What Grantham schools are already doing
Somewhere in a staffroom in Grantham this term, a teacher is deciding whether to let a pupil use an AI tool to draft an essay plan — and whether that counts as help or as doing the work for them. It is a small, everyday decision, but it sits at the centre of something schools are only beginning to work out.
The King's School Grantham and Kesteven and Grantham Girls' School (KGGS) have both introduced AI in ways that treat it as a supplement rather than a substitute. Computing lessons and STEM workshops are the primary settings — contexts where pupils examine how the technology works, not just what it produces. At KGGS, AI ethics has also been brought into debate and curriculum activity, which means students are being asked to question the tool, not simply use it. That framing — AI as a subject of critical enquiry — is a meaningful choice, even if it does not yet extend across every classroom.
The clearest piece of hard local evidence comes from The Westminster School, which published a formal AI Policy in March 2025 covering staff, governors, and pupils — explicitly including generative chatbots such as ChatGPT and Google Bard. A published policy is not the same as settled practice, but it signals institutional intent, and it is a relatively rare primary source from inside South Kesteven. What it reflects, across all three schools, is a cautious, opt-in approach: deliberate moves into specific areas rather than any wholesale shift in how teaching works.
The rules teachers are working inside
Behind those individual classroom decisions sits a defined national framework. The Department for Education published its first formal AI guidance for English schools and colleges in June 2025, updated again in August, and it draws a clear line: teachers may use AI to help with lesson planning, resource creation, marking, feedback, and administrative tasks — but professional responsibility for every output stays with the teacher, not the tool.
The guidance is explicit about what cannot be handed over. Statutory contributions to Education, Health and Care Plans for SEND pupils are a named example: AI may draft initial support material, but the professional judgement required for a statutory EHCP must remain the teacher's own work. The same principle applies to safeguarding decisions and any context where identifiable pupil data is involved — staff are prohibited from entering such information into public AI models, and many popular tools carry an 18-plus age requirement that schools are obliged to respect.
Lincoln College has gone a step further, adopting a bespoke AI framework for staff and students that is subject to annual review — one local institution that has chosen to build above the national floor rather than rest on it. The broader picture for Lincolnshire teachers, then, is not a vacuum: it is a defined baseline with room for local discretion above it.
Most teachers want to use AI and don't know how
The numbers from a DfE survey published alongside the June 2025 guidance are striking in combination: 43% of teachers rate their AI confidence at just 3 out of 10, and more than 60% explicitly asked for help applying AI to planning and support tasks. Teacher adoption, meanwhile, has nearly doubled since 2023 — rising from 31% to 58% in two years, according to the National Literacy Trust's 2025 survey of nearly 3,000 UK teachers. That pairing tells its own story. Uptake has accelerated well ahead of the confidence needed to use AI well.
This matters beyond the staffroom. A teacher who is uncertain about a tool is caught between two risks: adopting it uncritically, handing over tasks without interrogating what comes back; or avoiding it entirely, leaving pupils to encounter AI outside school without any professional framing. Neither outcome serves the classroom well. The gap between willingness and capability is not a personal failing — it is a structural one, the product of very rapid change against a professional development infrastructure that has not yet caught up. Almost all respondents in the DfE survey called for safety guidance and direct training, which suggests the demand for support is clear; what is still being built is the supply.
When AI becomes the subject, not the shortcut
There is a meaningful difference between a pupil who uses AI to answer a question and one who understands how AI arrives at an answer. Hyett Education, a STEM provider working across Lincolnshire, tries to create the latter. Their 'Train Robots with Machine Learning' workshop — available to schools from KS2 through KS4 — gives pupils as young as 7 the chance to train image-recognition models that then control LEGO robots, with no prior coding experience required. The point is not to produce engineers; it is to make the technology legible.
The sessions are funded through RAF Youth & STEM and Defence Nuclear Enterprise partnerships, which removes the direct cost barrier for Lincolnshire schools — a practical reason why this kind of specialist provision can reach schools in rural or less well-resourced areas rather than remaining confined to institutions with larger budgets.
That distinction between using AI and understanding it has particular weight given what national data shows about how young people already engage with generative tools. Two in three UK pupils aged 13 to 18 now use generative AI, and 61% do so specifically for homework help or question-answering. When use becomes that habitual, there is a real risk that the technology functions as a black box — producing outputs students accept without being equipped to test them. Workshops that expose the mechanics offer a counterweight: a pupil who has trained a model is considerably better placed to interrogate one.
Does using AI actually stop students from thinking?
One study cannot settle a pedagogical debate, but a neuroscience experiment at the University of Tokyo has at least sharpened the question. Researchers found that students using ChatGPT to work through problems showed measurably reduced brain activity compared with those tackling the same tasks independently. The finding is suggestive rather than conclusive — a single study under controlled conditions does not map directly onto a Lincolnshire classroom — but it points to something educators are already grappling with intuitively: that the act of struggling with a problem may itself be part of what builds the capacity to evaluate any answer, AI-generated or otherwise.
The calculator analogy is useful here. When pocket calculators became widespread, exam boards did not ban them or simply ignore them — they created parallel tracks: a calculator paper and a non-calculator paper. Some educators now argue that AI may require the same structural response: assessment modes that explicitly demand unaided reasoning alongside those that permit AI assistance, so that foundational skills are built and tested rather than assumed.
The dependency paradox sits at the centre of all of this. AI is only genuinely useful if the person using it can judge whether the output is correct. But if students reach for AI before they have developed that judgement, they may end up less equipped to catch the tool's mistakes — not more. The risk is not that students use AI, but that they use it before the foundations are in place to evaluate what it returns. That is less a moral problem than a sequencing one — and sequencing is precisely what good teaching is designed to manage.
How schools are trying to keep thinking visible
The practical response taking shape across English schools is not a ban and not a free pass. Instead, assessment is being redesigned around the learning process itself: teachers are requiring students to submit drafts at multiple stages, to defend their work verbally, and to take part in viva-voce discussions where they must demonstrate understanding in real time. A finished essay can be generated; a fifteen-minute conversation about how a student arrived at a conclusion is considerably harder to fake.
Universities are moving in a different direction. UCL's law faculty is among those that have returned to 100% closed examinations — a response driven partly by a documented rise in AI-related malpractice cases. Some educators regard this as the only reliable guarantee of integrity; others see it as a retreat from the kind of learning that actually prepares graduates for work. Both positions are held with conviction, and neither has settled into orthodoxy.
Schools across South Kesteven are making these calls without a finalised national template — course by course, year by year, often policy document in hand but specific answer still pending. The teacher from the opening scene, deciding whether an AI-drafted essay plan counts as help or as the thinking itself, is not an outlier facing an unusual dilemma. She is doing exactly what every school in the county is doing: working out the answer in real time, with real pupils, and no settled model to copy.