Using ChatGPT for Physical Therapy: Referring-Doctor Updates and Outreach

Use ChatGPT to write the template, not the patient update itself. Ask it for referring-doctor messages with square-bracket placeholders (visits attended, goal, next step), then fill the brackets from your own records. Outreach to new referrers needs no patient detail at all. Either way, a clinician reads and sends every message.

That split is the whole method. The chatbot is good at tone, structure and consistency, and a clinic that writes referrer updates well gets noticed. The patient's details are the part it never needs, so they never go in. The sections below build the template library, show the outreach emails, and set out the checks that stop confident invention.

Follow me on Instagram@sagnikteaches

Four messages worth templating

Referring doctors hear from a clinic in a few predictable moments. Each one deserves its own template, because each answers a different question for the person reading it.

Connect on LinkedInSagnik Bhattacharya

Only the first three are patient-specific, and for those the chatbot writes the shell. This is also where an EHR merge field or a saved text snippet earns its keep: once the template is approved, a receptionist or clinician can insert the details in a minute without touching AI again.

Subscribe on YouTube@sagnikteaches

A prompt for a progress-update template

Give ChatGPT the structure you want, your tone and the length, and tell it explicitly that no real details will appear. Paste in two of your own past updates (with all patient details removed) so it learns your voice.

I run a physical therapy clinic. Write a progress update email template to a referring doctor about a patient at around visit 5. Use square-bracket placeholders for anything specific: [patient initials only], [referral reason in general terms], [visits attended of planned], [main functional goal], [progress against goal], [next step], [anything we need from you]. Do not invent any outcome, number or clinical detail. Keep it 120 to 160 words, warm but efficient, plain English, written in first person from the treating therapist. Put the single most important point in the first sentence. Here are two examples of my usual tone: [paste].

Illustrative output:

Subject: [Patient initials]: progress at visit [x] of [y]

Dear Dr [surname],

[Patient initials] is making [good / steady / slower than expected] progress towards [main functional goal]. [Progress against goal in one sentence.] They have attended [x] of [y] planned visits.

Our plan for the next [number] weeks is [next step]. If you are happy with this, no action is needed from you. [If relevant: We would be grateful if you could [anything we need from you] by [date].]

Please call me on [direct number] if you would like to discuss.

Kind regards,
[Your name], [role]

What you edit: ChatGPT often adds a closing line such as "Thank you for the opportunity to care for your patient", which sounds like a form. Remove it. Change "good progress" wherever you cannot back it with a measure, and replace any vague phrase with the measure itself: "walking tolerance up from 5 to 15 minutes". That kind of figure comes from your record, which is why it goes in after the template is written.

Main example: a clinic with 14 referrers and a quiet month

Say an eight-clinician orthopaedic and sports clinic receives about 60 new referrals a month (illustrative). A tally from last quarter shows 14 referring doctors. Three of them send 36 of the 60, six send 19, and five send about one a month each. Two practices that used to send regularly have not sent anything for ten weeks.

A weekly 30-minute routine, built around templates:

1. Same-day acknowledgement for every new referral using the "referral received" template, filled by the front desk from the system (2 minutes each).
2. Progress updates at visit 5 for every patient referred by a doctor in the top nine, drafted from the template by the therapist (3 to 4 minutes each).
3. Discharge notes at the end of care, one per referral.
4. One outreach email a week to a quiet referrer, written from the introduction template and personalised by the clinic lead.

The tally above contains no patient data: it is a count by referring practice. You can paste that table into ChatGPT to rank who to contact first, as long as you use practice labels, not people's names, unless your policy treats referrer contact details as safe to share with the tool. For the top three referrers, a short in-person visit every quarter will beat any email, and the templates simply keep the day-to-day communication reliable so that you have time for it.

Outreach to a referrer who does not know you yet

This is the one message where ChatGPT can do nearly all the writing, because nothing in it is about a patient. The aim is to be useful, short and specific. Most outreach emails fail by describing the clinic, when the reader cares about what happens to their patients and their own workload.

Write a 130-word email from a physical therapist to a primary care doctor at a practice I have not worked with. Purpose: introduce myself and offer a quick way to refer. Facts to use, and no others: we see new referrals within three working days; the doctor gets a short written update at visit 5 and at discharge; we see low back pain, knee and shoulder problems and post-operative rehabilitation; referrals can be sent by [method]. Do not claim outcomes, rankings or satisfaction scores. Tone: collegial and brief. End with one low-pressure next step: a 10-minute call or a one-page referral guide.

Illustrative result, after editing:

Dear Dr [surname],

I am [your name], a physical therapist at [clinic]. We see new referrals within three working days, and I send the referring doctor a short written update at visit 5 and at discharge, so you are never left wondering how a patient is getting on.

We treat low back pain, knee and shoulder problems and post-operative rehabilitation. Referrals can be sent by [method].

If it would help, I can send a one-page referral guide, or arrange a 10-minute call at a time that suits your practice.

Kind regards,
[your name]

Check every claim against your reality before sending. "Within three working days" is only a promise if your schedule delivers it; if it does not, say "within a week". An outreach email is the start of a relationship that a broken promise ends quickly. And no "best", "leading" or "top-rated" unless you can prove it, since AI drafts reach for those words by habit.

Keeping patient details out, including the sneaky ones

Removing a name is not enough. Details that look harmless can identify someone, especially in a small community: a rare diagnosis, an unusual job, a combination of age and date, a mention of a local sports club. Set simple rules for what may be typed into any AI tool, and train everyone to the same rule:

• Never type a patient's name, initials, date of birth, address, phone number or record number.
• Never describe a real case, even loosely. If you want an example to test a template, invent one.
• Do not upload referral letters, notes or discharge reports to a general chatbot.
• Use an account on a business plan with training on your content off by default, and check the settings yourself rather than relying on a colleague's memory.

If you want the AI to critique a real update, redact it by hand until nobody could tell which patient it is, then check the redaction by reading it as an outsider. Where your clinicians already use an AI scribe or note tool covered by a signed agreement, drafting inside that tool is a different matter, and your policy should say so. For the clinical side of what needs sign-off, see AI Use in Physical Therapy: Which Uses Need Clinician Sign-Off.

Stopping the chatbot from inventing results

The riskiest sentence in a referrer update is the one that sounds good and is not true: "excellent progress", "fully recovered", "no further treatment needed". Models write these because they are common in the genre. Three habits control it.

1. Placeholders for every outcome. The template asks for [progress against goal] and the therapist writes it from the record in measurable terms.
2. A ban in the prompt. "Do not invent any outcome, number or clinical detail" costs nothing and works most of the time. It is not a guarantee.
3. A read-through on every send. The sender checks each factual sentence against the record. If you cannot confirm it, delete it.

A realistic mistake: a clinic reuses a template whose sample text still says "good progress" and a therapist sends it for a patient who has plateaued, so the referrer hears the opposite of the truth. The fix is to keep sample text out of templates altogether: only placeholders, with the choices ([good / steady / slower than expected]) listed so the sender must pick one on purpose.

A template library you build in an afternoon

Create a shared document with six sections: referral received, progress update, change of plan, discharge note, missed-appointments notice and introduction to a new referrer. For each, run the prompt, edit by hand, and save the final text with its placeholders. A "change of plan" template, for instance, opens with the change, the reason in one sentence, and what the referrer is asked to approve (a filled-in example follows below). A missed-appointments notice states attendance factually ("2 of 5 booked visits attended") and offers a plan, with no judgement. Put the owner's name and the date of last review on the top of the page. Review the library every quarter and whenever a referrer tells you something in a message was unclear.

Here is that change-of-plan template once a therapist has filled it from the record (invented details): "Subject: [Patient initials]: plan change at visit 6. Dear Dr [surname], I would like to extend [patient initials]'s care by four visits over three weeks. Knee bend has improved from 70 to 95 degrees, but stair descent is still painful and the main goal, managing stairs at home without a rail, is not yet met. If you are happy with the extension, no reply is needed; if you would prefer to review them first, please let me know by [date]. Kind regards, [your name], [role]". The prompt that produced the shell asked for the change first, the measure that justifies it, and an explicit default (no reply needed), because a referrer who has to work out what you want will often do nothing.

Two things to check in a filled change-of-plan note: the measures must be the latest ones in the record, not those from the previous update, and the extension must match what the patient's cover allows, which is a front-desk check, not a ChatGPT one.

If you also use ChatGPT for messages to patients, keep the two libraries separate; a patient message has a different reader and a different tone, as covered in Physical Therapy Patient Retention: AI Messages Between Visits. For routine referral and insurance questions on the phone, AI Receptionist for Physical Therapy: Referral and Insurance Questions covers the other end of the same workflow.

Re-engaging a referrer who has gone quiet

Two of the 14 referrers in the example have sent nothing for ten weeks. The cause is usually mundane: a change of staff, a bad experience with a slow first appointment, or a competitor who answered the phone faster. Do not assume offence. Write a short, no-pressure message that offers something useful and asks one question.

Write a 90-word email from a physical therapist to a doctor who used to refer patients regularly and has not for about ten weeks. Do not mention numbers of referrals or imply fault. Say we have changed [one concrete thing, e.g. new-patient waiting time is now two working days] and ask whether anything about the referral process could be easier. Offer to send the one-page referral guide again. Tone: friendly and brief.

Illustrative output, edited: "Dear Dr [surname], it has been a little while since we heard from you and I wanted to check that we are still easy to refer to. We have shortened new-patient waits to two working days, and I would value hearing if anything about the process could be simpler. I am happy to resend our one-page guide. Kind regards, [your name]". Fill the bracketed change only with something that is true today. If you have changed nothing, the honest alternative is a question: "Was there anything about how we handled a recent referral that we should fix?"

A thank-you and quarterly review for your top referrers

Your three biggest referrers deserve more than template updates, and ChatGPT can prepare the quarterly note if you give it only counts. For example, paste in "Practice A: 14 referrals this quarter, average days to first visit 2.4, visit-5 update sent for 13 of 14" and ask it to turn that into a short, factual note with no flattery. The figures must come from your own records, and the note must say only what they say. Do not add outcomes you have not measured, such as "great results for your patients". An honest note, say 120 words, reads: "In the last quarter we received 14 referrals from your practice, saw each new patient within an average of 2.4 working days, and sent an update at visit 5 for all but one. If a different update schedule or format would suit your team better, please tell me." Then follow it with a call or visit. The note does not replace the relationship; it gives you a reason to start the conversation.

Discharge notes that referrers read

A discharge note is the last chance to be remembered well, and it is often the one a doctor files without reading. Keep it to three parts: what was done and for how long, where the patient ended against their main goal, and what to do if symptoms return. The template prompt is the same shape as the progress one, with placeholders for [visits attended], [starting measure], [ending measure], [self-management advice given] and [when to re-refer].

Illustrative filled version, with invented details, showing the level of specificity to aim for: "Patient completed 9 visits over 7 weeks. Walking tolerance improved from 5 to 25 minutes and the stair-climbing pain score fell from 6 to 2 out of 10. We have given a home programme and advised re-referral if pain returns above 5 for more than a week." Every number there came from the record, not from ChatGPT. That is why a measure-based template beats a prose-only one: it forces the therapist to record the numbers and gives the referrer something to act on.

A quick readability test before you send

Referrers read on a phone between appointments. Test each template against five checks:

1. Can the main point be understood from the subject line and first sentence alone?
2. Is it under 180 words, with no paragraph longer than three lines?
3. Is there one clear request, or a clear statement that nothing is needed?
4. Does it avoid abbreviations that only a therapist would know?
5. Would you be comfortable if the patient read it over the doctor's shoulder?

Ask ChatGPT to run the same checks on the template you have written and list any that fail, with the exact sentence responsible. It is good at that sort of editing. Keep the final judgement for yourself, particularly on the fifth.

Asking for referrals without crossing a line

There is a gap between good communication and inducement. Telling a doctor how quickly you see patients is communication. Sending lunches, gifts or anything of value in return for referrals may breach rules for health professionals and payers, and the rules vary by place and by payer. Keep outreach to information, access and reliability, and ask your professional body or adviser before going further. A good test: would you be comfortable if the referring doctor's patient read the email? If not, rewrite it.

Avoid also implying a patient should be sent to you rather than to a competitor in a way that undermines patient choice. A line such as "we are happy to accept referrals when a patient asks to come to us" respects the patient and reads as less pushy than "send us your back pain cases".

Checking whether it brings referrals

Measure at 90 days, not at two weeks. Referral relationships move slowly, and a doctor may only think of you the next time a suitable patient appears. Track four numbers per quarter: referrals received by practice, the number of active referrers (sent at least one in the quarter), updates sent within the promised time, and the number of introduction emails that got any reply. If the first goes flat while the third is high, the messages are not the constraint, and a visit or a call will do more than another template. If the replies are low, the email is too long or too general, and the fix is to cut a sentence, not add one.

Done well, this takes a clinic lead about half an hour a week, and the same templates serve whoever is covering. The writing help from ChatGPT is real. The relationships, as ever, belong to the people who send the messages.

Questions clinic owners ask about referrer emails and ChatGPT

Can I paste a patient's progress into ChatGPT to write the update?

Not if it identifies the patient, and not unless your policy and the tool's terms allow it. The safer method is to have ChatGPT write the template with placeholders, then fill the details in from your own record system. The chatbot never sees the patient.

How short should a referrer update be?

One screen, about 120 to 180 words. A busy referrer wants three things: whether the patient attended, what the plan and expected timeline are, and whether you need anything from them. Longer reports belong in the attachment, not the message.

Is it acceptable to offer referrers something in return?

Be careful. Offering gifts or anything of value linked to referrals can breach professional or legal rules that differ by place and payer. Offer information, clear communication and quick feedback, and ask your professional body or adviser before offering more.

Should outreach emails come from me or the clinic?

From you, signed by name, because referrers choose people. Have ChatGPT draft in your voice from two or three of your own past emails, then edit. A message that sounds like a brochure gets deleted by the person who most needs to read it.

Further reads

• Physical Therapy Prior Authorisation: Track Visits With AI — Track visits against authorisations, the other admin load.
• Best AI Scribe for Physical Therapy: Four Tools Compared — Better notes make better referrer updates.
• How Physiotherapists Use AI to Write Treatment Notes Faster — Write treatment notes faster, with the same care.
• Physical Therapy Policies and Procedures: A Clinic Manual Drafted With AI — Put your AI rules into the clinic manual.
• Will Payers Accept AI-Written Physical Therapy Documentation? — Check how payers view AI-assisted documentation.
• ChatGPT for Medical Billing: Appeals and Claim Checks Without PHI — A related no-patient-data method for billing letters.
• Best AI to Help With PT Documentation: Avoiding Cloned Notes — Use AI for PT documentation without cloned notes: what auditors look for, a per-visit detail checklist, prompts and a therapist review step.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI's ChatGPT plan pricing page (no training on Business data by default), checked November 2026; the clinic figures and messages in this tutorial are invented for illustration.