AI for a Music School: Lesson Notes and Parent Emails Worked Through

Teachers type four or five rough bullets after each lesson, using a student code and never a name. AI turns them into a short parent note, and the same method drafts missed-lesson, practice and recital emails. A teacher edits every message and adds the first name before sending. Consent comes first, and no child data enters consumer AI tools.

Most advice on AI for teachers is written for classrooms: lesson plans for thirty children, marking, worksheets. A music school's week looks different. It is thirty-minute one-to-one lessons, a note home after each, parents who pay by the term and a recital to organise. So the AI never plans the teaching here. It tidies a teacher's own five-line record into a note a parent will read, and the time saved comes from doing that dozens of times a week, not from any single note.

Follow me on Instagram@sagnikteaches

A school of 96 students and the week we will follow

Take a school with four teachers (piano, guitar, violin and voice), 96 students aged about 6 to 16, a weekly 30-minute lesson each, and a Saturday ensemble of 14. The head of the school does the office work. Parents have asked for a note after every lesson, so 96 notes go out a week. Say each takes about seven minutes to write properly, which teachers squeeze in at the end of a long day, so the notes come out late and thin: "good lesson, keep practising scales".

Connect on LinkedInSagnik Bhattacharya

These are illustrative figures with round numbers. This particular week has four jobs: 96 lesson notes, a handful of emails the head writes (missed lessons, practice worries, fee reminders), a recital invitation to 62 families because the recital is in three weeks, and a digest and group note for the Saturday ensemble. Nothing here replaces teaching judgement. The AI turns a teacher's bullet points into readable sentences. It is never asked what a child needs.

Subscribe on YouTube@sagnikteaches

Consent first: the enrolment wording, and when a parent says no

Ask before you start, and ask in plain words. If no child's name, photo or voice goes into an AI tool, you might argue that no personal data is being processed. I would still tell parents, because a note about a coded child's playing can sometimes point back to one child in a small school, and because trust is worth more than the minutes saved. Data-protection law such as the GDPR says children merit specific protection with regard to their personal data, and a one-line addition to the enrolment form costs nothing.

AI tools and lesson notes
Our teachers sometimes use an AI writing assistant to turn their own bullet
points into a clear lesson note or email. We never type your child's name,
photo, voice or any health or family information into it. We use a code. A
teacher reads and edits every message before it reaches you, and the AI makes
no decisions about your child's lessons.

[ ] I am happy for my child's lesson notes to be written this way.
[ ] I would prefer notes to be written without AI tools.

Parent or guardian: ____________   Date: ____________

The second box matters as much as the first. A family that ticks it gets hand-written notes with no penalty, and the register carries a flag: AI = no. Put that flag at the very top of the stub template, so a teacher typing at 7pm can't miss it. Ask everyone again each September and whenever you change tools. Consent given for one assistant, on one set of terms, isn't automatically consent for the next.

Three rules for what goes into any AI tool

Rule one: codes, never names. A code such as P-014 means nothing outside the school's own register, and the register stays in the school's system. First names go in at the last step: the teacher types them into the email after the draft is written, or a mail merge does it. The prompt uses [first name] and [parent name] as placeholders. Keep stubs about the music, not the child. A stub that names the school a child attends, or a rare combination such as the only tuba player in the under-9 group, can identify a child even with a code attached.

Rule two: no recordings. Don't record or transcribe children's lessons or performances with a consumer AI tool. A child's voice is personal data, a recording can capture a sibling or a parent in the corridor, and the saving is small because a lesson note needs only four or five facts. The alternative is cheap: after the child has left, the teacher types or dictates their own bullet points.

Rule three: welfare never goes through a prompt. If a child says something worrying, arrives upset, or a parent mentions a health condition or a family change, that goes to your designated safeguarding lead under your own policy and into your own record. It doesn't go into a chat window, even with a code, and it never appears in a parent email drafted by AI. The same goes for additional needs and anything medical: write what you need in your own words, in your own system.

Which plan: five seats, and what each default means

ChatGPT's pricing page lists model training as "opt-out available" on its Free, Go, Plus and Pro plans, and Claude's pricing page lists "opt-out" for Free, Pro and Max. So on a personal plan your chats can be used to improve the models unless you switch that setting off. It works if every teacher switches it off and keeps it off, which is a lot to ask of five people. Business plans reverse the default: Claude's pricing page says Team plans carry no model training on your content by default, and OpenAI says the same of ChatGPT Business. Both need at least two seats.

For this school, with four teachers and the head, five seats at the annual list price of $20 per seat per month comes to $100 a month, or $125 on monthly billing at $25 a seat. Those are the vendors' list figures in October 2026, so confirm them for your account. A business plan lowers the risk. It doesn't make a child's name suitable to paste, and I would keep the codes rule even there. If you want to see how dedicated music-school software compares, which AI features in music school software are real checks the claims, and what a solo teacher's admin stack costs covers the one-person version of this decision.

Set-up: the stub, the standing instructions and the school's policies

Create one Project for the school in ChatGPT or Claude, or one per teacher, and put three things in it: the voice rules, the stub template and the policy text (make-ups, fees, recital dates). Student data never goes in. The stub is the real discipline. It forces a teacher to say what happened in the lesson in five short lines, and it gives the AI nothing to guess.

AI OK: yes / no        (if no, write this note by hand)
Code:
Instrument and level:
Worked on:
Went well:
Needs work:
Practice task:
Next lesson:
Do not mention:        (anything the parent should hear from me directly)

The standing instructions turn that stub into something a parent will want to read. They set the voice, and more importantly they shut doors: no invented praise, no exam talk, no softening of weaknesses.

You write lesson notes and parent emails for a small music school.
Voice: warm, plain and specific, in the school's usual spelling. No exclamation marks.
A lesson note is under 120 words.

Use only the facts in the stub. If something is not in the stub, leave it out.
Never add praise, exam talk, predictions, comparisons with other students,
or reasons you were not given.
Report weaknesses plainly and kindly. Never turn a weakness into praise.

Refer to the student as [first name] and the parent as [parent name].
Never guess a name.
If the stub contains what looks like a real name, a surname, a school or a
phone number, do not write the note. Say what you found and stop.

End lesson notes with the practice task and what happens next lesson.

That last block of rules is a second line of defence, not a control. A model can miss a name, so the real control is the teacher's habit and the sentence you added to the enrolment form. Test the instructions before you rely on them: paste one stub with a made-up surname in it and check that the assistant refuses.

Monday: six piano lessons, six stubs, six notes

The piano teacher has six lessons after school with a two-minute changeover between them. The stub gets typed on a phone during that gap, so the notes are drafted from what just happened instead of from memory at 8pm. Here is one stub, for a student working on early Grade 3 pieces.

AI OK: yes
Code: P-014
Instrument and level: piano, early Grade 3 pieces
Worked on: Minuet in G (page one, hands together); C and G major scales
Went well: left hand steadier; kept the beat through bar 8
Needs work: right-hand fingering bars 9 to 12; scales too fast and uneven
Practice task: bars 9 to 12 hands separately, 5 minutes a day; scales slowly, metronome at 60
Next lesson: start page two if bars 9 to 12 are secure
Do not mention: nothing

With the standing instructions in place, the draft that comes back is close to usable. This is illustrative output, and the teacher adds the first name before sending:

This week [first name] worked on the first page of Minuet in G with both hands together. The left hand is steadier, and the beat held all the way through bar 8. The right-hand fingering in bars 9 to 12 needs more work, and the C and G major scales are still going too fast to stay even. For practice: bars 9 to 12 with hands separate, five minutes a day, and the scales slowly with a metronome set to 60. If those bars feel secure, we will start page two next lesson.

Compare it with what a loose prompt produces. Without the rules, asking for "a friendly note for a parent" gives something like this (also illustrative):

What a fantastic lesson! [first name] is well on the way to Grade 3 and should be very proud of the progress on scales. A little more work on the right hand in the minuet and we'll be there. Keep it up!

Count the problems. "Fantastic" and "proud" say nothing a parent can use. "Well on the way to Grade 3" turns a level into an exam promise. "Progress on scales" turns a written weakness (too fast and uneven) into praise, which is the commonest failure of all: the teacher's criticism disappears and the parent never hears it. "We'll be there" promises an outcome. And the practice task, the one thing a parent can act on, has vanished. Each of those became a line in the standing instructions, and the first read of every draft is a check for exactly these.

The timing works like this. A stub takes about a minute and a half, the AI takes seconds, and reading and adjusting the note takes another minute and a half. Sending takes half a minute. That is about 3.5 minutes a note against the 7 in the example, and the note is more specific than a tired end-of-day one, because it was written while the lesson was fresh.

Tuesday: a missed lesson and a make-up slot

At 2pm a parent texts the school to say the 4pm lesson can't happen. The head does not paste that text into an AI tool. Messages like this often carry the child's full name and a reason, and an illness is health information. Instead the head types a stub:

Code: V-031
Missed: Tuesday 4:00 pm lesson; notice given 2 hours before
Policy: under 24 hours' notice means no make-up unless the teacher has a free slot
Free slot: Thursday 5:30 pm with the teacher
Reason given: not recorded
Tone: warm, brief

The reply, as an illustrative draft:

Hi [parent name], thanks for letting us know that [first name] can't make today's lesson. Because it was less than 24 hours' notice it falls outside our make-up policy, but [teacher] has a free slot on Thursday at 5:30pm if that suits. Reply to confirm and we will hold it. We look forward to seeing [first name] soon.

Left to itself, the model can end that reply with "We hope [first name] is feeling better soon". The stub says the reason wasn't recorded, so the line assumes illness, and if the real reason was a school trip or a clash with football it lands badly. The rule is that sympathy appears only when the stub states the facts that justify it. The policy sentence is accurate because the policy text lives in the Project: the first time you set this up, ask the assistant to quote your make-up policy back, and compare it with the real wording.

Wednesday: the practice email nobody wants to write

A guitar student, in their first term, hasn't practised for three lessons running. The teacher is worried about sounding like a telling-off. The stub states the goal as well as the facts, which is what stops the AI improvising a tone.

Code: G-008
Instrument and level: guitar, first term
Situation: no practice reported for three lessons running; quick and keen in the lesson
Goal of the email: agree a small routine, without blame
Options to offer: fixed 10-minute slot three evenings a week; or two 5-minute sessions on busy days
Do not mention: other students, reasons for missed practice

An illustrative draft, before the teacher's edit:

Hi [parent name], I wanted to share something about [first name]'s guitar. For the last three lessons we have started from roughly the same place, because there hasn't been time to practise between sessions. [first name] picks things up quickly and enjoys the lesson, so a little regular practice would show fast. Would a fixed ten-minute slot, three evenings a week, be realistic? If the week is busy, two five-minute sessions still help. Let me know what would work for your family and I will plan the next few lessons around it.

Without the standing instructions, a draft can open with "I'm concerned about [first name]'s commitment", add "Children often lose motivation at this age" (a generic claim that isn't in the stub), and say that most students in the group practise, which the model has no way of knowing. All three come out in the edit. Then the teacher reads the email aloud once, with one test: would I say this to the parent's face? Anything that fails that test is rewritten by hand. A practice email is a relationship message, and I would never send one unread.

Thursday: a recital invitation and eight fee reminders

These two jobs need no student data in the AI tool at all, which makes them the safest place to start. The recital invitation is one email to 62 families. The AI writes it once from the facts (date, venue, arrival time, guest numbers, dress, how to reply), with [first name] and [parent name] left as merge fields that the school's email tool fills in. The fee reminder is the same idea: one gentle template with [amount], [due date] and [payment link], and the AI never sees which families are late. A person pulls the list of eight from the billing system.

Subject: [first name]'s recital on [date]

Hi [parent name], the school recital is on [day and date] at [time], at [venue]. Please arrive by [arrival time]. [first name] will play [piece], and each family can bring [number] guests. Please let us know by [reply date] whether you can come, using [reply link]. We suggest [dress]. If [first name] is nervous about performing, please tell [teacher] so we can talk it through. [school name]

Hi [parent name], a gentle reminder that term fees of [amount] for [first name] were due on [due date]. You can pay at [payment link]. If you have already paid, thank you, and please ignore this note. If there is a problem, reply to this email and we will find a way through it together. [school name]

Watch the fee reminder closely. A draft can add "A late fee of $10 applies after seven days". If the school has no such fee, an invented penalty in writing is worse than no reminder. Rule: money terms appear only if you pasted them into the Project's policy text. The recital invitation has a similar trap. Before you merge, send three test emails to staff addresses, including one for a family with two children, and read each greeting. Many "AI mistakes" turn out to be merge mistakes. Planning a dance recital has a running order and parent-email checklist that carries over to music.

Friday and Saturday: the ensemble digest and the group note

The Saturday ensemble is the easy case, because the content is about the group and the music. On Friday the teacher pastes the rehearsal plan (bars 17 to 32 of the piece, sectional work for the lower strings, arrive 15 minutes early with a tuned instrument) and gets a four-line digest for the 14 families. On Saturday evening the teacher pastes three bullets about what the group achieved and gets a short group note. The one thing to supply yourself is every logistical fact: times, rooms, what to bring. Left to fill a gap, the model will invent a plausible room number.

Sunday: twenty minutes checking the week

The week ends with a review. It is new work and it takes time, so it appears in the sums below. The head or a lead teacher does the following.

1. Pick six sent notes at random and compare each with its stub, line by line. Look for praise that isn't in the stub, weaknesses that have softened and promises that nobody made.
2. Scan the week's chats for any student's real name, using the chat search if your tool has one. A hit means a name slipped in, and the rule needs repeating.
3. Check that no note went out for a student flagged AI = no.
4. Add one line to the standing instructions for anything you caught this week.

If you find yourself sending every draft untouched, slow down and compare two of them with their stubs. Drafts this tidy don't prove the rules are working. They can mean nobody is reading closely.

The week in numbers

Here are the illustrative figures for the whole school. Teacher time is the main saving. The head's emails matter less in minutes but a lot more in dread.

The school gets back about 6.6 hours a week, roughly 28 hours a month, spread across five people. Set-up costs about three hours for the head (consent wording, the stub, the standing instructions and the policy text) and twenty minutes per teacher to learn the stub. Against that, five business seats are $100 to $125 a month. If an hour of staff time is worth even $15 to the school, 28 hours is about $420 a month, which is a comfortable margin. The bigger gain is better notes that go out on the same day. Treat every figure here as a starting estimate, and replace it with your own after two weeks of timing real notes.

Four failures to plan for

These are illustrations of what goes wrong in practice, not reports from a particular school.

1. A name slips in. A busy teacher pastes a parent's text, complete with the child's full name and school, to get a quick reply. On a personal plan that content may be used for training unless the setting is off. The response: delete the chat, check the setting, tell the head, and ask your data-protection adviser whether a slip like this needs recording. The prevention is a rule that nobody pastes a parent's message. A person summarises it into a stub.
2. Details drift between stubs. Six stubs pasted into one request can swap details, with a piece title from one student's note landing in another's. One stub per request avoids it, at the cost of a few more clicks.
3. An exam promise. Suppose a note says a student is "ready for the June exam" when no exam has been entered. A parent may reasonably act on that. The standing instructions ban exam talk unless the stub contains it, and the teacher's read-through checks for it.
4. The merge, not the model. Two siblings share a surname and the greeting in one email names the wrong child. That is a mail-merge setting, not an AI error, and the three-test-emails habit catches it.

Jobs I would keep away from AI in a music school

Anything about a child's welfare stays out, for the reasons in rule three. Parent complaints stay out too: the facts may be disputed, the wording carries weight, and a person should own the reply. Exam entries and results are a human job, with no predictions in either direction. Notes about a student's additional needs or health conditions belong in your own records, written by the teacher. Fee hardship is a conversation, not a template. And never put photos or recordings of children into an AI tool. For a wider look at children's data in a similar setting, a one-hour AI briefing on children's data for nursery staff is a good model for a short staff session.

Questions music teachers ask about AI and children's data

Can I record lessons and have AI transcribe them?

I wouldn't, for children's lessons. A child's voice is personal data, a recording can capture other people too, and the saving is small because a lesson note needs only four or five facts. Dictate your own bullet points after the child has left. Recordings belong with a business-grade tool, a data-processing agreement and written parent consent.

Do I have to tell parents that AI helped write the notes?

Tell them. Put a short line on the enrolment form and repeat it in the first note of term: a teacher writes each note with an AI writing assistant, using a code instead of your child's name, and checks it before sending. Rules on explaining how you handle personal data vary by place, so ask an adviser.

What if a parent says no to AI?

Flag the student in your register and write their notes by hand, with no penalty and no awkwardness. Put the flag at the top of the stub template so a busy teacher can't miss it. Ask every family again each September, because families change their minds and your tools change too.

Is a student code enough to protect a child's identity?

No. A code removes the obvious identifier, but a stub that mentions the only tuba player in the under-9 group can still point to one child. Keep stubs about the music, not the child's circumstances, and treat the code list itself as personal data that stays in the school's own system.

Will every AI-written note sound the same?

It can, and parents notice. The cure sits in the stub, not the prompt: name the piece, the bars and the task. If two notes read alike, the stubs were alike. Add one detail that only that lesson had, such as the bar where the pulse slipped, and the note carries it through.

Further reads

• How Nurseries Use AI to Write Daily Reports for Parents — The same stub-to-report method for younger children.
• Can AI Write Swim Lesson Progress Reports? A Level-by-Level Template — Level-by-level progress reports for another lesson-based school.
• How Tutoring Centres Use AI to Write Parent Progress Reports — Parent progress reports in a tutoring setting.
• Can AI Draft a Martial Arts School Waiver for Children? — Drafting consent wording for children's activities.
• AI for Driving Instructors: Lesson Notes and Progress Summaries — Lesson notes and reminders for one-to-one instruction.
• Martial Arts School Software: Gradings and Parent Updates Compared — Compare martial arts school software on belt tracking and parent updates, then see where AI-written messages help and what data must stay out.
• AI Based Marketing Tools for a Music Shop: Lessons and Instrument Hire — A music-shop shortlist of scheduling, email and design tools for filling lesson slots and promoting instrument hire across one school term.
• Dance Studio Handbook: Parent Policies Drafted With AI — Draft a dance studio parent handbook with AI: a decision sheet, sections for attendance, costumes and behaviour, and a promise audit before sign-off.
• Martial Arts Summer Camp Ideas: Sign-Ups and Parent Emails With AI — A four-week camp plan for a martial arts school: 20-place pricing, daily timetable, safeguarding and pick-up rules, a sign-up page and six parent emails.
• Music Store Marketing Ideas: Lessons, Repairs and Used Gear, With AI — Nine ideas that fill lesson slots, workshop capacity and used-gear stock, each with a ChatGPT prompt, a watch-out and a monthly email that ties them together.
• AI Receptionist Kiosk or Phone AI: What a Walk-In Business Needs — Phone AI or a counter screen? A four-week tally, a music school worked through, list prices and the failure modes to test before you pay for either.
• AI for Dog Trainers: Session Notes and Homework Plans for Clients — A session-note template, a voice-memo prompt and a homework sheet owners will follow, with rules against diagnosis and for keeping client data safe.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT and Claude pricing pages (seat minimums, training defaults); the text of GDPR Recital 38; OpenAI and Anthropic business-plan terms as stated on their own pages (checked October 2026). The school and its figures are an illustration.