How to Use AI to Mark Physical Exam Papers at School

One of the most common misconceptions about AI marking tools is that they only work for typed submissions. Teachers imagine students uploading Word documents or copying and pasting their essays into a text box - and dismiss the whole category as irrelevant because their students write by hand, on paper, the way they will in every real exam. That assumption is understandable. It is also wrong.

For many UK secondary teachers, handwritten papers are exactly what they would want AI marking for. The sections below cover how it works in practice, what to know about entering mark schemes for AQA, Edexcel and OCR, and GradeOrbit's workflow from photograph to marked feedback.

Why Handwritten Marking Is Still the Norm

Despite the growth of digital learning platforms, the overwhelming majority of formal assessment in UK secondary schools still happens on paper. GCSE and A-Level mock exams are written by hand because terminal exams are written by hand, and training students to work under exam conditions matters as much as the content they produce. Internal assessments at KS3 and KS4 often follow the same format. Even coursework components - NEAs, controlled assessments, fieldwork write-ups - are frequently submitted as handwritten documents for subjects including Geography, Religious Studies, and Art.

This means that any AI marking tool designed exclusively for typed text is solving the problem for a small subset of teachers. Most secondary school teachers carry home physical scripts, not digital files. The marking workload that consumes evenings and weekends is almost always a pile of paper, not a collection of PDFs.

How AI Marking for Physical Papers Works

GradeOrbit's workflow for physical exam papers has two entry points. The first is a direct upload from your desktop or laptop - photograph the papers with your phone's camera app and transfer them to your computer to upload in bulk. The second is a QR code camera link that appears in the GradeOrbit interface: scan it with your phone, and you can photograph papers directly into the platform without any file transfer step.

Once the images are uploaded, GradeOrbit's AI reads the handwritten text straight from the page images and marks it in the same pass. It handles a wide range of handwriting styles, including the kind of compressed or inconsistent writing that appears under exam pressure. For particularly difficult scripts - very small handwriting, heavy crossings-out, or non-standard letter formation - compare the transcript of what the AI read with the scan before you accept the marks. In maths, an answer whose handwriting is hard to read gets a second reading, and the question is flagged if the two readings disagree.

After transcription, the marking process is identical to the typed workflow. You provide your mark scheme, the AI generates criteria-referenced feedback, and you review and approve the output before it is finalised. The handwriting recognition step adds a small amount of processing time, but it does not change the marking quality for legible scripts.

Inputting Your Mark Scheme

The quality of AI marking depends almost entirely on the quality of the mark scheme you provide. GradeOrbit does not apply a generic rubric - it marks against the specific criteria you give it, which means it works for AQA, Edexcel, OCR, Eduqas, WJEC, or any internal assessment framework your school uses.

For marks-based mark schemes: common in GCSE Maths, Science, and Geography - you paste the point-scoring criteria directly into GradeOrbit. The AI identifies which points the student has addressed, flags omissions, and suggests a mark based on the criteria met. This is particularly effective for structured questions where the answer is either present or absent.

For levels-based mark schemes: the norm for extended writing in subjects like English, History, Sociology, and Psychology - you paste the levels descriptors, including assessment objective weightings where relevant. GradeOrbit reads the student's response against each level descriptor and suggests a band placement with a rationale. AQA, Edexcel, and OCR all use slightly different levelling structures, and because you are providing the actual published mark scheme rather than a generic framework, the output reflects the specific criteria the examiner would apply.

For mock exams using past papers, you can often find the official mark scheme on the exam board's website and paste it directly into GradeOrbit without any modification. For internal assessments with bespoke criteria, you enter your own criteria in the same way. The platform does not require the mark scheme to be in any particular format - it reads plain text descriptions and applies them.

Redacting Student Information Before Processing

UK GDPR requires careful handling of any data relating to identifiable individuals, and student work - particularly named exam scripts - falls squarely within that scope. GradeOrbit includes a client-side redaction tool that allows you to draw black boxes over any personal information in an uploaded image before it is processed.

In practice, this means covering the student's name, candidate number, and any other identifying information that appears on the paper. On an individual teacher account, the redaction is applied directly to the image on your device before the image leaves your browser - GradeOrbit's servers never see the unredacted version - and student work is processed for the purpose of generating feedback and then discarded. Nothing is stored after the session ends. School accounts work differently: work is retained under the school's signed Data Processing Agreement for an agreed retention window.

For anonymous marking, where papers are already identified only by candidate number rather than name, redaction may not be necessary at all. But having the tool available means you can use GradeOrbit with named scripts when anonymous marking is not possible, without compromising your GDPR obligations.

What the AI Returns - and How You Use It

GradeOrbit's marking output for a handwritten paper is the same structured breakdown it produces for any submission: a criteria-referenced comment for each section of the mark scheme, a suggested mark or band placement, and a summary feedback paragraph. For each piece of work, you see the transcribed text alongside the feedback, which makes it straightforward to verify that the transcription was accurate before accepting the marking output.

The teacher's role in this workflow is review and approval, not generation. The AI produces a first draft of the feedback; you read through it, adjust anything that needs adjusting, and approve it for the student's record. With familiar content, much of the drafted feedback may need only light editing. For less familiar content, or for responses that sit at the boundary between mark scheme levels, your professional judgement does more work - which is exactly how it should be.

An A-Level essay that might take 12–18 minutes to mark from scratch is quicker to check once a drafted mark and comment exist, and structured GCSE responses are usually quicker still. Across a class set of 30 papers that difference adds up - and the feedback quality, because it is criteria-referenced rather than impressionistic, is often more consistent than when marking is done late in the evening against an increasing cognitive load.

For more on the handwritten marking workflow, see our guide to using AI marking for handwritten student work.

From Photograph to Approved Mark

For a paper-based class set, the steps in GradeOrbit are: photograph the scripts, provide your AQA, Edexcel or OCR mark scheme, check the transcription against the original image, then review each drafted mark and comment. The transcription check matters most on the hardest-to-read scripts.