Skip to main content
GradeDrive. Send it!GradeDriveSend it!
Sign inStart marking free

AI in Education · 6 min read

Does AI Marking Follow Your Mark Scheme? What Happens When It Disagrees With You

Does AI Marking Follow Your Mark Scheme? What Happens When It Disagrees With You

GradeDrive marks against the mark scheme you upload, shows which criterion each mark came from, and records nothing until a teacher has seen it. If you disagree with a mark, you change it. This post shows what that looks like on screen, what happened across 216,000 marks last summer, and where you should expect to do more of the deciding yourself.

What "following the mark scheme" looks like on screen

You upload the mark scheme and a scan of the class set. GradeDrive reads each answer against the scheme criterion by criterion and suggests a mark for each question.

The review screen shows the student's script on one side and the model answer and marking criteria on the other. Where GradeDrive awarded a mark, it shows which criterion it matched. That is the part that matters when you are checking someone else's marking: you can see the reasoning, so you are not looking at a bare number.

GradeDrive does not bring its own criteria to the paper. The scheme you uploaded is the only standard it marks against, which is why it is not tied to one exam board.

What happens when you disagree

You change the mark. Any suggested mark can be changed, and nothing is final until you have confirmed it. Once you are happy with the marks, you export them as a CSV, a feedback sheet for each student, or a class analysis report.

GradeDrive does the first read-through, and the decision stays with the teacher.

What 216,000 marks showed

Across summer 2026, GradeDrive processed 216,000 marks over 2,300 submissions. Every one of them was shown to a teacher before it counted. 99.2% survived that review exactly as GradeDrive first suggested.

The other 0.8% is worth looking at. About 700 submissions had at least one mark changed, an average of around 2.5 marks per paper in that group, 1,750 marks in total.

Two cautions about what that number can and cannot tell you:

  • It is our own measure, drawn from our own production database. It is an override rate, and it is not an independent accuracy benchmark. There is no chief-examiner comparison behind it.
  • A changed mark tells you the teacher and GradeDrive disagreed. It does not tell you who was right.

For an independent read, two case studies on SSRN compared GradeDrive with official AQA marks, finding a quadratic-weighted kappa of 0.885 on GCSE Physics and 0.864 on A-level Maths. The accuracy page has the methodology and the limits of both.

Where you should expect to decide more

GradeDrive works best when the scheme states its criteria clearly. That is why it officially supports four subjects: Biology, Chemistry, Physics and Maths, with English still experimental.

Within those subjects, expect to change more marks where a scheme leaves room for judgement, such as a borderline level-of-response answer, or where the scan is hard to read. Handwriting and page edges that are clear in the scan give the best results.

If you want to test this properly, mark one class set and compare. Look at the marks you changed and ask whether the scheme was ambiguous, the scan was poor, or the suggestion was simply wrong. That tells you how much to trust it on your own papers.

Common questions

Can I stop it using criteria that are not in my mark scheme? It marks against the scheme you upload. The review screen shows the criterion behind each mark, so you can see when a mark does not match the scheme and change it.

Does it work for any exam board? It marks against whichever scheme you upload.

Is it free to try? Every account starts with a one-time free allowance. See pricing.

Related reading

Ready to reclaim your evenings?

Join teachers across the UK using GradeDrive to mark papers faster, more consistently, and without the Sunday-evening dread.

Start for freeSee how it works