Insights · Guides · From the classroom
The DeepMark Blog
Practical takes on AI-assisted marking, GCSE assessment, and giving teachers their evenings back — from the people building DeepMark.
LatestGCSE grades are often a coin flip
Ofqual measured how often students get the 'definitive' grade. In Maths it is 96% of the time. In English it is closer to 52%. The regulator published this in 2018 and almost nobody read it. Here is what it means for your Year 11 class, and what it means for the accuracy question everyone asks about AI marking.
By Geoff Waugh
We built an AI detector and binned it
We tested AI detection on 449 real GCSE answers and 61 written by AI. The free methods caught 1 in 61. Here is the data, and why we are not building it.
By Stuart Bourhill
The best AI marking tools for GCSE in 2027 (it's not us)
A comparison post that refuses to crown a winner. What each AI marking tool is genuinely built for, what Ofqual actually said, and the six questions that decide which one you want. Written by a vendor, with the bias declared.
By Stuart Bourhill
From governess to grammar school: what is next for British education
For 150 years we have known that one to one teaching works, and never managed to scale it. Why AI's real job in education is not marking but the data that makes personalised teaching possible again, and the honest limits of AI marking today.
By Geoff Waugh
It's all about the mark scheme
How accurate is AI marking? It's the wrong question. Any mark is only ever as good as the mark scheme behind it. A deep dive into assessment objectives, level descriptors, and why consistency beats accuracy.
By Stuart Bourhill
AI marking is dangerous
It's the wild west of AI marking right now. DIY ChatGPT, vibe-coded apps, and copyright landmines. Here's what's actually at stake for your data and your students' work, and the checklist to run before you let any tool near an assessment.
By Stuart Bourhill
AI marking that doesn't suck
Most AI marking tools aren't built for teachers. Here's how DeepMark approaches reliable, mark-scheme-anchored marking, with the repeatability data to back it up.
By Geoff Waugh