Teachomatic what to automate, and what not to

A Prompt Library for Your Subject

The time goes on re-explaining. Every request starts by supplying the same context — the year group, the specification, the level, the format you want, the things you never want. Most people retype it every time, get an average result because they left something out, and conclude the tool is mediocre.

A prompt library fixes that, and it is not a technical undertaking. It is a text file with ten paragraphs in it.

For a workplace-oriented comparison outside education, this guide offers another way to look at measurement, workload, behaviour, or accountability.

Why this and not prompting technique

The advice circulating about clever phrasing has a short shelf life and is mostly marginal. What actually determines output quality is how much of your situation you supplied, and that is a knowledge problem, not a phrasing problem.

Your context does not change between requests. Your year group, your exam board, your class's reading level, your subject's conventions, your house style for worksheets, the things you always have to correct. Writing it down once and reusing it is the whole technique.

For an external perspective on lesson planning, teacher workload, and classroom technology, see OpenAI.

Which means the library is really a written record of your teaching context — the same knowledge that decides what does and does not hand over — and building it is useful even if you never use a tool again.

The ten that earn their place

Not a definitive list. The shape to aim for.

Levelled text. Rewrite this at [reading age], keeping all subject terminology and all connectives, same content, same register for a [age] reader. The failure modes to check are here.

Practice questions. Fifteen questions on [topic] for [year], in the style of these two examples, ranging from recall to application, with question 12 the hardest.

Misconception hunt. What are the five most common misconceptions [year] students hold about [topic], and what wrong answer does each produce? Genuinely useful for a topic you have not taught before.

Explanation, five ways. Explain [concept] for a [age] student five different ways, using different analogies, no more than 80 words each.

Feedback frame. Against these criteria, identify the single change that would most improve this piece, phrased as an instruction the student can act on. One change, not five.

Parent email. Your standard framing, tone and closing, ready to fill.

Rewrite for audience. Rewrite this for parents / for students / for the leadership team.

Worked example. Produce a worked example of [problem type] showing every step, with the reasoning stated at each step.

Model answer plus flawed answer. A strong response and a plausible weaker one, so students can compare. The flawed one is the teaching material.

Meeting notes to actions. Who does what by when.

The block to put at the top

The part that does most of the work. One paragraph, reused with every request:

I teach [subject] to [year groups] following [specification]. My students' reading ages range from [x] to [y]. Assume no prior knowledge of [topic area] beyond [what they have covered]. Keep all subject terminology. Use [British/American] spelling. Do not use bullet points unless I ask. Do not add encouraging commentary. If you are unsure of a fact, say so rather than guessing.

That last sentence is worth including every time. It does not make fabrication impossible — nothing does, and it will still happen — but it makes hedging more likely where hedging is warranted, and hedged output is easier to check.

Building it

Start with three, not ten. The three things you do most often.

Add one whenever you type the same context twice. That is the signal.

Keep it somewhere you will actually open — a note on your phone, a pinned document, a text file. Not a folder structure.

Write down what did not work too. A short "things that produce bad output" list saves you rediscovering them each term.

Share it with your department. The single biggest saving available. One person writes the levelled-text prompt properly and everyone stops reinventing it — the same argument as a shared comment bank, and just as underused.

What a library will not fix

It will not tell you what the difficulty is for your class. It will not choose the example. It will not know which student needs which version.

A library removes the friction around the mechanical work. It does not touch the judgement, and a very good prompt applied to a badly-conceived task produces a polished version of the wrong thing.

The short version