For the last two years, the education sector has been caught in a "Generative AI" fever dream.
We’ve marvelled at tools like ChatGPT that can write lesson plans in seconds or draft emails to parents. But as we enter the 2026 academic cycle, the novelty is wearing off. School leaders are asking a tougher question: Why can't these tools grade high-stakes exams reliably?
The answer lies in the difference between Horizontal AI (Generalist) and Vertical AI (Specialist). And for schools, that difference is the line between a helpful assistant and a liability.
From "Generative" to "Logical"
The biggest misconception about AI in schools is that its primary value lies in "content generation." But teachers don't just need more content; they need logical verification.
The demand is already there. According to Twinkl’s 2025 AI Survey, 39% of teachers explicitly list "marking and grading" as the top tool on their AI wishlist.
Why is this number so high? Because while 60% of teachers use AI for planning lessons, they still don't trust it to mark them. There is a vast chasm between an AI that can "chat" about a Physics concept and an AI that can trace the step-by-step logic of a student’s handwritten method marks.
Why "Generalist AI" Can't Mark Exams
At a recent gathering of the UK’s leading AI minds at Barclays Eagle Labs, the consensus was clear: the era of “do everything mediocrely" AI is ending.
In high-stakes environments like Finance, Medicine, and Education, "quite good" isn't enough. Generalist models (like standard Large Language Models) are designed for fluency; they are built to chat convincingly. They are not designed for accuracy or strict adherence to a mark scheme.
When a generalist AI hallucinates a fact in a poem, it’s a quirk. When it misses a marking criterion in a mock Physics paper, it erodes teacher trust and creates AI waste - bad data that requires human cleanup.
What is Vertical AI in Education?
Vertical AI refers to models that are deep-trained on a specific, narrow domain. Instead of knowing "a little bit about everything," they know "everything about one thing."
For Excelas, that "one thing" is examiner logic.
We moved away from general prompts to build a specialized engine trained specifically on mark schemes and handwriting recognition. The result is a 90%+ agreement rate with human markers.
- Horizontal AI: Can write a generic essay on Photosynthesis.
- Vertical AI: Can trace the step-by-step logic of a student’s handwritten calculation and award method marks according to awarding body guidelines.
In high-stakes assessment, that gap in accuracy is the difference between a system a teacher has to double-check, and a system they can trust.
Stressed teacher
Solving the "Feedback Lag" in 2026
The biggest search trend for school leaders isn't just "AI"; it's "Teacher Retention" and "Workload Reduction."
Currently, the "Marking Mountain" creates a three-week lag between a student sitting a mock and receiving feedback. By the time they get their paper back, the learning moment has passed.
Vertical AI provides the infrastructure of immediacy.
- Old Way: 3-week delay, generic feedback, exhausted staff.
- New Way: Instant analysis, specific interventions, energized staff.
We don't need AI to replace the teacher’s intervention. We need it to automate the mechanical task of marking so teachers can do what only humans can: mentorship, emotional support, and complex instruction.
The Bottom Line: Is Your AI Specialised?
As you plan your 2026 digital strategy, stop asking "Should we use AI?" and start asking "Is our AI specialised enough for the task?"
If you are using a chatbot to grade a mock exam, you are asking a generalist to do a specialist's job. It’s time to build a classroom infrastructure that values accuracy over chat.
Key Takeaways for School Leaders:
- Generalist AI (Chatbots) is built for fluency, not grading accuracy.
- Vertical AI (like ExamGPT) is trained on examiner logic to ensure compliance.
- AI waste (hallucinations) creates more work for teachers, not less.
- 90%+ Accuracy is the new benchmark for automated assessment.
Ready to see Vertical AI in action? See how ExamGPT grades handwritten exams with specialist accuracy.