Getting AI training right in UK schools

By Chris Calder, AI Education Consultant | Meta Pedagogy

The problem with most AI INSET

Most AI training for school staff fails in the same way. Someone demonstrates ChatGPT for an hour, staff nod politely and nothing changes. Two weeks later the same people are planning lessons exactly as they were before. The school has spent a day’s INSET time and a training budget on a session that produced no lasting change in practice.

Staff resistance gets blamed for failed AI training. Most of the time the training itself is the issue.

Lead with the work, not the tool

Staff do not care about AI. They care about their marking pile, their Year 9 class and the reports that are due. An INSET session that opens with “let me show you ChatGPT” has lost half the room before it starts.

Open with a task they recognise. “By the end of this session you will have differentiated your Year 9 text into three reading levels.” That is specific, useful and achievable in the time available. Staff can see the value immediately and they leave with work they can use the next day.

Structure for active use, not passive watching

The biggest mistake in AI INSET is too much demonstration and not enough doing. A session that is 50 minutes of showing and 10 minutes of trying produces spectators, not practitioners.

A better structure: 15 minutes of demonstration, 45 minutes of hands-on work. During the hands-on section every member of staff works on their own materials with support available. They are not following a script. They are building something they will actually use in their classroom. The demonstration should show polished examples from multiple subjects so that staff across departments can see themselves in the work.

Handle resistance honestly

In any staffroom you will have three groups. Some staff are already experimenting, often without oversight or guidance. Some are curious but cautious. Some want nothing to do with it.

The third group has reasons. They have heard years of technology promises that did not deliver. Interactive whiteboards were going to change education. Virtual learning environments were going to change education. Some staff are on their third or fourth cycle of being told that this time the technology really will make a difference.

Do not dismiss the scepticism. Address it directly. Be honest about where AI works well and where it falls flat. Acknowledge that there are real accuracy problems. Show the school’s actual policies on what is and is not acceptable. Clear boundaries do more for staff confidence than any amount of enthusiasm about what AI can do.

Cover safeguarding, not just productivity

An AI INSET that only covers lesson planning and marking misses the safeguarding dimension entirely. Under KCSIE 2026 all staff are expected to understand how filtering and monitoring apply to AI tools (paragraph 12). That includes non-teaching staff.

Build safeguarding into the session. Cover what staff must not enter into AI tools (student names, personal data, anything that identifies an individual student). Cover what AI-generated imagery means under the updated nudes and semi-nudes definition. Cover the school’s reporting process for AI-related safeguarding concerns. KCSIE 2026 makes this a statutory expectation.

Follow up, or it was wasted

The most expensive part of an AI INSET is not the training itself. It is the opportunity cost of a day where staff are not teaching. Without follow-up that cost produces nothing.

Two weeks after the session, check in. Ask each department to show one thing they have used AI for since the training. Share what staff have produced. Identify the staff who are most engaged and develop them as department-level champions who can support colleagues in their own subject context.

Track what matters: how many staff are using the approved tools, how much time they report saving, what quality issues they are finding. When you can report specific numbers, continued investment in AI becomes an evidence-based decision.

What a good AI INSET covers

A single 60-minute session cannot cover everything. Prioritise based on where the school is. For schools at the beginning, the first INSET should cover three things: what the school’s AI policy actually says (acceptable use, approved tools, what never to enter), one practical workflow staff can use immediately (planning, feedback or differentiation) and the safeguarding basics under KCSIE 2026.

Follow-up sessions can then cover subject-specific applications, assessment integrity, student-facing AI use and advanced workflows. A phased approach across a term produces better adoption than a single intensive session that overwhelms.

Common mistakes

Running the session in a room with no wifi or devices. It sounds obvious but it happens. Hands-on AI training requires every participant to have a working device with internet access to an approved AI tool. Check this in advance.

Using an external trainer who has never worked in a school. Staff know within five minutes whether the person at the front understands their reality. A consultant who references “the corporate sector” or “industry best practice” without translating it into classroom terms will not be taken seriously.

Treating all staff the same. The teacher who has been using AI for six months does not need the same session as the teacher who has never opened ChatGPT. Where possible, differentiate the training or offer parallel tracks.

Forgetting support staff. Teaching assistants, office staff, SENDCos and pastoral teams all interact with AI in different ways. A session that only addresses classroom teachers misses a significant part of the workforce.

What your school needs

If you are planning an AI INSET and want it designed around your school’s staff, context and policy framework, book a free consultation with Meta Pedagogy.