TL;DR

Across England, just 14% of classroom staff rate themselves confident on every dimension of AI literacy, according to Royal Society work built on weighted Teacher Tapp responses from over 9,200 primary and secondary teachers. The bigger finding sits underneath that headline: almost a quarter receive no support of any kind, and fewer than one in five have had formal training.

What “confident” was measured against

Respondents were tested on three things — grasping the mechanics, using the tools safely, and being able to interrogate the ethical and civic consequences. A third managed only the practical middle one. Another third were shaky on all three.

Support is where the survey stops describing a skills gap and starts describing a policy failure. Structured provision is rare: clear school-wide guidance reaches 15% of teachers, and formal courses 18%. A third fall back on informal advice from colleagues; another fifth are simply working it out alone. Where both training and guidance exist, teachers are around three times likelier to report full confidence — which makes the shortage the variable, not appetite.

Classroom coverage is thinner still. Under 2% report it running through most of what their school teaches. More than a third report it is not taught anywhere, and a fifth do not know. Add those together and roughly three in five cannot confirm their pupils learn any of this.

The gaps within the gap

The demographic splits are wide enough to matter. Male teachers report confidence at nearly three times the rate of female colleagues. Under-30s roughly double the over-50s. Headteachers score below the staff they lead, at 9%, and teachers carrying SENCO duties match that lowest figure while receiving the least structured support of any group measured.

Then the equity finding: in schools with more pupils on free school meals, 42% say AI literacy is absent from teaching altogether, against 33% elsewhere. Exposure to the subject is starting to track disadvantage.

Looking forward

The Royal Society’s argument is that this cannot stay a computing-department problem, and its review of existing frameworks found they lean heavily towards coding and algorithms while neglecting societal and environmental questions. It also notes the four UK nations are pursuing separate approaches with no shared baseline. That echoes an earlier finding that teachers who save time with AI simply absorb the hours elsewhere: without structure, the technology reshuffles work rather than improving it.