Safeguarding, shaped by schools.
Better questions lead to better protection.
From keyword alerts to the context of a lesson: how schools keep moving our safeguarding thinking forward.
The backstory
It starts with listening, not a feature list.
SchoolGPT is built around a simple division of expertise: educators understand their pupils and their classrooms; we bring the engineering. We do not tell schools how to teach. We ask what they need, build with them, and listen when the first answer is not enough.
Safeguarding is a good example. Schools told us about systems that raised an alert whenever a particular word appeared. That can become an accepted part of the working day: another alert to open, another false alarm to dismiss. We wanted to understand how to make that experience better, without overlooking a child who needs help.
Version 1
Spot the concerning words
Let us know when a pupil writes something concerning.
The first approach focused on keywords. But a word alone tells you very little: a surname containing 'gun' could trigger the same attention as something genuinely concerning. Staff needed a useful signal, not an ever-growing list of words to investigate.
What it taught usFinding a word is not the same as understanding a concern.
Version 2
Understand what is being written
Can we reduce the false alarms without losing the important ones?
We moved towards AI-based interpretation of the writing, rather than treating every keyword as an incident. Looking at meaning gives the system a better basis for distinguishing an ordinary reference from something that needs a member of staff's attention.
What it taught usRead the message, not just the word that caught your eye.
Version 3
Make room for creative writing
Our pupils are writing Gothic stories. The genre matters.
A classroom writing task can include unsettling themes without being a personal disclosure. Schools pushed us to account for genre and purpose. A fictional scene and a child describing their own situation need to be considered differently, without assuming that all fiction is harmless.
What it taught usThe task is part of the context, not a reason to ignore a concern.
Version 4
Bring the lesson into the picture
After PSHE, pupils ask questions about what they have just learned. Can the AI take that into account?
The next brief goes beyond the sentence and the genre: understanding what has been taught. A question prompted by a PSHE lesson should not automatically be treated as a disclosure. Equally, a real concern can surface during that lesson. The aim is better-informed escalation, not a blanket exemption for a subject.
What it taught usKeep asking what additional context would help staff make a sound judgement.
Why this matters
Schools keep raising the standard.
Every version began with a school explaining a real situation. That is what we mean by guided by schools: not asking educators to accept the limits of a tool, but working together to make it more useful.
Explore our teaching approachThis is the story of an evolving approach, not a promise of zero false alerts or missed concerns. AI supports safeguarding; it does not replace the designated safeguarding lead, professional judgement or school procedures.