The Prompt Is the New Lesson Plan

Watch two teachers sit down with the same AI tool to plan the same topic, and you will witness something revealing.

The first types: "Create a lesson plan on fractions." Out comes a lesson that could belong to any classroom on earth, which is another way of saying it belongs to none. The second types something longer: the form level, the class size, the misconception her students carried out of last week's lesson, the fact that her strongest student finishes everything in ten minutes and her weakest cannot yet see that two quarters make a half, the forty minutes she actually has after assembly eats the first ten. Out comes something startlingly close to teachable.

Same tool. Same topic. The difference is not typing skill. The difference is that the second teacher knows exactly what she is trying to achieve, for whom, and why. Her prompt is not a request. It is a compressed lesson plan, and the AI is merely decompressing it.

This is the thesis I want to offer the profession: the prompt is the new lesson plan. Not because it replaces planning, but because it exposes it. A prompt is pedagogical thinking made visible, and its quality mirrors, with uncomfortable precision, the clarity of the thinking behind it. Show me your prompt, and I will show you your pedagogy.

Consider what actually goes into a good brief. To tell the AI what kind of activity you want, you must know why that kind of activity works: whether this concept calls for direct instruction or guided discovery, whether your students need retrieval practice or elaboration, whether the cognitive load of the task fits the working memory of a Tuesday afternoon. These are not technology skills. They are learning theory and pedagogical design, the supposedly dusty content of teacher education courses, suddenly cashed out as the difference between a usable lesson and a generic one. Research on how teachers actually prompt these systems confirms the pattern: the quality and specificity of teachers' task-specific prompting practices vary widely, and outputs vary with them (Moorhouse et al. 2025). A teacher who cannot articulate what makes learning happen cannot brief a machine to support it. The risk is not that such a teacher gets no lesson. The risk is worse: they get a plausible lesson, full of familiar-looking activities, that produces no significant improvement in students, replicated endlessly because it looks like teaching. You cannot brief what you do not understand.

Here is the second, deeper reason the teacher remains the author. The AI does not know your students. It cannot. Their backgrounds, their home situations, the quiet feud between two boys in the third row, which student blossoms when called to the front and which one shrinks, who just lost a grandfather, who reads two levels below and hides it well: none of this exists anywhere in the machine's knowledge, and most of it exists nowhere in writing at all. It lives in one place, the teacher's head. This is precisely why experienced teachers make planning decisions that look arbitrary to outsiders but never are: every choice of grouping, pacing, example, and analogy is based on the actual situation of actual students. It is also why a brilliant lesson from one classroom so often falls flat when transplanted into another. Lessons are not recipes; they are fitted garments. The scholars who gave us the TPACK framework recognised this when they revisited it for the generative AI era, placing contextual knowledge at the centre: the teacher's knowledge of their specific students, school, and community is exactly the knowledge that no general-purpose AI possesses, and exactly what teachers must inject into every exchange with it (Mishra et al. 2023). When a teacher writes a rich prompt, they are not feeding the machine instructions. They are feeding it the context it is permanently blind to.

Which brings us to the proper division of labour. In an earlier column I explained what these systems fundamentally do: they predict the most plausible continuation of text, based on statistical patterns across the writing of millions of strangers (McCoy et al. 2024). They do not know your syllabus is different this year. They do not know anything, in the sense that matters. This is why I have taken to describing AI to my trainee teachers with two words: brainless brilliant. Brilliant, because the fluency, speed, and range are genuinely astonishing, and pretending otherwise fools no one under forty. Brainless, because there is no understanding underneath, no intention, no knowledge of children, no stake in whether your students learn. A brainless brilliance is a magnificent thing to have on your planning desk and a catastrophic thing to put in charge. The AI is here for brainstorming, drafting, and producing learning artefacts: the worksheet, the quiz items, the reading passage at three difficulty levels, the analogies you had not thought of. It is not here to be the sage we consult for what good teaching looks like. The ship still needs the teacher's brain at the wheel, and studies of teacher education confirm the healthy pattern: when pre-service teachers evaluate AI-generated lesson plans, it is precisely their content and pedagogical knowledge that lets them see the gaps, the missing scaffolds, the assessments that measure nothing, and fix them (van den Berg & du Plessis 2023).

Some will hear all this and conclude that AI makes teacher expertise less necessary, since the machine now drafts what novices once struggled to write. The truth runs exactly the other way, and it is worth stating carefully because it will define the next decade of our profession. Working with a system whose inner workings you cannot inspect demands more judgement, not less. Education researchers describe this as the new pedagogical challenge of our time: learning, and teaching students, to work with the black box, exercising evaluative judgement over outputs whose origins we cannot see (Bearman & Ajjawi 2023). Evaluative judgement about lessons is built from exactly the things we sometimes treat as optional garnish in teacher development: mastery of one's subject, fluency in learning theories, command of pedagogical methods, an understanding of curriculum design. The teacher who holds these can co-author with the machine and get the best of both. The teacher who lacks them is not co-authoring at all. They are transcribing, and their students will feel the difference long before any inspector does.

So no, the lesson plan is not dying. It is migrating into a new form, written conversationally, at speed, in dialogue with a brainless brilliant assistant. What that form demands of us is the oldest requirement in the book: teachers who deeply understand learning, their subject, and, above all, the particular children in front of them, because that understanding is the one ingredient the machine will never supply.

The prompt is the new lesson plan. Which means the pedagogy behind it has never mattered more.