What is actually changing in Grantham classrooms
On a wet Tuesday in October 2025, a secondary teacher in Grantham stays behind after the last bus and opens a generative AI tool to speed up the evening’s routine: draft ten retrieval-quiz questions for a Year 8 science class, and turn a pile of short answers into a set of tidy feedback comments. One line that comes back is crisply phrased — “Explain why the change happens, not just what happens” — and it is tempting to paste it straight into the markbook because it sounds so much like a “good” teacher comment.
That small convenience is also a quiet shift in who gets to shape quality. Educational assessment is often described as a systematic process of collecting and using data about what learners know and can do; once an AI-generated score, summary, or comment becomes part of that evidence, it can start steering what counts as a strong paragraph, a clear method, or a “complete” explanation in everyday schoolwork. When pupils revise to match the phrasing they see repeated — “add evaluation”, “use precise terminology”, “show working” — the tool is not only saving time; it is nudging the class towards its own pattern of “good work”.
The Cambridge classroom guide that introduces UNESCO’s teacher framework describes AI as software that can recognise patterns, process language and visual data, and make decisions — in other words, a decision-support system whose outputs still need human interpretation and oversight. In the ordinary rhythm of a South Kesteven week, though, the line between “support” and “standard-setting” can blur when a suggested comment becomes the default comment.
Rather than treating this as a chain of rhetorical questions, the focus is one practical tension: as AI tools arrive in Grantham-area classrooms, who is effectively defining “good work” — the teacher’s professional judgement, the software’s learned patterns, or the wider standards the tool was built to mirror?
There is little published material that documents day-to-day AI marking and feedback in Lincolnshire classrooms specifically, so the rest of this piece draws on national and international guidance alongside local observation to make sense of what is at stake for families, staffrooms and pupils when technology and humanity meet in the homework routine.
How do AI tools learn our idea of good work
In 2024 the UK government put £4 million into a “content store” designed to make generative AI classroom tools more “trustworthy” by giving them a bank of official material to learn from: curriculum guidance, lesson plans, teaching standards and anonymised pupil assessments, with an explicit aim of supporting marking, resource creation and routine admin in schools. In practice, it is a national decision about what counts as a reliable reference library for tools that will comment on pupils’ work in places like Grantham.
Most classroom AI does not judge like a human marker reading for meaning; it typically works by learning patterns from examples. Feed enough past marked answers, mark schemes, rubrics and model paragraphs into a system, and it becomes good at producing the kinds of sentences that look like familiar feedback (“define the key term”, “justify your conclusion”, “use evidence from the text”). That pattern-matching approach sits behind a lot of “AI in education” applications, including automated feedback and assessment, even when the interface feels like a chat.
Once official documents become training material, national expectations can be pulled directly into day-to-day comments. A Year 9 paragraph in a South Kesteven exercise book may be praised for the same features that appear repeatedly in the guidance the tool absorbed: coverage of required content, subject-specific vocabulary, and the sorts of structures exam questions reward. This can improve consistency, but it also means the tool’s default sense of “good work” is likely to resemble what the national system already values.
The 2024 announcement also leans heavily on the idea that “better data” from authoritative sources is what makes AI work properly for marking and planning, and it reports that parents tend to support AI when it reduces teachers’ out-of-hours workload and increases face-to-face time. Both points sound reassuring, yet they strengthen the pull of system-level standards: the more “authoritative” the source, the more it becomes the yardstick the tool reaches for.
Less visible is the implementation detail: public information is limited on how specific vendors will encode these materials, how strongly different sources will be weighted, and how clearly criteria will be shown inside the product. Without that transparency, it can be hard for a Grantham family — or even a classroom teacher — to see why a particular phrasing is rewarded, or why a surprising misconception is flagged, when the feedback is partly shaped by training choices made far from the classroom.
Where does the teacher’s judgement fit now
A marking decision can be made in seconds now: accept an AI-suggested comment, tweak two words, move on to the next book. That speed is exactly where professional judgement has to become more visible, not less—because the “teacher–AI–student” dynamic described in UNESCO’s 2024 AI Competency Framework is not just a new gadget in the classroom, but a new relationship in which judgement is shared, negotiated, and sometimes quietly deferred. Rather than re-listing the framework’s full structure, the useful point here is what it implies about daily practice: teachers are expected to keep hold of the evaluative “why” behind a mark, even when the tool is fluent and confident.
UNESCO frames its teacher framework around “protecting teachers’ rights” and “enhancing human agency”, and it sets out 15 competencies with progression levels (“Acquire, Deepen, Create”). Read plainly, that is a claim about who defines “good work”: not the model’s default, but a teacher who understands enough about AI to question it, and enough about ethics to notice when “consistent feedback” starts becoming narrow feedback. The same UNESCO text is explicit that AI does not remove the teacher; it changes what competence looks like when judgement is made with a tool on the desk.
Cambridge’s practical guide, built from the same UNESCO direction, makes that expectation feel closer to a South Kesteven staffroom: safe, effective and ethical AI use is presented as possible “regardless of teachers’ initial level of technological expertise”, as long as outputs are interpreted and overseen rather than treated as final. That “regardless” matters in real timetables, because confidence with the tool is likely to vary within one Grantham department in the same week that mock marking is due.
The international message is similar. The U.S. Department of Education’s report on AI in teaching talks about AI augmenting—not replacing—educators, and it explicitly frames classroom practice as a spectrum of “who is in control”. In other words, the policy intent is that the teacher remains responsible for the core decision, even if the AI drafts the rubric-aligned wording at 9pm.
That policy intent runs into a practical dilemma: workload pressure and accountability targets can make “accept suggestion” feel like a rational shortcut, especially with larger classes and repetitive question types. When an AI comment conflicts with what a teacher knows about a particular pupil’s misunderstanding, judgement shows up as a concrete action: overriding the phrasing, changing the next task, or refusing to convert a confident paragraph into a definitive mark.
One awkward question is still left hanging by many high-level documents: if an AI-influenced comment is challenged—say, for being biased, inaccurate, or simply misleading—who is accountable in a Grantham classroom? The teacher who clicked “accept”, the school that procured the tool, or the vendor that designed the model’s behaviour are all involved, but the line of responsibility is often less clearly spelled out than the reassurance that the teacher stays “in control”.
Who else sets the rules behind classroom AI
In a Grantham English classroom, the most visible moment is a single line of AI feedback on a Year 10 paragraph — “develop your evidence” — but the less visible part is the chain of organisations that quietly decides what counts as “evidence”, and how confidently a tool is allowed to say so.
One of the furthest “upstream” influences is UNESCO. Its 2024 AI Competency Framework for Teachers is positioned as a global reference intended to guide national frameworks, teacher training programmes and even “assessment parameters”. In other words, an international body is not setting a mark scheme for Lincolnshire, but it is trying to shape what competence, responsibility and human agency are expected to look like when assessment is mediated by AI — the kind of expectation that can filter into national guidance and procurement rules. UNESCO’s own phrasing about a “teacher–AI–student” dynamic matters here: it implies that the standards around good work increasingly include standards around how judgement is exercised when AI is present, not just the finished piece of writing.
Closer to home, parliamentary scrutiny points to the kinds of constraints that can end up as practical “rules of the road” for tools used in schools. A UK Parliamentary POST briefing notes that AI is being used to support lesson planning and marking, but flags concerns about fairness, bias and transparency; it also indicates that high-stakes assessment still tends to retain substantial human marking and oversight when AI is involved. That combination nudges classroom AI towards a co-pilot role: fast, helpful, but not treated as the final authority when the stakes rise.
The OECD frames the same tension at system level: AI can disrupt traditional homework and assessment models, so it calls for guidelines and guardrails to make use effective and equitable, rather than simply automating existing practice. In a South Kesteven context, those “guardrails” are not abstract. They are the difference between an AI comment being a vague instruction (“be clearer”) and being a contestable judgement with traceable criteria — for example, whether the system is required to show what standard it is applying, and whether a school can review patterns of bias or inconsistency across a year group.
Expert debate in higher education reinforces that direction. A Queen Mary University of London piece dated 12 January 2026 argues that “the real promise of AI in assessment is not to replace human markers, but to amplify their capacity to give rich, timely and meaningful feedback”. This is still a normative claim rather than a guarantee about how products behave, but it adds weight to a professional expectation that AI should support human judgement, not substitute for it.
Taken together, these actors — UNESCO, parliamentary briefings, and international policy work like the OECD — shape what vendors build and what schools feel comfortable adopting long before an AI tool is opened on a laptop after the last bus home. What remains thin in the public evidence is fine-grained, day-to-day observation in typical UK school settings: when an AI tool and a teacher disagree on what “good work” looks like in a specific Grantham exercise book, policy documents signal that the human should remain the final arbiter, but they say less about how that disagreement is logged, audited, or appealed in ordinary classroom practice.
What should count as good work when AI is everywhere
Assessment starts to look different once the baseline assumption shifts from “no tools” to “tools are in the room”. A 2024 Department for Education report on generative AI records that some experts want assessment redesigned on the basis that students will have access to AI and other digital aids, drawing on “computer-based math” approaches rather than trying to pretend the tools do not exist. In that framing, the question becomes less about catching AI use and more about specifying what counts as legitimate help, and what evidence of learning is still required.
One likely consequence is that “good work” may be judged as much by process as by product. A fluent, tidy paragraph on Macbeth that offers three “points” and a quotation can look convincing at 9pm — even if it has the tell-tale smoothness of an AI draft and quietly misreads a scene. Under an assessment design that assumes tool access, the mark may hinge on things that sit alongside the final paragraph: a short log of prompts used, a note of what was accepted or rejected, and a brief check against the text itself (“Act 2, Scene 2 doesn’t support this claim”). That kind of “AI audit” can be as plain as a Year 9 appendix: “Prompt 1: summarise the argument; Prompt 2: give two alternative quotes; I didn’t use quote 2 because it wasn’t in my copy.”
UNESCO’s 2024 AI Competency Framework for Teachers is aimed at “enhancing human agency” and builds ethics directly into what competence should look like when AI is present. Alongside UNESCO’s wider competency work for students and teachers, that points towards AI literacy and ethical awareness becoming part of the curriculum itself: recognising bias, understanding limits, and being able to justify responsible use rather than treating AI output as neutral or automatically correct.
In a Grantham classroom, that could translate into tasks where AI use is explicit and inspectable, alongside moments where it is deliberately switched off:
- a geography write-up using an AI summary of a local flooding article, followed by a required paragraph titled “Where the summary might be wrong, and how I checked” (date-stamped sources, local place-names such as the River Witham);
- an English assignment that includes a first draft, an AI-suggested revision, and a 150-word explanation of which changes improved clarity and which dulled the pupil’s own voice;
- a “no-AI” retrieval quiz every Friday to keep core knowledge and sentence control visible without scaffolding.
The tension is not just technical; it is social. The more schools normalise AI in homework, the harder it can feel — for parents, teachers and pupils — to know what is genuinely a student’s own work and what is the tool’s polish. Yet a blanket ban may also feel unrealistic in 2024–2026 classrooms, and may leave pupils less prepared for courses and workplaces where using digital tools thoughtfully is expected.
Questions Grantham can ask about AI and good work
In the schools around Grantham and South Kesteven, the awkward moment is rarely the first draft on screen; it is the quiet second step, when an AI suggestion becomes a real comment in a markbook. At that point, “good work” is being co-produced: teacher judgement, the criteria encoded in curriculum and standards, the design choices of a software team, and—more and more—the way pupils learn to use (or resist) the tool.
Local schools cannot rewrite international frameworks or national procurement, but they do control the day-to-day conditions under which AI is allowed to influence learning. The Cambridge guidance built on UNESCO’s 2024 framework puts teacher rights and student agency at the centre—AI use should support safe, effective and ethical practice, and students should retain the ability to “act and make choices independently”. That framing matters in Lincolnshire precisely because it treats AI as something to be governed in a classroom culture, not just installed on a device.
A short set of questions can keep that culture concrete—especially given the UK Parliamentary POST warnings about fairness, bias and transparency, and the OECD’s call for clearer “guardrails” around AI and assessment:
- For school leaders and staff (2024–2026): What is the tool trained on, and which standards or rubrics does it default to? Where are the criteria visible to staff and pupils when feedback is generated? What is the named route for human override—and how is disagreement recorded when an AI suggestion feels wrong?
- For parents and carers (at a Grantham parents’ evening): Which parts of homework or marking use AI, and which do not? How can a pupil see how feedback was produced (not just the final grade or comment)? What happens if a family challenges an AI-influenced judgement—who reviews it, and on what evidence?
- For pupils (Year 8 to Year 11): What does the teacher say “good work” looks like with AI and without it? What proof of thinking is expected—drafts, prompt notes, corrections, or checks against sources—so the work is more than polished text? When does an AI answer need challenging, and what counts as a reliable way to check it?
A simple takeaway holds across most of these: in an AI classroom, “good work” is not just what scores well—it is work where the standard, the evidence, and the final human decision stay visible enough to be explained and, when needed, contested.