AI marking for teachers: how Rubrical works

Rubrical is an AI marking assistant OpenKit built for the Department for Education, starting with GCSE Geography. It drafts rubric-aligned feedback that the teacher edits and signs, and teachers report a 52% reduction in marking time in the developer’s tests, while the moderation trail stays defensible. This guide is the honest version of how it works, how accurate it is, and where a teacher stays in control.

Last updated 10 August 2026.

Why marking is the problem worth solving

Workload is the reason teachers most often give for leaving, and marking is one of its heaviest components, a pattern the DfE’s own Working Lives of Teachers and Leaders survey keeps reporting. Long-form subjects like geography make it worse, because grading an extended answer well is slow and grading it fast usually means grading it thinly.

The Department for Education ran a programme to test whether AI could take the administrative weight off marking without taking the judgement away from teachers. OpenKit was selected to build a tool for it, and the brief was specific: assess handwritten work and hand-drawn geography maps, give useful feedback, and never let the machine be the one that decides a grade.

What Rubrical does

Rubrical takes a student’s work, typed or handwritten, and returns draft feedback aligned to the marking rubric. The teacher reviews it, changes what they disagree with, and signs the grade. That is the whole loop, and the order of it is the point: the model drafts, the teacher decides.

It reads handwriting through text recognition, so a scanned script goes in the same way a typed one does. It syncs classes from Google Classroom and Microsoft Teams, or runs on a manual setup for schools that use neither. And it reports back across a class, so a teacher can see where a whole cohort missed the same thing rather than rediscovering it thirty times.

The accuracy had to be earned, not claimed

Before building anything, we benchmarked. We took GCSE Geography long-form responses already graded by senior teachers, held them out, and scored generic AI tools against the same rubric. They matched the human grader 67% of the time. That is fine for a demo and indefensible at moderation.

Rubrical, a private retrieval-augmented system answering from the DfE content store and evaluated continuously against real marks, reached 92% on the same held-out set. That gap is the argument for building against your own domain instead of prompting a general model. The programme’s public target was a 50% reduction in marking time, and teachers report a 52% reduction in the developer’s tests, which the Department for Education records on page 98 of its February 2026 schools white paper, Every child achieving and thriving. Feedback came back around 2.5 times more detailed than teachers had time to write by hand, with satisfaction at 95%.

The teacher signs every grade

The design principle that survived every iteration is that nothing reaches a student unsigned. Rubrical drafts, the teacher overrides and signs, and the moderation trail records all of it. In a setting where a trust has to answer to governors, parents, and exam boards, an audit trail of who marked what is part of what the school is actually buying.

That is also why Rubrical runs on a private stack in a UK Azure tenant rather than a public model. Student work stays inside controls a school can evidence: OpenKit holds ISO 27001 and ISO 9001, the deployment operates to UK GDPR, and it passed a DfE data protection impact assessment before it scaled.

As Reuben McQueen, OpenKit’s CTO, put it at the programme announcement: “Our goal is reducing marking time by 50% while maintaining high-quality feedback, with level-appropriate explanations and follow-up exercises tailored to each student.”

Where this fits

If you want the engineering detail, inside the Rubrical build covers how the private stack and the evaluation framework were put together, and the full Rubrical case study carries the numbers and the rollout. The same pattern, a private model answering from approved content with a human signing every decision, is how we approach AI in education generally, not just marking.

Start with an audit

Most engagements start with an AI Audit and Transformation, a fixed-scope, fixed-fee piece of work that finds where AI earns its place in your business and where it does not, then puts the first of it into service. You leave with a written report and a prioritised 12 month roadmap.

Start the conversation ↗ How the audit works