AI for Acupuncturists: Notes, Reminders and Reviews Without Claims

A solo acupuncturist can use AI for three admin jobs without making health claims: tidying treatment notes you have already written, drafting appointment reminders, and wording review requests. The rule is simple. AI edits your words and your templates, and you decide what gets published, sent or filed.

Most trouble in this trade comes from a fourth job people slip in: asking AI to describe what acupuncture treats. A chatbot will happily write that a session "relieves migraines" or "reduces stress hormones", and your advertising regulator may treat that as a claim you must be able to prove. Keep AI away from outcomes and it is a very good admin assistant.

Follow me on Instagram@sagnikteaches

Where AI helps in an acupuncture clinic and where it must not

An acupuncture clinic is a clinic, not a spa or a wellness lounge. You take a history, record findings, treat, and keep records that may be read by a regulator, an insurer or a solicitor years later. That changes which jobs are safe to share with a chatbot.

Connect on LinkedInSagnik Bhattacharya

The last row matters. You are the practitioner of record. Anything AI suggests about a point prescription, contraindication or red-flag symptom is not evidence, and a note that quietly absorbs a chatbot's guess becomes your clinical statement.

Subscribe on YouTube@sagnikteaches

Free versus paid: what each tier can and cannot be trusted with

The free plans of ChatGPT and Claude cost $0. ChatGPT Plus is $20 a month and Claude Pro is $20 monthly or $17 a month billed annually. ChatGPT Business and Claude Team both need at least two seats, so a one-person clinic would pay for two, or stay on an individual plan. Check the vendor pricing pages before you buy, because plans move.

The practical difference for you is not speed, it is what the vendor does with your chats. Consumer plans let you switch off use of your conversations for model training in the privacy settings, and it is worth doing on day one. Business plans do not train on your content by default. Neither tier is a substitute for a written data agreement if your rules or your insurer require one for patient information.

• Free plan: fine for anything with no patient information in it. Reminder templates, review-request wording, FAQ drafts and form layouts all qualify.
• Individual paid plan: higher usage and upload limits (free plans allow only a limited number of uploads), useful for longer templates. Same data caution as free.
• Business plan: the place to consider anything near patient records, after you have read its data terms and asked your regulator or insurer what they require.

If you are unsure which side of the line a task is on, the confidentiality checklist for AI note tools gives you questions to ask any vendor in ten minutes.

Treatment notes: tidy what you wrote, never invent what you did

The safest note workflow keeps you as the author. You write or dictate quick shorthand during or just after the session. AI turns the shorthand into your clinic's note layout. Nothing in the output should contain a fact you did not supply.

Strip identifiers first

Before any shorthand reaches a chatbot on a free or personal plan, remove the name, date of birth, address, phone, email, employer and any rare detail that would identify the person. Use a patient code such as "Pt 214". Keep the code key in your practice software, not in the chat. Dates can become "visit 3". If stripping feels like it destroys the usefulness of the note, that is the sign you need a business plan with proper terms, or a different tool.

A note-tidying prompt

Save this in a Project so you paste only the shorthand each time. ChatGPT and Claude both include Projects on their free plans (Claude's free plan allows up to five), so this step costs nothing.

You are formatting my own clinical shorthand into a treatment note. Rules:
1. Use only facts I give you. Never add findings, points, advice or outcomes.
2. If something is missing from the template below, write "not recorded".
3. Do not write diagnoses, prognoses or statements about effectiveness.
4. Keep my wording for anything I quoted from the patient.
5. End with a list headed "Questions for the practitioner" for anything unclear.

Template: Presenting complaint (patient's words) | Relevant history changes | Observations I recorded | Treatment given (points, technique, duration) | Advice given | Plan and next visit.

My shorthand: [paste here]

Rule 5 is the useful one. It stops the tool papering over gaps, and the questions it raises are usually gaps in the shorthand rather than the AI.

A before and after, illustrative

Shorthand: "Pt 214 visit 3. Neck stiff mornings, says better than last week, sleeps 6h. Tongue pale, pulse thready (my reading). Needled GB20 bil, LI4 bil, retained 20 min. Advised heat pack, shoulders. Rebook 1 wk."

Output, after the prompt: the six headings filled from that text, "not recorded" under "Relevant history changes", and a question asking whether "better than last week" refers to pain intensity, range of movement or both. That question is the value. A note saying only "improved" is weaker than one that records what the patient said and which measure they meant.

What you would fix by hand: if the tool wrote "neck stiffness resolving", delete it. "Resolving" is a prognosis. If it replaced "thready" with another descriptor, restore your word. Compare each line of output against the shorthand for ten seconds before you save it into the record.

Consent and record-keeping

Three habits keep this defensible.

1. Say it in the consent form. One sentence such as: "I may use computer tools to format my own notes and to write reminders. I remove identifying details first, and I review everything before it is filed." Add it to the intake form and mention it at the first visit.
2. Keep the original shorthand until the final note is signed off. If a regulator asks what you wrote at the time, the shorthand and the final note should match in substance.
3. Never keep the chat as the record. The record lives in your practice software or the paper file. Delete the chat once the note is filed, and check how long the tool retains chats on your plan. Anthropic's privacy centre, for example, says a deleted Claude chat leaves its back-end storage within 30 days, while chats used for model improvement (with that setting switched on) can be kept in de-identified form for up to five years.

If you are weighing a dedicated scribe instead of a general chatbot, read recording versus dictation for therapists first. Recording a consultation raises its own consent questions that dictating your own summary does not.

Appointment reminders that do not drift into advice

Reminders are where AI earns its keep with no patient data at all. You write one template per situation, with merge fields your booking system fills in. AI only polishes tone and trims length.

The four reminder templates worth having

• Confirmation, sent at booking: time, place, arrival advice and how to change.
• 48-hour reminder: time, a one-tap confirm or reschedule link, cancellation window.
• Same-day nudge: a short line with the door code or parking note.
• Missed-visit follow-up: kind, no blame, a rebooking link.

Prompt to draft the set:

Write four short appointment messages for an acupuncture clinic: confirmation, 48-hour reminder, same-day nudge and missed-visit follow-up. Use these merge fields exactly: {first_name}, {time}, {date}, {link}, {clinic_phone}. Each under 45 words, warm and plain. Do not mention symptoms, conditions, treatment outcomes or what the session will do for them. Include the cancellation window: 24 hours. Output as a table with the message type and the text.

The "do not mention symptoms" line is deliberate. Reminders get screenshotted and forwarded. A line like "see you for your sciatica session" is a health detail on a lock screen, and a promise of a result sits right beside it.

A sample output, and what to fix

The 48-hour message might come back as: "Hi {first_name}, a quick reminder of your appointment on {date} at {time}. Please confirm or change it here: {link}. If you need to cancel, we ask for 24 hours' notice. Questions? Call {clinic_phone}." Good. The fix is usually the sign-off: tools add "We look forward to helping you feel your best", which is a soft claim about outcome. Delete it.

For timing, deposit and no-show patterns that carry across appointment businesses, cutting no-shows with reminders and deposits shows the structure. A deposit policy needs to be on your booking page and consent form before it is enforced.

A simple sum

Say you see 24 patients a week and 2 do not turn up. At $70 a session that is $140 a week, around $7,000 over a 50-week year. A reminder sequence that rescues one missed visit a week brings back about $3,500 over the same year. The sequence costs nothing in tools if your booking software already sends messages; the AI only wrote the words once. Figures are illustrative; use your own fee and rate.

Review requests that follow platform rules

Asking patients for a review is fine. The danger is the wording and who you ask. Google's contribution policy, as of November 2026, bars businesses from offering incentives such as discounts or a free session, from selectively asking for positive reviews (review gating), from discouraging negative ones, from asking for specific content to be included, and from pressuring people to review while they are on the premises. That last rule matters in a clinic: handing a patient a tablet at the desk and waiting while they post is the pressure the policy describes. Check the current policy page before you finalise wording.

For a health service there is an additional trap. A patient writing "cured my back pain" in a review on your profile is not your claim, but if you share, pin or reuse it in your marketing, it can become one. Treat reviews as the reviewer's property, ask permission before quoting, and do not repost testimonials that describe outcomes.

A compliant request, drafted with AI

Write a short, friendly message asking a patient to leave a Google review of our acupuncture clinic. Rules: ask everyone the same way, offer nothing in return, do not suggest what to write, do not mention symptoms, conditions or results, and do not ask only people who seem happy. Include the placeholder {review_link} and {first_name}. Under 50 words. Give me three versions.

Sample output, illustrative: "Hi {first_name}, thank you for visiting. If you have a moment, an honest review helps other people find us: {review_link}. No pressure either way." That passes. A version saying "if you were happy with your treatment, we'd love 5 stars" fails: it gates by satisfaction and steers the rating.

Send the same message to every patient after their third visit, for instance, rather than hand-picking. A consistent trigger is also your evidence that you did not gate. For replies to reviews you receive, never confirm that someone is a patient, because that discloses their health information. The tutorial on auto-posting review replies explains why a human should press send.

A claims check you can run on anything AI writes

Run this checklist on every website line, social post and handout. Paste the text into a fresh chat with the prompt below, then read the result yourself.

Act as a strict advertising reviewer for a health service. Below is draft text for an acupuncture clinic. List every sentence that: (a) names a condition, symptom or disease, (b) says or implies a treatment result, cure, relief or prevention, (c) mentions research, studies or "evidence", (d) uses "natural", "safe" or "proven", (e) quotes a testimonial about outcomes. For each flagged sentence, quote it and give a neutral replacement that describes only what the clinic offers or what a session involves. Do not add new claims.

Text: [paste]

Three examples, illustrative:

• Before: "Acupuncture relieves chronic back pain." After: "Initial appointments include a full health history and an examination, followed by a treatment tailored to the patient."
• Before: "Safe, natural fertility support." After: "Appointments are available for patients who are also under the care of their doctor or fertility clinic. Please tell us about your care team when booking."
• Before: "Our patients report less stress." After: delete. Reported outcomes are testimonials in disguise.

The tool is a net, not a judge. It will miss subtle implications, so a person with the clinic's advertising rules in front of them reads the final text. If you are drafting content plans around the same idea, the 30-day dental content plan with claim checks uses the same approach in another health setting.

A week in a solo clinic, with the numbers

Take an illustrative practitioner who sees 22 patients across four days, works alone and does admin on Friday afternoon. Before AI, Friday admin ran close to three hours: 22 notes finished from scribbles (around 5 minutes each, 110 minutes), reminder texts written by hand when someone cancelled (30 minutes), and a review chase she kept forgetting (20 minutes).

After the setup above, the notes step changes shape. She types shorthand between patients, which takes about two minutes, strips identifiers as she types, and pastes the batch into the pinned prompt on Friday. Reading and correcting each output takes roughly three minutes. That is 22 notes at about five minutes in total each, so the notes saving is small, and she should say so honestly. The gain is consistency: every note has the same six headings, gaps are flagged as "not recorded", and a reviewer opening the file sees the same layout in March as in November.

The larger saving is the other two jobs. Reminder templates are sent by the booking system, so the 30 minutes disappears. The review request fires automatically after visit three, so the 20 minutes disappears and, more important, it actually happens. Net: roughly 50 minutes a week back, plus fewer forgotten follow-ups. If a tool promised to halve your note time, treat that with suspicion. Notes are where you carry clinical responsibility, and the saving there should be modest.

A mistake that shows up in real clinics

Here is a realistic failure, described generically. A practitioner asks a chatbot to write a "welcome" page for new patients. The draft includes a sentence that acupuncture "supports the body's natural healing" and a bulleted list of conditions. She likes it, pastes it onto her site and moves on. Six months later a patient screenshots the list when complaining about an unresolved problem, and the practitioner is asked to substantiate every condition on it.

Nothing about the AI was unusual. It produced fluent marketing copy, as it is built to. The gap was the missing review step. The fix is dull and reliable: every public sentence goes through the claims prompt, then a human check against your regulator's advertising guidance, then it is dated and saved in a "published wording" folder so you can show what was live and when.

A one-page rule sheet for you and any assistant

This is a filled-in example. Adapt each line to your clinic and keep it where anyone who touches admin can see it.

• Allowed in a chat: reminder templates, review-request wording, form layouts, my own shorthand with identifiers removed.
• Never in a chat: names, dates of birth, contact details, health history in full, photographs, anything a patient told me in confidence.
• Never from AI: diagnosis, point selection, advice on medication, prognosis, claims about results.
• Sign-off: I read every note, message and public sentence before it leaves my hands. Nothing is scheduled from a chat I have not read.
• Review cycle: first Monday of each quarter, I recheck privacy settings, plan terms and the review policy.
• If something slips: note what went where, delete the chat, tell the patient if their information was involved, and ask my insurer or regulator what else I must do.

Setting up the clinic's AI admin in one afternoon

A realistic first session takes about three hours and costs nothing in tools.

1. Privacy settings (10 minutes). Turn off model-training use of chats on whichever account you use.
2. Consent wording (20 minutes). Add the AI sentence to the intake form and the website privacy page.
3. Reminder set (30 minutes). Generate, edit and load the four templates into your booking system.
4. Review request (20 minutes). Generate, edit, schedule one message for every patient after visit three.
5. Note prompt (45 minutes). Save the prompt, test it on three old notes with identifiers stripped, and compare the output with the originals.
6. Claims check (30 minutes). Run your existing website text through the checker and fix what it flags.

Then write a one-page rule sheet and pin it near the desk: what may go in a chat, what may never go in, who signs off wording. If you hire a receptionist, they follow the sheet from day one.

How to tell it is working, and what to do when it is not

Measure three things over a month. Time per note: time a few before you start, then after. Missed visits: compare the no-show count to the previous month. Review requests sent versus reviews posted: a healthy rate is a fraction, so do not chase a number.

Warning signs worth acting on:

• A note that contains a sentence you cannot trace to your shorthand. Stop, tighten rule 1 in the prompt, and re-check recent notes.
• A reminder that names a condition. Rewrite the template and review what was sent.
• A patient asking why you use AI. This is an opening, not a problem. Answer with the consent sentence and offer to keep their notes fully manual.
• A vendor change. Plans, retention periods and data terms move; recheck them every quarter.

If you want an outside view on which jobs are worth automating first, that is what my AI implementation consultation covers; you can reach it through the contact page. Otherwise the three jobs above are a safe place to start on your own.

Questions acupuncturists ask before trying AI

Can I paste a patient's notes into ChatGPT?

Not into a free or personal account. Free and individual plans may use chats to improve models unless you switch that off, and they are not set up for clinical records. Use fragments with no identifiers, or a business plan whose terms and data agreement your regulator and insurer accept.

Do I need to tell patients I use AI?

If AI touches anything about their care or their data, yes. Put one plain sentence in your consent form and say it aloud at the first visit. If AI only edits your own generic reminder wording, a notice is still good practice but carries less weight.

Can AI write 'helps with back pain' on my website?

It can write it, but you shouldn't publish it unchecked. Condition claims fall under your regulator's advertising rules and need evidence you can cite. Describe the service and what a session involves instead, and run every sentence through the claims check in this tutorial.

Is a free AI tool enough for a solo clinic?

For reminders and review-request wording, yes, because no patient data goes in. For note drafting from dictation, a free tool is the wrong place for identifiable information, so you either strip identifiers or move to a paid plan with proper data terms.

Further reads

• Dental Practice AI Mistakes: Consent, Data, and Over-Automation — Consent, data and over-automation errors another clinic type makes.
• How Physiotherapists Use AI to Write Treatment Notes Faster — How a neighbouring hands-on clinic structures treatment notes.
• Should Therapists Tell Clients They Use AI? — Wording for telling clients AI is part of your admin.
• AI Tools for Physiotherapy Clinics: What Each One Actually Does — What each clinic tool actually does, without the sales copy.
• Can Therapists Use ChatGPT for Session Notes? — Where a general chatbot is and is not suitable for notes.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT pricing pages (individual and business), Claude pricing page, Anthropic privacy centre articles on data retention and model training, Google contribution policy on fake engagement (all checked 4 November 2026).