How Tutors Can Plan a Week of Lessons With AI in an Hour

Spend the hour on planning, not on writing materials from scratch. Keep a one-line roster per student, generate every lesson skeleton in a single batch prompt, create new materials only for the two or three lessons that need them, and spend the last 15 minutes checking answer keys. Twelve to fifteen weekly students fits comfortably.

What breaks the hour is asking AI to invent a full worksheet for every student. Most weeks, the right starter or practice set already exists in your own files or a textbook. AI is fastest at sequencing and adapting what you have; generating new questions is slower, because each one needs checking.

Follow me on Instagram@sagnikteaches

What an hour can and can't cover

An hour covers: a skeleton for every lesson (objective, starter aimed at last week's error, main activity, check for understanding, homework), a handful of new question sets, and a note of what to bring. It doesn't cover: designing a new course, writing a full mock paper, or marking. Keep those separate, or the planning hour becomes a three-hour evening again.

Connect on LinkedInSagnik Bhattacharya

Be realistic about the first week. Setting up the roster and the standing instructions takes most of an extra hour on its own, so the first run is closer to 90 minutes or two hours. By the third week, with the roster updated after each lesson and a folder of checked materials to reuse, a tutor with 12 students is usually inside the hour, and often well inside it in weeks where most students are continuing a topic rather than starting a new one.

Subscribe on YouTube@codingliquids

Before the hour: the roster AI plans from

This is the only preparation, and it's mostly copied from your session notes. One row per student, updated after each lesson. An illustrative roster for a maths and science tutor with 12 weekly students might start like this:

Use IDs or first names only. If you use a personal ChatGPT, Claude or Gemini plan, switch off the model-training setting in privacy settings; the business plans don't train on your content by default.

Minutes 0-10: set the week's brief

Open the Project (or Gem, becoming a skill from November 2026) where your standing instructions live and paste the roster. The standing instructions, written once, look something like this:

I'm a private tutor. Lessons are 60 minutes, one-to-one, online.
Every lesson plan uses this structure:
- Objective (one sentence, student-friendly)
- Starter (8 min): targets the "last session" observation directly
- Main (35 min): uses my listed resources first; say which pages
- Check (7 min): 3 questions that prove the objective was met
- Homework (15-20 min of work), from my resources where possible
- Materials needed: list anything I must create, marked NEW
Keep each plan under 150 words. British spelling.
Never write answer keys unless I ask; I'll request them separately.

Keeping answer keys out of the batch step is deliberate. It keeps the batch fast, and it means every key you generate later is one you've asked for and will check.

Minutes 10-25: every lesson skeleton in one pass

Ask for all plans at once: "Plan this week's lessons for every student in the roster." One pass is quicker than twelve chats, and the model keeps the format consistent. For S01, an illustrative output:

S01 – Solving quadratics by factorising. Objective: solve a quadratic by factorising and setting each bracket equal to zero. Starter: show a worked solution that stops at (x + 3)(x − 5) = 0 and ask "are we finished?"; then two where the student completes the last step. Main: textbook ch.9 ex. 9C Q1-12, then Q14-16 (rearrange first). Check: solve x² − x − 12 = 0; x² = 5x; explain why x² = 5x can't be solved by dividing by x. Homework: ex. 9C Q17-24. Materials: NEW – three "unfinished" worked solutions for the starter.

Read through all twelve quickly and fix direction, not wording. Common fixes: a main activity that's too ambitious for 35 minutes, a homework set from the wrong exercise, or a starter that ignores the observation. Two or three corrections in one follow-up message ("S04: no calculators in the starter, build confidence with mental methods first") is quicker than rewriting.

S02 shows the kind of fix that's worth the follow-up. The batch gave this illustrative starter: "Recap the rule: subtracting a negative is the same as adding. Five quick questions, e.g. −3 − (−5), 4 − (−2)." It reads fine and misses the point. The roster says S02 guesses when unsure, and a rule recited without understanding is exactly what a guesser forgets under pressure. The correction message: "S02: no rules in the starter. Use my number line cards: start at −3, show 'subtract −5' as turning round and moving 5 the other way, then two for the student to do on the cards before any written questions." The regenerated starter used the cards, and the rule came at the end of the lesson, once the student had seen why it works.

The batch also copes with a week that changes halfway through. If S01's school announces a test on Friday covering completing the square, don't rerun the whole roster. Send one message: "Re-plan S01 only. School test on Friday: completing the square and the quadratic formula. Keep the starter on setting brackets to zero, shorten the main to 25 minutes, and add 10 minutes of mixed exam-style questions from past paper booklet 2." Five minutes, and the other eleven plans stay as they were.

Minutes 25-45: new materials only where they're needed

Scan the plans for items marked NEW. In a typical week that's three to five small pieces: a set of unfinished worked solutions, a short card sort, six word problems in a context the student cares about. Generate each one with an explicit spec:

Create 6 percentage problems for S04 (adult, preparing for a nursing
course, anxious with calculators). Contexts: dilutions, doses per kg,
percentage increase in a bill. Whole numbers or one decimal place
only. Solvable mentally. Put the answers with full working in a
separate section at the end, so I can check each one.

For S04 that produced six problems such as "A patient weighs 70 kg and the dose is 3 mg per kg. How many milligrams is that?" The tutor swapped one question that needed a calculator (a 17.5% change) for a 15% version and kept the rest.

Two rules keep this stage inside 20 minutes. First, don't regenerate a set because one question is weak; edit that question. Second, ask for working, not just answers, because working is what lets you spot a wrong key in seconds.

Minutes 45-55: the check that stops a wrong answer reaching a student

This is the part people skip, and the part that matters most. AI answer keys are usually right, and occasionally wrong with total confidence. A realistic example: in a factorising set, the key gave the solution to x² + 2x − 15 = 0 as x = 3 and x = 5. The working showed (x + 5)(x − 3) = 0 on one line and then lost the sign on the next. Correct answers: x = −5 and x = 3. A student marking their own homework would have "learned" the wrong result.

A ten-minute check for the week's new material:

1. Work the first and the hardest question yourself, without looking at the key.
2. Follow the key's working line by line for the rest; don't just compare final answers.
3. Check every question is solvable with what the student knows now.
4. Check numbers are sensible in context: no negative lengths, no 140% of a class passing.
5. For science, check units on every line; unit slips are the most common AI error in calculations.

Step 5 is the one S03's set would have failed. An illustrative question from the new moles practice: "How many moles are in 2.5 kg of calcium carbonate (Mr = 100)?" The key's working read "moles = mass ÷ Mr = 2.5 ÷ 100 = 0.025 mol". The formula is right and the arithmetic is right, and the answer is out by a factor of 1,000, because the mass needed converting to 2,500 g first: the correct answer is 25 mol. It's also the exact error in S03's roster row, so a key like that wouldn't just be wrong, it would reinforce the habit the lesson is meant to fix. Checking units line by line caught it in under a minute.

For the longer version of this routine, including reading level and copyright checks on worksheets, see how tutors should check AI-made worksheets.

Minutes 55-60: package the week

Paste each plan into wherever you'll look during the lesson: the student's notes page, your calendar event, or a single weekly document with a heading per student. Attach or link materials. Send homework or pre-reading to students or parents who get it in advance. Then note one thing to change in the standing instructions next week; that small habit is what brings the hour down to 40 minutes over a term.

The note to parents can come out of the same chat. Ask for "a three-sentence message to S01's parent: what we're working on, the homework and how long it takes, and one thing to look for", then add the names yourself. An illustrative result, lightly edited:

Hi [first name], this week [student] and I are finishing quadratics
by factorising. Homework is exercise 9C, questions 17-24, about 20
minutes. If you glance at it, the thing to look for is whether each
bracket gets set to zero at the end - that's the step we're fixing.

Parents rarely want more than that, and the last sentence gives them something specific to notice rather than "please encourage practice".

The weekly change note is worth keeping as a running list, because it shows why the hour shrinks. Three illustrative weeks for the maths and science tutor:

Week 1 (105 min): plans for adults too long -> add "max 3 main tasks
                  for adult learners" to standing instructions
Week 2 (70 min):  homework page numbers from wrong textbook edition ->
                  add edition and ISBN to the resources column
Week 3 (55 min):  starters too similar week to week -> "vary the
                  starter format: error hunt, card sort, mini-quiz"

The same hour for a piano teacher and a language tutor

The routine transfers to other kinds of tutoring, with different inputs.

A piano teacher with 18 weekly pupils. The roster's "resources" column lists each pupil's current pieces and scales, and the observation is musical: "bar 12, rushes the left-hand quavers". The plan's main activity becomes a practice plan: which bars, at what metronome speed, how many repetitions. AI is good at turning "rushes bars 12-16" into a seven-day practice chart that steps the tempo up gradually. It is not good at choosing repertoire, so keep that decision yourself. The time saving here is mostly in the practice charts, which pupils and parents actually use.

An illustrative chart for that pupil, after the teacher's edit:

Bars 12-16, left hand alone, then hands together
Mon  LH alone at 60, 4 times; hands together at 60, 3 times
Tue  same, then hands together at 66, 3 times
Wed  hands together at 66, 5 times
Thu  rest day for this passage; play the whole piece once
Fri  hands together at 72, 3 times; if it rushes, back to 66
Sat  hands together at 72, 5 times
Sun  whole piece once, slowly, before the lesson

The AI's first version climbed from 60 to 96 beats a minute over the week, a jump most pupils can't absorb without the rushing coming straight back. The teacher capped it at 72 and added the "back to 66" rule, which is the part that actually teaches steady quavers. Ask for small tempo steps in the standing instructions and you won't need the edit next time.

A language tutor with adult learners. The observation might be "avoids past tense, switches to present". The plan's starter becomes a two-minute scenario that forces the past tense ("tell me about your last holiday"), and AI can generate a short dialogue at the right level with the target structures highlighted. Ask it to mark any phrase it's unsure is natural, and check dialogues aloud; a model will occasionally produce a sentence that is grammatical but that no native speaker would say.

A tutoring agency can use the same structure centrally: one coordinator maintains the standing instructions and the resource list, and each tutor runs their own hour against their own roster, so plans look consistent across the agency without anyone sharing student notes between tutors.

Choosing a tool: a general assistant or a teacher tool

Free tiers change often and can cap how much you generate, so test your full hour on the free version before paying. If the hour works but hits limits, that's the point to upgrade.

When the hour keeps running over

• Vague observations in the roster. "Needs more practice" gives the model nothing to aim the starter at. Fix the notes, not the prompt.
• Too many NEW items. If more than five lessons need new materials, you're rebuilding resources you should be saving. Store every checked set in a folder by topic and list it in the roster next time.
• Re-prompting in circles. After two attempts at the same plan, write it yourself. It's usually one lesson where you know something the model doesn't.
• Checking left to the end of the evening. If you run out of time, cut new material, not the check.

The first fix is the one that pays back most. Compare two versions of the same roster note. "Needs more practice with fractions" produces a generic worksheet. "Adds the denominators when adding fractions (wrote 1/3 + 1/4 = 2/7); gets it right once shown a bar diagram" produces a starter built around that exact wrong answer and a main activity that starts from diagrams. The second note takes 20 seconds longer to write straight after the lesson, and it saves the re-prompting later.

The session notes you keep for planning double as the raw material for reports, which is where writing parent progress reports with AI picks up. If you're weighing whether any of this justifies a subscription, whether AI is worth paying for in a small tutoring business covers the sums, and Projects versus Gems helps you pick where to keep the standing instructions.

Planning with AI: what tutors ask next

Is it worth paying for a teacher-specific AI tool?

For a private tutor, a general assistant with a saved Project usually covers weekly planning, because the value comes from your roster and notes rather than the tool. Teacher tools such as MagicSchool add ready-made generators for rubrics, quizzes and differentiation; its free plan is enough to test whether those save you time before paying for Plus.

Can AI plan lessons for a student with special educational needs?

It can suggest adaptations if you describe the need and what has worked before, such as shorter tasks or visual supports. It cannot assess the need or replace advice from the student's school or specialist. Keep any diagnosis out of the prompt and describe only the practical adjustments you use.

How do I stop AI lessons feeling generic?

Feed it last session's specific observation for each student and ask for the starter activity to address that exact error. Lessons feel generic when the only input is a topic name. They feel personal when the input is what this student got wrong on Tuesday.

Further reads

• Tutoring Centre Admin With AI: Enquiries, Timetables, Invoices — Enquiries, timetables and invoices, if you run a centre rather than tutor solo.
• How to Set Up Human Review for AI Work Without Slowing Down — Make the answer-key check a habit rather than an afterthought.
• How to Build a Shared Prompt Library for Your Team — Keep your standing instructions and best prompts where you can reuse them.
• Gemini in Google Sheets for Small Businesses: 10 Practical Uses — Handy if your student roster already lives in Google Sheets.
• Can ChatGPT Read PDFs, Spreadsheets and Photos? What Breaks — What happens when you upload past papers and photos of work.
• AI Hallucinations Explained for Business Owners: Causes and Fixes — Why an answer key can be confidently wrong.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: MagicSchool pricing page (checked September 2026); vendor plan pages for ChatGPT, Claude and Gemini. Student rosters, timings and outputs are illustrative.