moderators currently verify teacher-assigned grades by comparing them with actual pupil scripts, a process costing about £100,000 yearly. From this year, some moderators will review output produced by the GPT-5 model instead. The trial remains limited: three collections will use AI material for one standardisation exercise, while full exercises in 2026-27 and 2027-28 will retain children’s scripts. Twenty experienced moderation managers will check the AI samples for authenticity. The department will decide in spring 2027 whether to continue or revert to external suppliers. Experts have raised ethical concerns, noting that non-real writing could alter perceptions of acceptable standards and might disadvantage neurodivergent or non-native English speakers. The DfE states that outputs will be heavily reviewed and edited before use.
