Most UK teachers are already using AI for lesson planning — but 2 years in, the output is still inconsistent. A 2025 National Literacy Trust survey of nearly 3,000 teachers found 58% use AI regularly, with lesson resources the most common task. The frustration isn't adoption. It's that the same prompt produces brilliant output one day and something barely usable the next. Here's why that happens — and how to fix it.
The real problem isn't ChatGPT
If you've used it for a while, you've noticed: the quality of what you get back varies enormously. Same topic, same rough request, different days — and the output can range from genuinely useful to something you wouldn't hand to a supply teacher.
That's not a bug in the model. It's a briefing problem.
ChatGPT — and any AI tool, Gemini and Claude included — is a statistical engine. It generates output based on the context you give it. Give it vague context, it fills in gaps with averages. Give it specific context, it has less to guess at. The variance drops. The output gets sharper.
Most teachers who've been using it for a year or two have learned this intuitively — but haven't fully systematised it. This is what systematic looks like.
What structured briefing actually involves
Before you write the prompt, have four things ready:
- The specific learning objective — not "water cycle" but "students can explain evaporation and condensation in their own words and give a real-world example of each"
- The class context — year group, ability range, anything that affects how they learn ("28 students, mixed ability, 4 with EAL, they respond well to kinaesthetic tasks, tend to struggle with extended writing")
- The format you need — full lesson plan, just the starter, a differentiation extension, a question set, a rubric
- What you don't want — this is the one most people miss
That last point matters. "Don't include group work — this class functions better independently. Don't suggest worksheets — I don't have colour printing." Negative constraints cut out a whole category of generic filler.
A prompt with all four components:
"Year 5 science, 60 minutes. Learning objective: students can explain evaporation and condensation in their own words and give a real-world example of each. Class is 28 students, mixed ability, 4 with EAL — they respond well to visual anchors and kinaesthetic tasks. I need: a 10-minute starter, 35-minute main activity, 10-minute plenary, 5-minute exit ticket. Include a higher-ability extension. No worksheets, no group work."
That takes 90 seconds to write. The output variance drops significantly.
Where experienced users still leave time on the table
If you're already using it regularly, the gain isn't in the lesson plan itself — it's in the surrounding tasks that eat time without requiring your expertise.
Differentiated materials. Paste your main task and ask it to rewrite at two reading levels — one simpler, one with an extension challenge. You review and adjust. Faster than building three versions from scratch.
Question sets. Give it the topic, the year group, the format (multiple choice, short answer, extended), and ask for 12. You pick the 8 that fit. The generation takes seconds; the curation is yours.
Exit ticket variations. Ask for three different formats for the same learning objective. Pick the one that fits the lesson's mood.
Marking rubrics. Describe the task and the year group and ask for a simple rubric. Edit from there — much faster than building criteria cold.
Parent-facing summaries. Ask it to summarise what the class is working on this term in plain English, under 150 words, for a non-specialist audience. Useful for newsletters, parents' evening prep, communication home.
The consistency fix: prompt templates
The teachers getting the most consistent output aren't writing new prompts every time. They've built a small library of prompt templates — one for lesson plans, one for differentiation, one for question sets — and they reuse and refine them.
After you get output you're genuinely happy with, save the prompt that produced it. That's your template. Next time, you swap in the new topic and objective. The structure does the work.
Three iterations and you have a prompt that fits your class, your style, and your school's format requirements. The first lesson plan takes 20 minutes. The tenth takes 5.
The underlying skill
What separates the teachers getting consistent, high-quality output from those still finding it hit-and-miss isn't technical knowledge of the tool. It's the ability to write a precise brief — context, objective, format, constraints.
That skill transfers. Assessment feedback. Parent communications. Revision materials. Reporting. Anywhere you need professional written output without starting from a blank page every time.
AgentTongue teaches structured prompting from the ground up — built around real tasks, not abstract theory. These techniques work with any AI tool: ChatGPT, Gemini, Claude. The briefing principle is the same. Unit 1 is free, no account needed. The full course is £39.
Start Unit 1 free at AgentTongue.com