Two years ago, ChatGPT's arrival triggered a panic in UK education circles. Schools scrambled for policy responses. Academics argued about whether AI-generated essays represented a new form of plagiarism. Examiners worried about the integrity of assessments. Government officials promised guidance.

The panic has subsided. In its place is something more complex: a profession and a system that is improvising its way through a transformation it did not design for and for which it still lacks adequate policy or professional infrastructure.

What's Actually Happening

The most honest answer to "what are UK schools doing with AI" is: everything, inconsistently, with significant variation by school type, teacher confidence, leadership approach and student age.

At one extreme are schools that have implemented effective AI literacy programmes — teaching students how large language models work, what their limitations are, how to evaluate AI output critically, and how to use these tools productively for learning rather than task-avoidance. These schools treat AI similarly to how a previous generation treated the internet: as a potentially powerful learning tool that requires explicit instruction in critical use.

At the other extreme are schools that have issued blanket bans on AI tool use for homework and coursework, which are discovering those bans are difficult to enforce. When every student has a smartphone with AI built into the keyboard, browser and productivity apps, prohibiting AI use is not a policy; it's a wish.

The majority of schools occupy the middle ground: informal norms have developed, some teachers explicitly allow AI assistance with appropriate attribution, others don't, and school-level policy is vague enough that coherent practice is impossible.

The Teacher Preparation Gap

The most significant finding from the 2026 Ofsted digital landscape survey is the gap between student AI use and teacher AI training. An estimated 65% of secondary school students use generative AI tools for homework at least monthly. Around 28% of secondary school teachers have received any formal training on AI tools — and this falls to around 11% in primary schools.

Teachers who have not used the tools themselves, in any meaningful way, cannot teach students to use them well. They cannot identify AI-generated work by pattern (the patterns shift constantly), cannot explain why AI sometimes produces plausible-sounding false information, and cannot help students understand the appropriate and inappropriate uses of these tools.

The professional development infrastructure — initial teacher training, in-service training through INSET days, CPD programmes — has not yet systematically incorporated AI literacy. Individual schools with motivated leaders and tech-interested staff have run excellent programmes. This remains the exception.

The Assessment Problem

The deeper structural challenge is assessment. The UK's assessment system — GCSEs, A-levels, vocational qualifications — was designed assuming that written work submitted by a student was produced entirely by that student. The significant growth in AI-assisted writing challenges this assumption in ways that exam boards are still working through.

Current approaches include:

  • Handwritten examinations: these remain largely AI-proof; universities and schools that have shifted weight toward in-person written assessment are responding rationally to the AI context
  • Oral examinations and viva components: adding verbal examination of understanding to written work; effective but resource-intensive
  • AI detection tools: unreliable, generating both false positives (flagging non-AI work by students with atypical writing styles) and false negatives (AI-generated work that passes detection)
  • Changed task design: shifting assessment away from "produce an essay on X topic" toward tasks that require demonstrable personal experience, local knowledge or real-time decision-making

No single approach solves the problem, and the exam boards have been cautious in publishing definitive guidance while the tools evolve. The Universities and Colleges Admissions Service (UCAS) personal statement format was redesigned in 2025 partly in response to AI — but the underlying question of how to assess learning authentically in an AI-assisted world has not been answered.

Where It's Working

The schools with the most positive experiences share certain characteristics: leadership that has engaged with AI tools genuinely (not just published a policy), teacher professional development that is practical and specific, and a pedagogical framing that treats AI literacy as a curriculum objective in itself.

In one London comprehensive, year 10 students complete a unit in which they are given a factually incorrect AI-generated essay and asked to identify and correct the errors using primary sources. The exercise teaches source criticism, AI output evaluation and the research skills that pre-date AI — simultaneously. Students report finding it engaging precisely because it is genuinely challenging. The AI's confident wrongness is motivating in a way that abstract fact-checking exercises are not.

Several primary schools have introduced age-appropriate AI literacy through practical exploration: children interact with a simplified language model, ask it questions, notice when it makes things up, and develop a practical intuition about what these tools are and aren't doing. This builds cognitive schema before students encounter the more powerful consumer tools in secondary school and beyond.

The Productivity Paradox for Teachers

Teachers are significantly more positive about AI as a professional tool for their own work than they are confident in using it as a teaching tool. AI assistance with lesson planning, differentiation materials, communication writing and administrative tasks is widely reported as genuinely time-saving by the teachers who use it.

This represents both an opportunity and a challenge. The opportunity: if AI tools reduce administrative burden on teachers — which the profession urgently needs given workload-driven recruitment and retention problems — that is a genuine benefit. The challenge: teachers developing productive AI habits for their own work, without also developing the pedagogical frameworks to teach students about those tools, risks deepening the teacher-student gap rather than closing it.

The emerging consensus among education technology researchers is that AI adoption in schools goes best when teachers are positioned as learners alongside students — exploring the tools together, developing shared frameworks for appropriate use, and building the kind of critical relationship with AI that a highly educated adult who thinks carefully about these tools has. That requires time, trust and investment that the education system has not yet fully provided.

What Comes Next

The Department for Education's AI in Education team has signalled that updated guidance is expected in late 2026, likely to include clearer frameworks for assessment integrity, professional development standards and curriculum requirements. Ofsted's inspection framework is expected to be updated to evaluate schools' AI readiness as a component of digital education provision.

The more immediate change agent is likely to be the devices themselves. AI is being progressively embedded in the operating systems and productivity suites that schools use — Microsoft Copilot in the Office suite, Google Gemini in Google Workspace for Education — creating situations where AI assistance is structurally available in the tools students use for schoolwork, regardless of any school policy. Managing AI in education increasingly requires engaging with the tools students actually have, not the tools schools imagined they were using.