AI for Naturopaths: Intake Summaries and Follow-Up Emails

Yes, a naturopath can use ChatGPT or Claude to summarise intake forms and draft follow-up emails, but only with identifying details removed or under a vendor agreement that covers health data. Keep names, contact details, diagnoses tied to a person, and any treatment or supplement advice out of the prompt.

The job AI does well here is turning a three-page form into a one-page briefing for you, and turning your visit notes into a tidy email. The job it must never do is decide what a client takes or does. Hold that line in three places: in the prompt, in what you paste, and in how you read the output, because a model will cross it unasked.

Follow me on Instagram@sagnikteaches

What the intake summary is for, and what it is not

A naturopathic intake form is long. It usually covers history, current symptoms, medicines, sleep, diet, stress, family history and goals. Reading it properly before a first consultation takes ten to fifteen minutes. A summary does not replace that reading. It gives you a one-page map so that you walk into the room knowing which sections to look at first.

Connect on LinkedInSagnik Bhattacharya

Think of the output as a briefing sheet with four parts: what the client said they came for, what they listed under medicines and allergies (copied, not interpreted), gaps in the form, and questions you may want to ask. Nothing in it should be a conclusion about cause, and nothing should name a product or protocol. That restriction is deliberate. A language model will happily produce plausible-sounding suggestions, and a plausible suggestion about a supplement in a draft that you skim at 8am is how a claim ends up in an email you never meant to send.

Subscribe on YouTube@sagnikteaches

The signed intake form and your own consultation notes remain the record. If a regulator, an insurer or the client asks what you knew and when, you point to those, not to a summary a tool generated. Keep the summary only as long as you need it for the visit, and decide in advance where it lives. A good default is that it sits in the same system as the form and is deleted when the file is closed.

The data rules to settle before you paste anything

Before any prompt, answer three questions. Which tool is it? What agreement sits behind it? What are you putting in?

Consumer chat plans (free and individual paid tiers) are built for general use; on Claude's individual plans, for instance, model training is a setting you opt out of. Business plans such as ChatGPT Business and Claude Team do not train on your content by default, which is a real improvement, but not training is not the same as having a health-data agreement. Whether your clients' records fall under health-privacy law such as HIPAA turns on whether you are a covered entity. For a provider, that depends on whether you send health information electronically for standard transactions such as insurance claims, so a practice that never bills insurers electronically may sit outside it. If you are covered, any vendor handling those records for you needs a signed business associate agreement, and the everyday business chat plans are not offered with one: Anthropic lists its HIPAA-ready offering on Enterprise only, not Team.

One shortcut you may have heard about does not apply to you. OpenAI's free clinician workspace, which lets an individual sign a business associate agreement in its settings, is currently open only to physicians, nurse practitioners, physician assistants and pharmacists, so naturopaths are not on its list. If you have clients in the EU, the GDPR treats health data as a special category, so you need a lawful basis, a clear privacy notice and a processor agreement with the vendor. Whether your practice is covered by either law is a question for your professional body, insurer or a qualified adviser, not one to settle from a web page.

The practical rule that works in nearly every case: redact first. Strip the name, date of birth, address, phone, email, employer, and any detail that could identify a person in a small town. Replace the name with "Client A". Then you are summarising a de-identified form, and the question of agreements matters much less. Redaction is not perfect, since rare conditions and unusual life stories can identify someone, so keep the stricter rule for unusual cases: if the story is distinctive, summarise it yourself.

If you want the full version of this thinking for any health practice, the patient data and AI confidentiality checklist goes through retention, access and vendor questions in more depth than fits here.

One intake form, from redaction to briefing sheet

Say a naturopathic practice with one practitioner and a part-time receptionist sees four new clients a week. The intake form is three pages. Reading and annotating it takes twelve minutes, so four forms cost about 48 minutes a week before the consultations even start. The practitioner decides to trial a redacted summary for two weeks.

The receptionist redacts the form in the booking system by copying the free-text answers into a blank document and deleting the name, contact and employer lines. Below is the illustrative redacted input.

Client A, late 40s. Reason for visit: "tired all the time, poor sleep, bloating after meals, want to feel like myself again."
Sleep: asleep by 11pm, wakes 3-4am most nights, up at 6:30.
Energy: worst 2-4pm. Coffee: 4 cups a day.
Digestion: bloating most days; bowel movements daily; no blood noted.
Stress: "high, busy job and two teenagers".
Medicines: levothyroxine 75 mcg daily (since 2019); ibuprofen occasionally.
Allergies: penicillin (rash).
Previous practitioners: GP, last blood test "about a year ago".
Goals: more energy, fewer headaches, better sleep.
Form left blank: family history, alcohol, exercise.

The practitioner's prompt asks for a briefing and explicitly rules out interpretation.

You are helping a naturopath prepare for a first consultation. Below is a de-identified intake form.
Produce a one-page briefing with exactly these headings:
1. Why the client came (their words, condensed)
2. Medicines and allergies (copy exactly as written; do not comment on them)
3. Patterns the client mentioned (list only what they wrote; no causes, no diagnoses)
4. Blank or unclear sections
5. Five open questions to ask in the consultation
Do not suggest supplements, diets, tests, treatments or diagnoses. If the form is unclear, say so instead of guessing.

[paste the de-identified form]

An illustrative output looks like this: heading 1 says the client wants more energy, better sleep and relief from bloating. Heading 2 copies "levothyroxine 75 mcg daily (since 2019); ibuprofen occasionally; penicillin allergy (rash)". Heading 3 lists wakes at 3-4am, afternoon energy dip, four coffees a day, bloating most days, high stress. Heading 4 lists family history, alcohol and exercise as blank. Heading 5 offers questions such as "When did the night waking begin, and did anything change around then?" and "What does a typical day of eating and drinking look like?".

What would you fix? Look at the third heading. A weaker model may slip in a line like "possibly related to caffeine intake". That is an inference the form does not support, so delete it, and tighten the prompt with "list only what is written". Also check that the medicines line is verbatim. If the summary writes "thyroid medicine" instead of the named drug and dose, it has paraphrased something you must have exact. Time cost after the first week: about three minutes to redact, one minute to run the prompt, three minutes to check the summary against the form. Seven minutes against the twelve minutes before, and the briefing sheet is better organised.

Follow-up emails that stay on your side of the line

A follow-up email after a consultation has four honest jobs: thank the client, restate what was agreed in the room, list the practical next steps (booking, forms, tests you ordered through the usual route), and say when you will next be in touch. All of those you can draft with AI. What the email recommends, you write.

The workflow that keeps this safe is to give the AI a list of facts you chose, not the whole consultation. You type the bullets from your own notes, using "Client A" and no clinical detail beyond what the email truly needs, and the AI turns them into a warm message. You then add the personal details at the sending stage.

Write a short, warm follow-up email from a naturopath to a client after a first consultation.
Facts to include (use only these):
- Thanks for coming in today.
- We agreed a food-and-sleep diary for 10 days; a template is attached.
- Follow-up appointment: 3 weeks from today, 45 minutes; booking link is [link].
- Bring the diary and any questions.
- Call the practice on [number] if anything worries you before then; for anything urgent contact a doctor or emergency services.
Do not mention supplements, diagnoses or treatments. British English. Under 150 words. Sign off as [your name].

Illustrative output: a friendly 120-word email with the facts in a short list and a sign-off. Check three things before sending: that the "urgent" line is intact (models sometimes soften it), that no recommendation has crept in, and that the placeholders were filled in with the right client. A placeholder left as "[first name]" in a sent email is a common, embarrassing failure, and a missing-name email to a health client lands worse than a typo elsewhere.

Three more messages worth templating

Once the first email works, the same method suits other routine messages. A reminder before a follow-up appointment: facts are date, time, length, what to bring. A diary-nudge halfway through the diary period: one sentence of encouragement, no coaching on content. A "we haven't seen you for six months" note: warm, no pressure, a booking link. Each of these runs from a fixed fact list that you paste in, and none needs health information beyond a first name.

Before and after: rewriting a message that overreaches

The riskiest sentence in any wellness email is the one that sounds helpful. Here is a realistic before and after, with the "before" being what a general-purpose tool produces when asked for a friendly follow-up without limits.

Before (illustrative AI draft): "It was lovely to meet you. Based on what you told me about your tiredness and bloating, I'd suggest cutting back on coffee and trying a magnesium supplement in the evening, which can support sleep and calm digestion."

After (edited): "It was lovely to meet you. As we agreed, please keep the food-and-sleep diary for ten days and bring it to your next appointment so we can go through it together."

The edit deletes a product recommendation, a causal claim and an invitation to self-treat, none of which the practitioner decided in that form. Keep a simple check: if a sentence says "can support", "helps with", "will improve" or names a product, it comes out unless you wrote it deliberately. Your professional body and the advertising rules where you work decide what you can say about outcomes, and this is a question for them. Related reading on claim limits is in AI marketing for chiropractors: claim limits, which covers the same trap in public marketing.

Setting up the routine for the whole practice

A routine that depends on one careful person is fragile. Write it down so the receptionist, a locum and you all do the same thing.

1. Decide the tool and the agreement. Name one tool, note which plan, and record whether it has a health-data agreement. If it does not, the rule is redact-only.
2. Write the redaction rules on one page. Use the table above. Add "when in doubt, summarise by hand".
3. Save the two prompts in a shared note or a Project. ChatGPT Projects and Claude Projects let you keep standing instructions so nobody retypes the "no supplements" line. ChatGPT can share a project on every plan, Free included; Claude offers projects on every plan (a Free account holds up to five) but shares them only on Team and Enterprise, so on an individual Claude plan each person keeps their own copy of the instructions.
4. Name the reviewer. One person reads every summary against the form before the consultation and every email before it is sent. That is you for the first month.
5. Set retention. Summaries are deleted when the file is closed. Chat history in the tool is cleared on a schedule you decide, because an old chat is also stored data.
6. Tell clients. Add a paragraph to your privacy notice and intake paperwork: what AI is used for (briefing sheets and admin emails), what is never entered (identifying details), and that a person reviews everything.

The client-facing wording can be short: "We sometimes use an AI writing tool to help prepare administrative summaries and emails. We remove your name and contact details first, a practitioner reviews everything, and AI is never used to decide your care." Have your adviser or professional body look at it before you rely on it. If you are weighing how and whether to tell clients, should therapists tell clients they use AI works through the reasoning for a neighbouring profession.

A twenty-form tally: is the briefing sheet earning its place?

For the first twenty intake forms, keep a tally on a sheet of paper next to the desk. Each time you review a summary, mark one of three outcomes: accurate, needed a fix, or unsafe to use. Count what you corrected: invented causes, paraphrased medicine names, skipped blank sections. After twenty forms you will know your real error rate and which prompt line to tighten.

Say seven of twenty summaries needed a fix and three of those were an invented cause. That is a prompt problem. Add "do not infer causes" in capitals and try again on the next ten. If the problem persists, change tools or go back to manual reading for complex cases. A tool that adds causes under a firm instruction is not saving you time, because you have become its proofreader for the one error that matters most.

Also track time. Write down minutes spent on a form before and after. If the saving is under four minutes a form, the redaction step is probably the bottleneck, and you may be better off asking the client to complete a shorter form that is already easy to summarise by eye. Not every practice benefits equally: if you see two new clients a month, the setup effort is not worth it, and your time is better spent on the follow-up email template alone.

Five ways a redacted-intake routine slips

A few realistic failure modes, each with how it shows up.

Merged clients. You paste two redacted forms in one chat to save time, and the summary of the second quietly inherits the sleep pattern of the first. Fix: one chat per client, and a new chat each time.

Confident gaps. A blank "family history" field produces a briefing line saying "no significant family history". Blank is not the same as none. The prompt line "list blank sections as blank" fixes it, and your check should look for it.

Voice-memo shortcuts. It is tempting to dictate consultation notes and have a tool turn them into an email. A recording of a consultation is identifiable health data. If you want that, you need a scribe product with a health-data agreement. See the questions in AI scribe security: audio, retention and BAA vendor questions before you trial one.

Over-friendly tone. Models drift towards cheerful certainty. Add one sentence to your standing prompt: "Calm, plain and professional; no exclamation marks; no promises of results."

Autocomplete in email apps. Built-in AI writing helpers in webmail see the client's real name and message thread. Turn them off for client emails, or check what the provider does with that content, rather than assuming it is covered by your chat tool's settings.

What stays out of AI entirely

• Anything that recommends, doses or names a supplement, herb or protocol for a person.
• Interpretation of lab results, imaging or medicine interactions.
• Diagnostic reasoning, including "what could this be?" prompts.
• Safeguarding concerns, risk assessments and anything involving a minor's disclosure.
• The signed record itself. Summaries support it, they never replace it.
• Clients' photos, identity documents, and lab PDFs.

If you are unsure whether something belongs on that list, it does. For a broader look at which jobs suit AI in practices like yours, AI tools for dietitians: admin, content and follow-up by job splits the work in a similar way, and the clinic data and consent checklist is worth reading if you ever add a chatbot.

A short weekly rhythm that fits a solo practice

For a solo practitioner, the routine need not be heavy. On Monday morning, the receptionist redacts the week's new forms and saves them as separate documents. You run the briefing prompt for each, one chat each, and read the output next to the original while the kettle boils. After each consultation, you type five bullets of facts for the follow-up email and run the email prompt. On Friday, spend five minutes tallying the week's corrections and deleting the chats. That is the whole system, and it fits in under half an hour on a four-client week.

If, after a month, it saves less than it costs, stop. The most useful outcome of a trial is knowing which two prompts are worth keeping, not owning a long list of them.

Questions naturopaths ask before trying this

Can I paste a full intake form into ChatGPT?

Not into a standard consumer or business chat plan unless you hold a signed data-processing or business associate agreement that covers it. Safer routes: redact first, or use a tool your regulator and insurer accept, then check the vendor's agreement terms before any real record goes near it.

Should follow-up emails mention supplements or protocols?

Leave them to you. Let AI draft the structure, appointment details and homework reminders, then add any recommendation yourself in your own words, within your scope of practice and your professional body's advertising and claims rules.

Do I need to tell clients I use AI?

Tell them. A line in your intake paperwork and privacy notice saying what AI is used for, what is never entered, and that you review every output is simple and honest. Data-protection rules in many places also require you to describe how data is processed.

What if the AI summary gets something wrong?

Treat the summary as a working draft. Always read the source form beside it, and never file the summary as the record. The signed intake form and your own notes remain the clinical record.

Further reads

• AI Note-Taking Tools for Therapists: A Confidentiality Checklist — A confidentiality checklist for any note-taking tool.
• Telling Patients You Use AI Note-Taking: Consent Wording That Works — Consent wording that works when AI touches client information.
• AI Charting for Dietitians: Drafting ADIME Notes With a Review Step — How a review step sits around AI-drafted charting.
• Nutritionist AI Prompt: Write Consent and Cancellation Wording — Consent and cancellation wording for a wellness practice.
• Personal Trainer Intake Forms: Build One With AI, Protect Health Data — Designing an intake form that collects less health data.
• AI Marketing Automation for Doctors: Recall, Reminder and Review Flows — Recall and reminder flows once your follow-ups work.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI ChatGPT Business pricing page and help centre articles on ChatGPT for Clinicians and Projects; Claude pricing and plan comparison page; the official HIPAA covered-entity guidance (archived copy). Check your own regulator's guidance.