A physiotherapy clinic can use AI to turn a clinician's short, written facts into a clear progress update for a referrer or GP: the clinician supplies what was found and what happens next, the AI shapes it into a letter, and the clinician reads, corrects and signs it. Consent covers the sharing, and nothing is sent unreviewed.
The important word is "facts". The tool writes the paragraph; it does not interpret the patient. A letter that reads well but contains a finding nobody recorded is worse than a plain letter, so the whole method rests on giving the AI only what the clinician has written and asking it to add nothing else.
What do referrers actually want to read?
A referring doctor or consultant scans a letter for three things: what you found, what you are doing, and what you need from them. Everything else is padding. A six-line update that says "improving, expected discharge in three weeks, no action needed from you" is more useful than a page of narrative. Before you build any template, ask two or three regular referrers what they want, and what they skip. Their answers shape the structure better than any prompt.
The same letter should answer: Has the problem changed since my referral? Is there something unexpected I should know about? Should I do anything? Will the patient be seen again, and when? A good template has a labelled line for each, which also makes the clinician's input quick.
The input is the clinician's bullet points
The method has one rule: the clinician writes the facts, the AI writes the sentences. A clinician spends two minutes after the session typing bullet points in a form, and the form is the input. A minimal version has these fields:
- Referral reason, in the clinician's words.
- Key findings at assessment or this review (the clinician's own measures and observations).
- Treatment so far, with the number of sessions.
- Progress, as the clinician judges it: improving, plateau or worsening.
- Any red flag or unexpected finding, yes or no, with details.
- Plan and expected next review or discharge.
- Any request of the referrer, such as imaging, medication review or a further opinion.
Nothing in the form is generated by the AI. If a field is empty, the letter says nothing about it, rather than inventing a plausible sentence.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA prompt that refuses to guess
You write progress letters from a physiotherapy clinic to a referring doctor.
Use ONLY the facts below. Do not add diagnoses, measures, advice, prognosis or reassurance that are not stated.
If a field says "none" or is blank, do not mention it.
Style: plain, professional, under 180 words, no headings, start "Dear [referrer name],". End with "Kind regards," and leave the signature to the clinician.
Patient reference: [initials or reference only]
Referral reason: [text]
Findings: [text]
Treatment so far: [text]
Progress: [text]
Red flags: [text]
Plan: [text]
Request of referrer: [text]Note the patient reference field. A short initials-and-reference pair is enough to let the AI write the letter, and the full name, date of birth and identifier are added afterwards in your practice system when the letter is merged. That keeps the sensitive identifiers out of the drafting step.
An illustrative progress update, with what the clinician fixes
Take a clinic with three clinicians. A clinician fills in the form for a patient referred with persistent shoulder pain after six sessions: referral reason "shoulder pain, 10 weeks, no injury"; findings "range improved from 90 to 140 degrees, pain on overhead reach reduced from 7 to 3 of 10, strength unchanged"; treatment "six sessions of exercise-based rehabilitation, home programme"; progress "improving"; red flags "none"; plan "four more sessions, then discharge with self-management"; request "none".
An illustrative draft:
Dear [referrer name], thank you for referring this patient with ten weeks of shoulder pain and no history of injury. After six sessions of exercise-based rehabilitation and a home programme, shoulder range has improved from 90 to 140 degrees and pain on overhead reach has fallen from 7 to 3 out of 10. Strength is unchanged. We have no red-flag concerns. We plan four further sessions and expect to discharge to self-management after that. No action is needed from you at this stage. Kind regards,
Checking it against the facts, the clinician finds the draft accurate. If the AI had written "good strength recovery", or "full range expected", those would be additions and out. If it had written "no concerns" where the form said only "red flags: none", that is a small softening to watch, since "no concerns" is a broader claim than "no red flags". The clinician's two-line check takes seconds once the habit exists.
The four letters most clinics need
Acknowledgement of referral
Short, sent by the admin team from a fixed template, with no clinical content: "We have received your referral and we will offer the patient an appointment within [your normal timeframe]." This one needs no AI at all, and its value is that referrers hear back. If you want it to vary slightly by referral source, use the template with a merge field.
First-assessment summary
The most detail-heavy letter. It restates the referral reason, the main findings, the working problem as the clinician has recorded it, the treatment plan and the expected number of sessions. Aim for 150 to 250 words. Use the same form, with an extra field for the clinician's working impression, which the AI must copy as written, without paraphrasing the clinical wording.
Progress update
The letter above. Send it at the point the clinician agreed with the referrer, or when something changes. A workable default is one update at about the midpoint of a course and one at discharge, plus an immediate contact when a red flag appears.
Discharge summary letter
The closing letter: what was done, outcome against the starting point, advice given to the patient to carry on, and whether to re-refer. Ask the AI for a closing line that tells the referrer what would prompt a re-referral, but only if the clinician has supplied that wording, as it is a clinical judgement.
A discharge summary, filled in
Discharge letters are where referrers want an outcome. An illustrative form for a patient with low back pain, discharged after eight sessions: referral reason "low back pain, 6 weeks"; starting measure "pain 6 out of 10, unable to sit longer than 20 minutes"; final measure "pain 2 out of 10, sitting tolerance 90 minutes"; treatment "eight sessions of exercise and advice"; advice given "continue home exercises daily, return if pain limits sitting again"; re-referral wording "none needed unless symptoms return". The AI's job is to convert this into four sentences. A draft that says "excellent outcome" is the AI praising the result, and the clinician replaces it with "pain fell from 6 to 2 out of 10 and sitting tolerance rose from 20 to 90 minutes", because the numbers speak better than an adjective and are the clinician's own.
The same form can add one extra field: "Outcome measure name and score, if used". If the clinic uses a named questionnaire, the clinician types the name and the two scores, and the AI repeats them exactly. If the field is blank, the letter says nothing about measures, rather than quietly inventing a scale.
Different referrers want different letters
The same facts need a different register depending on who reads them. A family doctor wants the short version with any action for them at the top. A hospital consultant who referred for a specific question wants that question answered first. An insurer or an employer, where the patient has agreed to share, wants dates, session counts and functional outcomes and nothing clinical beyond that. Keep one master form and three light variants of the prompt:
- Doctor: under 150 words, with a line beginning "Action for you:" even when it says none.
- Consultant: opens with the referral question restated, then the answer from the clinician's findings.
- Insurer or employer: dates, number of sessions attended, functional outcomes the clinician has recorded, expected end date. No diagnosis unless the patient has agreed and the form says so.
An illustrative error: the insurer variant is used by mistake for a letter that the patient has only agreed to share with their doctor. Add a "who is this for and has the patient agreed?" tick to the form so the variant is chosen on purpose, and let the admin team check that tick before the letter goes anywhere.
Messy bullets in, tidy letter out
Clinicians do not write tidy forms at the end of a long day. An illustrative before: "pt much better, rom nearly full, still stiff am, 3 more sessions then d/c, gp to review meds?" The AI can expand abbreviations and give structure, but the question mark is the problem. Is the clinician asking the doctor to review medication, or wondering aloud? A tool may well turn it into a firm request, which would be a clinical suggestion nobody made. The safeguard is a line in the prompt: "If any field is ambiguous, ask a question instead of writing it", and a habit for the clinician to answer. With that line in place, an illustrative draft comes back with "Clarify: do you want to ask the GP to review medication?" and the clinician either confirms or removes it.
Encourage clinicians to type full sentences for anything that ends up as a request. A request typed in a hurry is the most likely to be altered by a tool, and a request that changes in transit is the kind of error referrers notice.
A short audit once a month
Once a month, the clinic lead pulls ten sent letters at random and compares each with its form. The questions are narrow: does every number match, is there a sentence that is not supported by the form, and did the right variant go to the right recipient? Write the results on a single page with the date. If one in ten has an unsupported sentence, tighten the prompt and retrain the habit. If none do for three months, you can reduce the audit to five letters. The audit costs about twenty minutes and is the clinic's evidence that review is real and not a tick-box.
When a patient asks not to be written about
Occasionally a patient asks that their doctor is not sent an update, or that an employer receives less than usual. The clinic's process has to make that choice visible, because the quickest letter system in the world is no use if it ignores a preference. Add a field at the top of the form, filled from the patient's record: "Sharing preference: doctor yes or no, other recipient yes or no, anything withheld". If the answer is no, the draft button stays off and the clinician records the conversation. An illustrative case: a patient agrees to updates to their doctor but not to the employer who referred them; the form shows that, the admin member sees it before merging, and the employer receives only the attendance note the patient approved. Build the preference into the form once and the AI never has to be trusted with it.
Separating a red flag from a routine update
A system that produces polite letters in a minute creates one danger: an urgent finding written in the same calm tone as everything else. The form's "red flags" field should be a yes or no, and any yes bypasses the template. The clinician rings or writes a separate, immediate message in their own words, as they would have done without AI. A useful rule on the form: if red flags is "yes", the AI draft button is disabled, or the template carries a bright warning line asking the clinician to contact the referrer directly.
An illustrative failure: a letter draft arrives with "no significant change" for a patient whose new numbness the clinician mentioned verbally but did not enter in the form. The AI could not know. The clinician caught it on read-through only because the habit was to read the letter against the patient's session, not just against the form. That is the reason the clinician, not the admin team, does the final read.
Consent and what the patient is told
Sharing a progress update with a referrer is part of ordinary care in most clinics, and patients usually expect it. Using an AI tool in the process is a separate thing to be open about. Include one sentence in your consent wording and patient information: "We may use a computer-assisted writing tool to help draft letters about your care. Your clinician reads and signs every letter, and we do not give the tool your name or contact details." Check the exact wording against your professional body's guidance and your clinic's data-protection policy; the sentence is a starting point, not legal advice.
If you also use an AI scribe for notes, the consent step is larger and worth treating on its own. The tutorial on AI scribes for physiotherapists: consent and records checklist covers it, and the tutorial on how physiotherapists use AI to write treatment notes shows the note-writing side that feeds your forms.
Time and cost: a quick sum for a three-clinician clinic
An illustrative clinic sends about 30 letters a week across its three clinicians, and each takes about eight minutes to write from scratch, so four hours of clinician time. Say the structured form takes two minutes and the read-and-sign takes two minutes, so four minutes per letter, two hours a week. That is a saving of about two hours, spread over three people. Those figures are assumptions for the sum, and your own timing of 10 recent letters is the number to use.
The cost side is small. Any general chat assistant with a suitable plan or the writing feature of your practice software can do it, and the cost depends on the plan you already have. The larger cost is the setup, a form that clinicians fill in reliably, and the agreement of the referrers about what they want. Do the referrer conversation first, as any saving disappears if letters come back with "please add the outcome measure".
Templates that stay in the clinic's voice
Clinics have a voice, usually warmer or more clipped than a generic letter. Give the AI two or three of your best past letters, with identifying details removed, as style examples. The instruction "match this tone and length" is often enough. Update the examples when a clinician leaves or when referrers give feedback.
Keep a small checklist at the bottom of the form that the clinician ticks before sending:
- Every number in the letter matches my notes.
- Nothing is in the letter that I did not write on the form.
- If red flags is yes, I have contacted the referrer directly.
- The right referrer and the right patient identifiers have been merged.
The last line catches the most embarrassing error of all: a correct letter attached to the wrong patient. Merge identifiers in the practice system, never by typing them into a chat window, and check the name on the letter against the record every time.
Where it fits in the clinic's wider admin
Referrer updates sit at the end of a chain that starts with the referral itself. If you are tracking who has been referred, who has been seen and who has not turned up, the referral tracking and chase letters tutorial covers the front of the chain, and AI tools for physiotherapy clinics maps every other job. The tutorial on referral partner updates for mortgage brokers shows the same structure in a different trade, which is a useful way to see how little of the method depends on the sector.
A two-week trial inside the clinic
Pick one clinician and one letter type, the progress update. For two weeks, they write every update with the form and the AI draft, timing each, and also note any edit that changed meaning. At the end, put three questions to the referrers who received them: was it clear, was anything missing, was anything unnecessary? If the answers are good and the edits are small, add the first-assessment summary. If meaning-changing edits are frequent, tighten the prompt before you add anything. If you would rather design the process with someone, that is what my AI implementation consultation covers.
Common questions about AI-written referrer letters
Does the clinician still need to read every letter?
Yes, every time, and the clinician signs it. The letter goes out under their name and professional responsibility, so the review is part of the job. A draft that arrives in a minute is useful only if the reading takes a minute too.
Can I paste the clinical notes into the AI tool?
Only if your clinic has checked the tool's data terms, retention and any required agreement, and the patient has consented to this use of their information. Many clinics avoid the question by giving the tool only a short list of facts with no name, date of birth or contact details, then merging identifiers afterwards.
Should the patient see the letter first?
Offering a copy is good practice and some clinics send one automatically. Check what your professional body and local rules expect, and say so in your consent wording, so the patient knows who will receive information about them.
Can the AI add recommendations or prognosis?
No. Tell it to use only the facts given. Prognosis, advice and any change to a plan are clinical judgements, and they belong in the clinician's own words.
How many letter types do we need?
Four cover most clinics: an acknowledgement of referral, a first-assessment summary, a progress update and a discharge summary. Add a short non-attendance notice if referrers ask for one.
Further reads
- AI Receptionist for Physical Therapy: Referral and Insurance Questions — Referral and insurance questions at the front desk.
- Creating Home Exercise Programmes With AI for Physio Patients — Home exercise handouts for patients, drafted with AI.
- Marketing a Physio Clinic With AI: Posts, Tips, and Newsletters — Posts and newsletters that keep referrers aware of you.
- AI Receptionist for Physiotherapy Clinics: Cost and Booking Setup — The front-desk side of the same clinic.
- Physiotherapy Clinic Software Compared: Which Has Built-In AI? — Which clinic software has built-in AI.
- AI Secretary for Doctors: Referral Letters, Inbox, Insurer Forms — The other end: how doctors handle incoming letters.
- AI Marketing for Medical Practices Without Making Health Claims — Service pages, review replies, referrer updates and local search for a medical practice, drafted with AI and kept free of health claims.
- Chartered Society of Physiotherapy and AI: A Clinic Checklist — Turn the CSP's AI principles into a working checklist: what each asks of a clinic, who owns it, what evidence to keep and when to review.
- NDIS Occupational Therapy Report Template: A First Draft With AI — Report headings, an evidence table, a de-identified drafting method and a review checklist for an AI-drafted NDIS OT report you stay accountable for.
- Can Dietitians Use AI for Admin and Content? What Stays Human — Which dietitian tasks AI can draft (admin, handouts, posts) and which need the dietitian's own judgement, with prompts and data rules.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the clinic's own letter templates and consent wording as source text; general plain-English letter practice. No vendor claims are made in this tutorial.