# Top Marks AI — full LLM digest Source of truth for product copy on the redesign: packages/landing/content/site.json (also editable via Retool → Mongo website_content). Canonical public site: https://www.topmarks.ai Generated for AI crawlers / answer engines. HTML pages remain the human UI. ## SEO Title: Top Marks AI | AI marking for schools and MATs Description: Top Marks AI batch marks handwritten GCSE, A Level, KS3, IB and IELTS scripts, with 90%+ correlation to the chief examiner and detailed feedback for every student. ## Home # Mark a whole Year 11 mock in a morning — not over half-term. Top Marks AI batch marks handwritten scripts across GCSE, A Level, KS3, IB and IELTS — with 90%+ correlation to the chief examiner and detailed feedback for every student. Your teachers review the marks; they don't start from a blank pile. - Independently validated by Ark Schools & Community Schools Trust - Featured on BBC News ### Problem: Mock season swallows weekends and half-term. A Year 11 English mock can mean 30+ hours of marking per teacher — on top of a full timetable. Fix: Scan the whole cohort as one PDF. Every script is transcribed, split by barcode and marked — 3,000 pages in a single morning. ### Problem: Marks drift between teachers and classes. Even experienced markers land only ~45% of scripts within the board's tolerance. Moderation catches some of it — at a cost. Fix: Each tool is benchmarked to board standardisation materials, so every script is marked to the same standard — ~84% within tolerance. ### Problem: Feedback arrives too late to act on. By the time scripts come back, students have moved on — and comments are often too brief to be useful. Fix: Every student gets detailed, scaffolded ScaMP feedback per Assessment Objective, plus Whole Class Feedback for your next lesson. ## Workload claim What is marking really costing your school? Based on a UCL finding of ~230 marking hours per teacher per year, and Top Marks AI's estimated 55% reduction. FTE uses 1,265 hours of directed time. ## Accuracy # We publish our accuracy. Every tool, every board. Accuracy claims are only worth the study behind them. Ours are measured against board standardisation materials, published per question, and corroborated by the trusts that use us. - 0.94 — correlation · AQA GCSE English Language - 0.91 — correlation · OCR English Literature - 0.90 — correlation · Edexcel IGCSE English - Within tolerance (0–3 marks): human ~45%, Top Marks AI ~84% - Moderate difference (4–6 marks): human ~28%, Top Marks AI ~15% - Large difference (7+ marks): human ~27%, Top Marks AI ~1% Human baseline: research on experienced GCSE English examiners marking against chief-examiner scores (Fowles, 2009) found a Pearson correlation of around 0.65, with only ~45% of marks inside the board's tolerance. ### One fine-tuned model per tool Each of our 400+ tools has its own dedicated model, trained by an expert examiner for that qualification and question. ### Benchmarked to board standardisation Every tool is algorithmically benchmarked against official standardisation materials before it reaches a school. ### Measured, then published Correlation, Mean Absolute Error and % in tolerance — published per question, so you can check the tool you'll use. ## Case study — Wensleydale (BBC) Wensleydale School · Leyburn, North Yorkshire · State secondary One of the first UK state schools to AI-mark GCSE mocks in a live setting The school set out to test whether AI could reliably apply exam board standards. Teachers marked alongside the platform and compared every script against their own marking. Quote (Charlie Barnett, Deputy Headteacher): "This was never about replacing teachers. The purpose of the trial was to see whether AI could reliably apply exam board standards." BBC: https://www.bbc.co.uk/news/articles/c6207xenj1do - **Ark Schools:** Independently corroborated Top Marks AI's accuracy findings. Schools within the trust use AI-generated marks to inform target-setting. - **Community Schools Trust:** Independently corroborated our accuracy findings as part of a trust-wide evaluation of AI marking. ## About We're a group of AI and education specialists combining pedagogical expertise, teaching experience and machine-learning techniques to develop the next generation of AI-marking software. Our aim is to create an AI-marking platform that not only marks students' work as accurately as experienced teachers, but also provides innovative feedback that improves students' grades. ### Accuracy, consistency and fairness We not only test the accuracy of every single tool by marking sets of board-moderated essays — we publish the results on our blog. We want to be held to a high standard, and we're proud to say that our tools typically mark with greater accuracy and consistency than experienced teachers. This accuracy and consistency makes Top Marks AI perfect for moderating the marking of different teachers — whilst ensuring fairness for pupils. ### Pupil progress Our feedback is shaped by pedagogical theory so that pupils can genuinely understand new ideas and achieve better results. We're excited about creating new ways of working, and we offer innovative types of feedback that help to develop teachers as well as pupils. ### Alleviating the burden of marking Accurate marking and detailed feedback allow teachers to spend more time teaching and planning. We want to help schools to make teacher burnout a thing of the past, whilst also giving them the tools to support teaching and learning. ## Published accuracy studies (links) - [AQA GCSE] English Language Paper 2 — 16-mark comparison question — r=0.94 — https://www.topmarks.ai/ai-aqa-gcse-english-2-4 - [AQA GCSE] English Language — 40-mark creative writing — r=0.94 — https://www.topmarks.ai/ai-marking-outperforms-humans-aqa-gcse-english-creative-writing - [AQA GCSE] English Literature — Post-1914 drama and prose — r=0.94 — https://www.topmarks.ai/ai-gcse-marking-aqa-modern-plays-prose - [AQA GCSE] English Literature — 19th-century novel, 30-mark — r=0.97 — https://www.topmarks.ai/ai-gcse-marking-aqa-19c-prose - [AQA A Level] Politics — 25-mark extract question — r=0.96 — https://www.topmarks.ai/ai-marking-aqa-politics-a-level-extract - [OCR GCSE] English Literature — headline accuracy — r=0.91 — https://www.topmarks.ai/ai-marking-for-teachers-gcse-ocr-english-literature - [Edexcel GCSE] History — 16-mark question — r=0.90 — https://www.topmarks.ai/ai-gcse-marking-edexcel-16-history-explain - [Edexcel IGCSE] English Literature — literary heritage question — r=0.91 — https://www.topmarks.ai/ai-marking-top-marks-edexcel-igcse-literary-heritage - [Edexcel IGCSE] English — outperforming experienced human markers — r=0.90 — https://www.topmarks.ai/top-marks-outperforms-humans - [Edexcel A Level] Politics — head-to-head vs a competitor tool (0.48) — r=0.84 — https://www.topmarks.ai/ai-a-level-marking-politics-edexcel-no-source-30 ## Articles - [ASSESSMENT] Is AI Marking Fair? Bias, Transparency, and Trust Explained — https://www.topmarks.ai/is-ai-marking-fair - [LEADERSHIP] Classroom AI vs AI Marking Infrastructure: A Strategic Choice for MAT Leaders — https://www.topmarks.ai/classroom-ai-vs-ai-marking-infrastructure - [PEDAGOGY] ScaMP Feedback: Scaffolded, Modelled, Precise — https://www.topmarks.ai/scamp-feedback - [HOW WE BUILD] Why Our AI Marking Is Different — https://www.topmarks.ai/why-our-ai-marking-is-different ## Contact Discovery call: https://meetings-eu1.hubspot.com/jsnowden/discovery-call Login: https://app.topmarks.ai