Mark Zolotar · Beyond English
Updated：2026-09-30
In an earlier project, I used a program I had written to generate more than 5,000 pages of curriculum material. That number is impressive in exactly the way a very large pile of paper is impressive. It tells you how much there is. It tells you rather less about whether anyone should learn from it.
The project covered subjects across Grades 1–6. My aim was practical: assemble material I could examine, adapt, and use, including to support my son’s learning. I also adapted some material for teaching. I described the process in an earlier manuscript; this is a reflection on the design and editorial work, rather than a current software installation guide.
The program changed the scale of drafting
I organised prompts in a spreadsheet and sent batches through a program. That let me generate many related sections without manually conducting a separate conversation for each one. The gain was in the mechanics: issuing requests, collecting responses, and turning them into documents.
It did not establish that every topic was appropriate, every explanation was correct, or every sequence was suitable for a particular child. Those are educational judgments. A program that makes documents quickly does not automatically make those judgments well.
The first research step needed scrutiny
I initially asked AI to suggest countries with strong reputations in different subjects, then used those suggestions to structure further work. That was a convenient starting point, but it should not be mistaken for a validated international curriculum comparison. An AI-generated ranking is a lead to investigate. Official curriculum documents, assessment evidence, and the needs of the learner still need checking.
If I were designing a repeatable workflow around this project, I would require a source record for curriculum decisions before expanding the writing. Otherwise, a questionable assumption made in the first prompt can be reproduced with remarkable efficiency. Computers are wonderfully obliging in that respect.
Generation left an editorial job behind
My original account describes removing weak sections, replacing others, adding missing material, changing formatting, and asking for revised drafts. That work took substantially longer than generating the text. I also sought supplementary material, adapted worksheets, and checked them before use.
There is a useful distinction here. A curriculum draft proposes what could be taught and in what order. A lesson still needs an achievable objective, an explanation, examples, practice, a way to assess understanding, and a response when the learner does not understand. The learner is inconveniently real. The document cannot negotiate that inconvenience on its own.
A workflow I would recommend
- Define the learner and outcome. Specify what a learner should be able to do, not just the topic name.
- Check the source. Keep curriculum decisions traceable to material you have actually reviewed.
- Generate a small sample. Examine one unit before multiplying the format across a whole programme.
- Review for teaching. Check correctness, prerequisites, examples, task difficulty, and whether the assessment matches the objective.
- Try it with a learner. Observe where the explanation fails and revise the material.
- Expand what works. Keep versions and record why important changes were made.
This is a recommended process, not a claim that the entire original collection has passed an independent evaluation.
What the page count does—and does not—prove
The project showed me that the practical cost of producing a first draft had changed dramatically. It did not demonstrate that 5,000 pages produced 5,000 pages of educational value. Nor did it establish a measured learning outcome that I can honestly attribute to the software.
The valuable question is what a teacher can do with the time saved. In my case, the answer includes planning, checking, adapting, explaining, and working with the learner. I want tools that make those activities more achievable. I have considerably less interest in winning a competition for the longest document.
About this account
Adapted from my supplied manuscript, How I Generated 5,000+ Pages of Curricula Using AI. Historical project details are presented as my own account. The workflow recommendations are editorial reflections; no independent effectiveness study is claimed. This edition was prepared with AI-assisted editing.
Assessing advanced listening: turn evidence into the next lesson · More writing by Mark Zolotar · School and university work at Beyond