How an English Language School Cuts Lesson Prep and Reports with Claude Code

It was 10 p.m. on a Friday, and the handout for tomorrow’s lesson was still a blank page.

I know a teacher who runs a small English conversation school on his own. Fifteen students, and almost every one of them is at a different level with a different goal. His days are wall-to-wall lessons, so the only time he has to build materials is after class. On top of that, the progress reports he owes parents keep piling up. “I love teaching English,” he told me, “but the stuff that isn’t English is going to bury me.”

If you run an English school, you already know this trap: you spend more time preparing to teach than actually teaching. In this post I’ll walk through how much of that I was able to cut with Claude Code and generative AI, with everything I actually tested by hand.

Key takeaways

• Most of the workload at an English school isn’t the lesson itself. It’s building materials and writing reports.
• When I handed those two jobs to Claude Code, prep for one lesson shrank from about 30 minutes to roughly 10.
• Delegate the drafting, the rough cut, and the formatting. Level assessment and the final check stay with the teacher, always.
• Never feed a student’s real name or contact details to the AI. Swap them for an alias or a symbol before anything goes in.
• I’ve included copy-paste prompts and a runnable script that turns lesson notes into report templates in one pass.

Where the work actually piles up at an English school

Let me be clear about who this is for. I’m writing for the owner or teacher of a small school with roughly 10 to 50 students. You juggle one-on-one sessions and small group classes, you have no admin staff (or maybe one person), and you handle materials and parent communication yourself, in between everything else.

Lay out the workflow of an English school and it looks something like this:

1. Run a trial lesson and interview the student about their level and goals.
2. Prepare materials and handouts matched to that level, every single time.
3. Teach the lesson.
4. Record what happened and send a report to the parent or the student.
5. At month-end, summarize progress and plan the next month.

Steps 2 and 4 are the time thieves. The lesson itself (step 3) is the fun part, and it’s the part only a human can do. But if you write steps 2 and 4 from scratch every time, prep and follow-up cost 40 to 60 minutes per lesson. A teacher running 20 lessons a week loses well over ten hours to that alone.

The rework and headaches teachers keep hitting

Here are the “yep, that’s me” moments my friend described:

• Over-built a handout, then on the day it didn’t match the student’s level and half of it went unused.
• Rebuilt a near-identical vocabulary sheet every week instead of reusing past work.
• Put off reports until they stacked up, then wrote the whole month’s batch through gritted teeth.
• Changed the tone of every message per parent, spending 15 minutes on each one.
• Lost handwritten notes in a pile and couldn’t remember what a lesson covered three months ago.

The common thread is “writing from scratch every time” and “never reusing what you’ve already built.” That’s exactly the territory where generative AI shines, because rough drafts are its strong suit.

Use case 1: Draft a leveled handout in 10 minutes

The first use is drafting handouts. Tell the AI the student’s level and a theme, and it produces the whole skeleton at once: a vocabulary list, example sentences, practice problems, even a model conversation.

What I tried was a handout for a CEFR A2 student (roughly someone who has finished beginner-level grammar) on the theme of “ordering at a cafe.” Here’s the prompt I wrote:

You are a veteran English-conversation teacher. Draft a lesson handout under the following conditions.

# Student profile
- Level: CEFR A2 (finished beginner grammar)
- Age group: working adults, 20s to 40s
- Today's theme: ordering at a cafe

# What I want you to produce
1. 10 vocabulary words to learn today (a table with the word, a simple pronunciation guide, and the meaning)
2. One example sentence for each word
3. A model ordering conversation (staff and customer, about 6 exchanges)
4. 5 fill-in-the-blank practice questions (answers in a separate section)
5. 3 free-conversation prompts

# Constraints
- Avoid jargon and tricky phrasing; use expressions an A2 learner can say comfortably
- Add one note about a pronunciation or intonation point learners often stumble on

The draft that came back was about 80% usable as-is. The vocabulary table was clean, and the model conversation sounded natural. The remaining 20% was just teacher tweaks, like “this student learned ‘would like’ last time, so let’s work it in.” A 30-minute build from scratch became a 10-minute adjustment.

Claude Code makes this even easier. Keep your past handouts in the school’s folder, tell it “don’t repeat vocabulary from the previous handouts,” and it actually reads those files and avoids the overlap. That kind of folder-level work is something a one-off chat AI can’t do, and it’s where Claude Code is strong. If the tool itself feels intimidating, skim Claude Code for non-engineers first and you’ll trip up less.

Use case 2: Templatize lesson reports and produce them in bulk

The second job is lesson reports, the recap you send a parent or the student after class. Written from scratch every time, one report easily eats 15 minutes.

The trick is to hand over bullet-point notes and let the AI do only the clean write-up. The messy notes you jot during a lesson are enough:

- Student: K (alias)
- Today: past-tense questions, talking about a travel memory
- Went well: regular past-tense verbs were smooth
- Needs work: irregular verbs like go -> went are shaky
- Next time: review with irregular-verb cards
- Homework: write 3 sentences about the weekend

Attach those notes to the prompt below and they turn into a polished report for parents:

Turn the following lesson notes into a report to send to a parent.

# Tone
- Warm and encouraging. Lead with what went well; pair any weak point with a concrete suggestion
- No jargon; words a parent who isn't fluent in English will understand
- About 150 to 200 words, ending with a short word of encouragement for next time

# Lesson notes
(paste the notes above here)

Lock the tone to “warm and encouraging” and the time you used to spend agonizing over wording for each parent disappears. I cover this templatizing mindset more in advanced prompt engineering for Claude Code, and English-school reports are a textbook case of a task that benefits from a fixed template.

What to delegate to AI, and what you decide yourself

Get this line blurry and you’ll have an accident. Here’s where I draw it:

Task	Hand to AI	Human must decide
Drafting handout vocabulary and examples	Yes, produce the rough cut	Final check that the level actually fits
Generating practice questions	Yes, write questions and answers	Eyeball whether the answers are correct
Writing up reports	Yes, turn notes into prose	Confirm nothing misstates the facts
Assessing a student’s level	Only as a reference opinion	Teacher makes the final call
Promotion or class changes	No	Teacher/owner decides
Sending to parents	No	A human always reads it before it goes out

The rule is simple. Let AI “write” and “tidy up”; you keep “decide” and “send.” Generative AI will cheerfully slip in a plausible-sounding falsehood, so every answer and every fact gets a human read. Hold that line and you get the speed of the rough draft without the risk.

Security and privacy notes

An English school handles real names, contact details, payment info, and sometimes information about children. Tread carefully here.

• Don’t type a student’s real name, address, phone number, or email into the AI. Replace them with an alias or symbol, like “K” or “Student A.”
• Be especially careful with information about children. Don’t share photos or a specific school name or address.
• After generating a report, a human always reads it and fixes any wrong proper nouns or facts before sending.
• Document a working rule for the school. Put on a single page what’s okay to enter and what isn’t.

Sticking to aliases alone avoids most of the risk. For setting per-project rules, CLAUDE.md best practices is a useful reference. Write “use aliases for any personal data” right into the school’s folder, and Claude Code will work with that policy in mind.

Copy-paste: a runnable script that bulk-builds report templates

When notes pile up, pasting them one by one is tedious. So I wrote a script that reads notes as JSON and generates report templates in one pass. It runs with Node.js. It doesn’t use an Anthropic API key. It’s a template version meant to first confirm that the structure flows in cleanly.

// report-template.mjs
// Bulk-generate report templates from lesson notes (JSON)
import { readFile, writeFile, mkdir } from "node:fs/promises";

// Always keep student info as aliases (never the real name)
const notes = [
  { alias: "K", topic: "past-tense questions", good: "regular verbs were smooth", issue: "irregular verbs are shaky", next: "review with irregular-verb cards" },
  { alias: "M", topic: "giving directions", good: "turn left/right is solid", issue: "choosing the right preposition", next: "practice with a map" },
];

function buildReport(n) {
  return [
    `[${n.alias} - Lesson Report]`,
    ``,
    `Today we worked on "${n.topic}".`,
    `What went well: ${n.good}. They stayed calm and focused.`,
    `Next focus: ${n.issue}. Next time we'll practice together with ${n.next}.`,
    ``,
    `Keep up the great work. Looking forward to the next lesson!`,
  ].join("\n");
}

await mkdir("./reports", { recursive: true });
for (const n of notes) {
  const body = buildReport(n);
  await writeFile(`./reports/report-${n.alias}.txt`, body, "utf8");
  console.log(`Generated: reports/report-${n.alias}.txt`);
}
console.log(`Wrote ${notes.length} templates.`);

Running it is just this:

node report-template.mjs

A per-student template lands in the reports folder. From there, hand the text to Claude Code and ask it to “make the tone warmer,” and the clean write-up follows in one go. Lock the structure with code first, then have AI flesh out the prose. That order is the secret to reports that never fall apart. If you want to go faster still, see Claude Code productivity tips.

If the CEFR levels feel fuzzy, take one look at the British Council guide to CEFR levels and you won’t hesitate when telling the AI what difficulty to target.

A rough ROI estimate

Let’s put numbers on it. These are ballpark figures.

• Handout: 30 minutes the old way, down to 10 with an AI draft plus tweaks. About 20 minutes saved per lesson.
• One report: 15 minutes the old way, down to 5 with notes plus a clean write-up. About 10 minutes saved per report.
• A teacher running 20 lessons and 20 reports a week: (20 min x 20) + (10 min x 20) = about 10 hours saved per week.

At a notional hourly rate of $25, that’s roughly $250 of time freed up each week. You can pour that time back into filling trial lessons or caring for individual students. Getting time back to actually teach is, I think, the biggest payoff of all.

FAQ

Q. Can I hand AI-made materials straight to a student? A. As a rough draft they’re excellent, but I don’t recommend using them untouched. You’ll find answer mistakes and phrasing that doesn’t fit that particular student’s level. A teacher should always read and adjust them first.

Q. Won’t the English example sentences sound unnatural? A. Recent models are quite good at natural English. That said, you’ll see a mix of spoken and written registers, and regional differences (US vs. UK). For anything that bothers you, add a follow-up like “make this more casual and spoken” and it cleans up.

Q. Can I use this if I’m not great with computers? A. If all you do is copy and paste a prompt, you need no special skills. If setting up Claude Code itself feels daunting, start with the Claude Code getting-started guide and you’ll be fine.

Q. I’m nervous about entering student information. A. Don’t enter real names or contact details. Swap in aliases or symbols and you can run both report write-ups and material creation with no problem. Putting your input rules on a single page makes it safer.

Q. Does this work for a solo teacher and for a whole school? A. If you’re an individual learner or a side-gig teacher, start from the free PDF and learning materials. If you’re folding this into a school’s operations, start from a conversation about how to design the rollout. See the CTA below.

If you want to build this into your school’s operations for real

You can try every prompt and script above with your own hands first. If you want to learn and use it as an individual, the free PDF and learning materials are the easy place to start.

On the other hand, if you run a school with multiple teachers and you want to “standardize how we build materials and reports as a system” or “tighten up how we handle personal data,” designing it beats winging it. For a school-wide rollout or teacher training and operational consulting, reach out through training and consulting.

What I confirmed when I tested this

To close, here’s an honest account of what I checked by hand.

For the handout, I actually generated the A2 “ordering at a cafe” theme. It produced the full set: vocabulary table, model conversation, fill-in-the-blank questions. About 20% needed touching up. It was clearly faster than writing from scratch. That said, one of the fill-in-the-blank answers came back slightly off, which drove home that the eyeball check is non-negotiable.

For reports, I actually ran the script above and confirmed it output two aliased templates into the reports folder. Once the structure is locked by code, the variation in write-ups disappears. I also tested the alias approach and confirmed a report comes out complete without ever entering a real name.

The conclusion: a lot of the “everything but teaching” time at an English school really can be cut with generative AI and Claude Code. But what you can cut is the drafting and the formatting. The parts where a teacher’s skill shows, level assessment and the final check, stay firmly in place. And that’s exactly why you can take the time you saved on the machine work and give it back to the students in front of you.