Rubric-aligned AI marking

What “rubric-aligned” means

Every exam board rewards different things. An essay that scores full marks under one board’s criteria can underperform under another’s. In Guidelight, you upload the exam-board rubric itself when you create an essay task. The AI then marks against those exact criteria — band by band, criterion by criterion — rather than against a generic notion of “good writing”.

The workflow

• Upload the rubric. Paste or upload the exam-board marking criteria when creating the task.
• Model essay. The AI drafts a model answer aligned to the top band of your rubric, so students can see what excellence looks like.
• Student writes. The task is delivered as a single-question essay with the integrity controls you choose.
• AI marks. Each attempt receives a mark and written feedback tied to specific rubric criteria.
• Teacher reviews. Nothing reaches students until you have reviewed and released the marks. You stay the examiner of record.
• Rewrite loop. Students revise against the feedback and resubmit, turning one essay into a genuine learning cycle.

Why teacher review is built in

AI output is probabilistic; assessment is consequential. Guidelight therefore treats AI marks as drafts: fast, consistent and rubric-referenced, but provisional. Teachers review, adjust and release. Every mark is then stored against the student’s profile, feeding weakspot analysis and exam-readiness probabilities across the term.

Where the AI runs

All marking runs on Cloudflare Workers AI, at the edge, on our own domain. Class data is never sent to OpenAI, ChatGPT or any external AI provider — which keeps the feature usable in every region and straightforward under GDPR and school data policies.