Using AI in a Yoga or Pilates Studio While Keeping Your Voice

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Using AI in a Yoga or Pilates Studio While Keeping Your Voice.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Using AI in a Yoga or Pilates Studio While Keeping Your Voice.

A yoga or Pilates studio keeps its voice with AI by giving the AI its own words first: a one-page voice sheet of the words you use and avoid, how you describe classes and how teachers talk, plus a few examples of your best writing. A person edits every public piece, and teaching advice and personal replies stay human.

Without that groundwork, AI produces the average of all the wellness copy it has ever seen, and the average is very recognisable: "sanctuary", "journey", "nourish your mind, body and soul", "transform your body". Students can't always say why a studio's newsletter feels off, but they notice when it sounds like every other studio's. Your voice lives in specifics, such as the teacher who always says "soft knees", the way you describe your 7am class, or the fact that you never promise anyone a flatter stomach, and AI only knows those specifics if you write them down.

Follow me on Instagram@sagnikteaches

Why AI drafts all sound like the same studio

A chat assistant writes by predicting the most likely next words, based on an enormous amount of text it learned from. Ask it for a yoga class description with no other information and "most likely" means the phrases that appear most often in yoga class descriptions everywhere. It isn't being lazy; it's doing exactly what it was built to do.

Connect on LinkedInSagnik Bhattacharya

Here's what a bare prompt produces. The output is illustrative but typical:

Subscribe on YouTube@codingliquids
PROMPT: Write a class description for our Tuesday evening slow flow
yoga class.

OUTPUT:
Unwind and reconnect with our Slow Flow class, a nourishing journey
for mind, body and soul. Move mindfully through gentle sequences
designed to release tension, build strength and cultivate inner peace.
Suitable for all levels, this sanctuary of calm is the perfect way to
end your day. Leave feeling restored, balanced and renewed.

There's nothing false in it, and nothing true either. It doesn't say what time the class is, who teaches it, how slow "slow" is, whether there are props, or what a student will actually do for 60 minutes. Every sentence could appear on any studio's website, which is exactly the problem.

A voice sheet the AI reads before every draft

A voice sheet is a single page that tells the AI, and new staff, how your studio talks. Keep it short enough that you'd actually update it. Here's an illustrative filled-in one for a small reformer Pilates studio with four teachers:

STUDIO VOICE SHEET (reformer Pilates studio, 6 reformers, 4 teachers)

WHO WE'RE FOR: people who want to get stronger without the gym. Lots
of desk workers, a few runners, several over-60s. Many are nervous
beginners.

HOW WE SOUND: warm, plain, a bit dry. Like the founder at the front
desk, not like a spa. Short sentences. We explain what happens in the
class, not how it will make you feel.

WORDS WE USE: reformer, springs, footwork, "your first few classes",
strong, steady, control, "we'll set the springs for you"
WORDS WE NEVER USE: sculpt, tone, burn, shred, bikini, detox, journey,
sanctuary, "long lean muscles", "all levels" (we say who a class suits)

CLASS LEVELS:
- Foundations: first 6 classes, lots of set-up help, slower pace
- Mixed: comfortable on the reformer, no jumping board
- Strong: jump board, faster transitions, not for beginners

CLAIMS WE DON'T MAKE: no promises about weight loss, posture
"correction", back pain, pelvic floor recovery or any medical outcome.
We say "many students find..." at most, and point anyone with an
injury or condition to talk to a teacher first.

HOW WE SIGN OFF: "See you on the reformer, [teacher name]"

EXAMPLES OF US AT OUR BEST: [paste 3 short pieces: a newsletter
intro, a class description, a reply to a nervous beginner]

The "never use" list does more work than the "use" list. It stops the most common drift in one line, and it's also the easiest part for staff to check. Building a banned-phrase list for your brand covers how to grow it from your own edits.

A yoga studio with a particular lineage needs different entries. A studio rooted in a traditional Hatha or Ashtanga teaching line might add: which Sanskrit names you use and how you spell them (and whether you translate them), how you refer to your teachers' training, and topics you treat with care, such as chanting or philosophy. AI tends to either strip that context out or overdo it with borrowed spiritual language, so spell out the middle ground you want.

Filled in for an illustrative Hatha studio, those extra lines might read: "Pose names: English first, Sanskrit in brackets the first time in a piece, for example downward-facing dog (adho mukha svanasana); no accents or diacritics. We chant a short opening in the Sunday class only and say so in the description, so nobody is surprised. We mention philosophy when a workshop is about it, never as decoration in a timetable notice. We don't sign off with 'namaste'. Teachers' training is described by hours and school, never as 'master' or 'guru'." Five lines, and they stop the two opposite kinds of drift at once.

Store the sheet where every draft starts from it. On business plans, ChatGPT and Claude both offer shared projects, so the whole team's drafts can begin with the same voice sheet and examples. On a solo plan, paste it into custom instructions or keep it as the first thing you paste into a new chat.

The same class, before and after the voice sheet

Run the Tuesday class through again, this time with the voice sheet (a yoga version) and the basic facts about the class. An illustrative result:

PROMPT: Using our voice sheet, write a class description. Facts:
Tuesday 7:30-8:30pm, Slow Flow, taught by [teacher]. Slow pace, about
8-10 poses held for 5-8 breaths each, blocks and bolsters used, 10
minutes of savasana at the end. Suits people who find vinyasa too
fast, and beginners after their first 3 classes. Not heated.

OUTPUT:
Slow Flow, Tuesdays 7:30pm with [teacher]
An hour of yoga at half the speed of our vinyasa classes. [Teacher] takes
you through eight to ten poses, holding each for five to eight
breaths, with blocks and bolsters to make every shape comfortable.
The last ten minutes are savasana, lying down, lights low. It suits
you if our flow classes feel rushed, or once you've done your first
three classes with us. Not heated. Bring water and something warm
for the end.

What you'd still edit: "every shape comfortable" is a slight promise, so "comfortable for more bodies" is more honest; and the teacher might want to add the one thing she always says in that class. But this version tells a student exactly what they're booking, which is what makes them book. The difference didn't come from a cleverer tool. It came from the facts and the voice sheet.

Jobs to hand to AI, and jobs to keep

JobAI's partWhat stays with you
Class and workshop descriptionsFirst draft from your facts and voice sheetFinal wording, and anything about who shouldn't attend
Monthly newsletterStructure, timetable changes, workshop listings, a draft introThe opening paragraph, which students read as the founder's voice
Replies to routine emails (parking, what to bring, class packs)Drafts from your FAQA read before sending
Instagram captionsA week of drafts in one sittingChoosing photos, cutting clichés, tagging teachers
Timetable change noticesClear, complete wordingChecking dates and names
Teacher biosTidying the teacher's own wordsThe teacher approves every word
Injury or health questionsNothing beyond "a teacher will reply"All of it
Messages about grief, pregnancy loss, illnessNothingAll of it, personally
Class sequencingAt most, a brainstormThe teaching

The newsletter row is the one where the split matters most. Here is an illustrative October newsletter opening as the AI drafted it from the voice sheet and a list of the month's news, and as the founder rewrote it:

AI DRAFT:
October is here, and with it a new season of classes. We're
excited to share our updated timetable, a new Foundations course
and our autumn workshop. Read on for everything that's happening
at the studio this month.

FOUNDER'S VERSION:
The heating's back on, which means the Monday 7am crew have stopped
bringing jumpers. Three things this month: Thursday Mixed moves to
6:15pm so fewer of you have to run from the station, a new
Foundations course starts on the 12th, and our one-day workshop on
the jump board has six places left.

The AI draft is tidy and follows the voice sheet: no "journey", no "sanctuary". It still says nothing. The founder's version has a detail only the studio could know and puts the three pieces of news in the second sentence. Let the AI draft the timetable section and workshop listings underneath; the first paragraph is the part students read as you.

Routine email replies have their own trap. A student writes: "Do I need anything special for my first reformer class?" An illustrative draft with the voice sheet loaded: "Just comfortable clothes you can move in and grip socks. If you don't have any, we sell them at the front desk for $12. Arrive 10 minutes early and we'll set the springs for you." It sounds exactly like the studio. But nobody told the AI you sell grip socks or what they cost, and in this illustration the studio lends them free for first classes. When a draft contains a fact you didn't supply, check it before sending, and add the right answer to your FAQ so the next draft has it.

Captions show the same pattern in miniature. Ask for a week of posts with the voice sheet and a list of photos, and the illustrative draft for a photo of the reformer room at 7am comes back as "Rise and shine! Start your day strong with our energising morning reformer class." The edit keeps what's true and swaps the rest for detail: "7am, springs set, coffee after. Monday Mixed has two reformers free this week." Shorter, specific, and useful to someone deciding whether to book.

For captions specifically, writing Instagram captions with AI that sound like you goes further.

Teachers have their own voices too

A studio voice is partly the teachers' voices, and teacher bios are where AI flattens things most. Ask it to "write a bio for our teacher [first name], who has 10 years' experience and loves yin" and you'll get a paragraph that could describe any of 10,000 teachers.

A better method is to start from the teacher speaking. Ask each teacher to record a voice note answering five questions: how they came to teaching, what a first-timer should expect in their class, the cue they say most often, what they do outside the studio, and what they'd say to someone nervous about coming. Transcribe it (most phones and chat assistants can), then ask the AI to shorten it to 90 words using only phrases from the transcript. An illustrative result:

FROM THE TEACHER'S VOICE NOTE (excerpt):
"...honestly I came to yin because I was a terrible runner with
terrible hips. I'll always tell people, it's not a stretching
competition. The cue I say the most is probably 'find the edge,
then back off a bit'..."

BIO DRAFT:
[First name] came to yin as a runner with very tight hips and has
taught it for ten years. In these classes you'll hold poses for three to five
minutes, supported by bolsters, and you'll hear "find the edge, then
back off a bit" more than once. It's not a stretching competition.
Outside the studio, [first name] is still running, badly, and still
stretching afterwards.

The teacher will recognise that as themselves, which is the test. Send every bio to the teacher for approval before it goes live.

Health claims are where studio AI copy goes wrong

Wellness copy online is full of health promises, so AI reproduces them readily. A realistic mistake: a studio asks AI to "write a newsletter section promoting our new Pilates for Back Care course". The draft says the course "realigns your spine, relieves sciatica and fixes years of bad posture". It goes out unedited. Two students with diagnosed back conditions book on the strength of it, and one emails after the first session to say it made things worse and the newsletter promised it would help.

Claims like that invite complaints, can fall foul of advertising rules on health claims, and can put someone with a real condition in the wrong class. Build in three safeguards:

  • List banned claims in the voice sheet: cures, fixes, relieves, corrects, detoxes, and any named condition.
  • Describe what happens in the class, not the outcome: "slow, supported movements for the back and hips, with options for every exercise" rather than "heals back pain".
  • Any course aimed at a condition says clearly who it isn't for and asks students to speak to a teacher, and their own health professional where appropriate, before booking.

Applied to the back care course, the rewrite reads: "Pilates for Back Care: a six-week course of slow, supported reformer work for the back and hips, with options for every exercise and no more than four students per teacher. It suits people who've been told gentle exercise is fine for them. If you have a diagnosed back condition, recent surgery or pain that's getting worse, please talk to [teacher] before booking, and check with your own health professional." Every sentence describes the class or who it's for, and none promises an outcome.

Replies to injury questions ("I've got a herniated disc, which class should I do?") never come from AI. The AI's only job is to route the message to a teacher and tell the student when to expect a reply.

Keeping one voice when several people use AI

In a studio with an owner, a front-desk manager and four teachers, the voice rarely drifts because of the AI. It drifts because each person uses a different tool, a different prompt, or no voice sheet at all. A realistic version: the owner drafts newsletters in a shared project with the voice sheet loaded, while the front-desk manager answers emails using the free chat assistant on her phone. Within a month, students are getting warm, plain newsletters and replies that open with "Thank you for reaching out! We're thrilled you're considering joining our community." Nobody wrote anything wrong. The studio just started speaking in two voices.

Three habits prevent it. Everyone who writes for the studio uses the same shared project or pastes in the same sheet. One person, usually the owner, owns the sheet and is the only one who changes it. And once a month, that person looks at the edits people made to AI drafts. Any phrase that keeps getting cut goes on the "never use" list; any phrase that keeps getting added goes on the "use" list. After a few months the sheet reflects how the studio actually talks, rather than how you described it on day one, and the drafts need fewer edits.

An illustrative first monthly review, from the owner's notes on 23 edited drafts: "nourish" cut six times, "welcome to our community" cut four times, "all levels" replaced three times with the actual level, "invigorating" cut twice. On the other side, "we'll set the springs for you" added to five drafts, and "your first few classes" to three. The first four go on the "never use" list; the last two move up to the top of the "use" list with an example sentence each. Fifteen minutes' work, and the following month's drafts come back needing noticeably less cutting.

One more rule worth writing down: teachers who post on their own accounts are free to write however they like. The voice sheet is for the studio's channels. Trying to make every teacher sound identical removes the variety students come for.

A read-aloud check before anything goes out

The quickest voice test is to read the draft aloud, imagining the founder saying it to a student at the front desk. If a sentence would sound odd spoken, it's probably AI residue. Then run through five checks:

  1. Any word from the "never use" list? Search the draft for them; it takes seconds.
  2. At least one specific detail per paragraph? A time, a teacher's name, a prop, a number of breaths. Generic paragraphs get cut or rewritten.
  3. Any health promise? Remove it or turn it into a description of what happens in the class.
  4. Names, times and Sanskrit spelled correctly? AI is inconsistent with class names and spellings; check against your timetable. Check day-and-date pairs too. A realistic slip in a timetable notice: "From Tuesday 12 October, Thursday Mixed moves to 6:15pm", when the 12th is a Monday. Students who read "Tuesday" start asking whether Tuesday's classes are changing as well. Assistants often get the weekday for a date wrong, so check each one against a calendar.
  5. Would a nervous beginner know what to do next? A link, a class to start with, what to bring.

Studios that do this for a month usually find their edits shrink, because the "never use" list grows and the examples improve. If you want a broader framework beyond the studio, building a brand voice guide that AI can follow is the general version, and turning Pilates intro offers into memberships shows the voice sheet at work on one of the studio's most important messages.

Studio owners also ask

Should I tell students we use AI for our emails and posts?

You don't need to label every newsletter, but don't pretend otherwise if asked, and never present AI-written text as a teacher's personal words. Many studios add a line to their website saying they use AI tools for drafting and that every piece is edited by the team. Personal replies, especially about injuries or life events, should come from a person.

Which AI tool is best for a small studio?

Any mainstream chat assistant will do the drafting. What matters more is keeping your voice sheet and examples in one place the whole team uses, such as a shared project on a business plan or saved custom instructions, so every draft starts from the same rules. Your booking software may also have built-in AI for emails; test it against your voice sheet before using it.

Can AI write class sequences for our teachers?

It can suggest a structure, but sequencing for a real room of students is a teaching skill that depends on who turns up, their injuries and your tradition. Treat any AI sequence as a brainstorm a qualified teacher checks and changes, never as a plan to teach from. Don't use it for individual students with health conditions.

Further reads

Sources: ChatGPT and Claude business plan documentation on shared projects. Example studios, drafts and outputs are illustrative.

Want AI drafts that sound like your studio?

On a 1:1 call we'll build your studio's voice sheet from your best existing writing, set it up in the tool your team uses, and decide which jobs AI drafts and which stay with teachers.

Book a 1:1 call with me