Gains:
- Ability to distinguish where AI saves real time in the physiotherapy workflow and where the clinical decision and responsibility should remain with the competent physiotherapist, based on the level of risk
- Ability to implement a multi-layered validation discipline that tests each AI output against clinical examination, evidence and physical therapist validation
- Ability to acquire the habit of anonymizing the context to protect patient data and choosing safe tools in terms of KVKK and professional confidentiality.
Physiotherapy is a long chain of decisions that changes a person's movement, pain and daily life. In this chain, you examine, touch, feel, observe the patient and devise the most appropriate treatment for him. Some links in this chain (writing reports, preparing exercise narration, setting reminder messages) can quickly become automated; Others (diagnosis, clinical decision, tactile assessment, responsibility) can never be delegated to a machine. Artificial intelligence (AI for short; software that learns patterns from data and produces text, images or data) is a very powerful accelerator in this chain: it drafts programs, writes text, organizes tables, summarizes data. But AI is not a physical therapist; He does not examine the patient, does not take responsibility, and does not sign the treatment. In this unit you will learn where to use AI in physiotherapy work, where to stand and how to validate each output. The goal is to make you a clinical professional who oversees AI, not dependent on it.
What AI can and cannot do in physiotherapy
The real power of AI is to generate language, patterns and outlines. A home exercise program outline, an evaluation report draft, a patient education brochure, a session note layout, a motivational message; These come off in seconds. In these tasks, AI saves you hours each week and reduces burnout. It reduces the time spent in front of the keyboard and increases the time spent with the patient.
What AI cannot do is decision making, which requires clinical examination and responsibility. Sensing a red flag (e.g. suspicion of cancer, cauda equina syndrome, sign of thrombosis), manually evaluating the end-feel of a joint, finding the real source of pain through differential diagnosis; None of this can be determined by the logic of "this is how it usually happens." AI hallucinates because it learns from texts on the internet: that is, it can make up a number of repetitions, a contraindication, a source, or an angle that appears real but is false. In physiotherapy, this type of error is not just a typo; It means that the patient is injured, deteriorates, or a real disease is overlooked.
Caution: The AI output is a draft, not a clinical opinion. Diagnosis, differential diagnosis and treatment decisions are based on the examination of a competent physiotherapist. No safety-critical decision can be made without the approval of the physiotherapist (and physician when necessary). AI does not replace this consent.
Three regions according to risk level
The most practical way to use AI safely is to divide each task into three zones based on risk level. This distinction quickly and accurately answers the question “should AI do this?”
Region
Sample tasks
The role of AI
Verification level
Green (low risk)
Brochure language, reminder message, e-mail, title, session note layout
Free draft production
Quick review
Yellow (medium risk)
Exercise program draft, evaluation report, patient education text, evidence summary
Draft + proposal
Physiotherapist check + source/clinic confirmation
Red (security-critical)
Diagnosis, differential diagnosis, red flag interpretation, contraindication decision, referral decision
Pre-screen/checklist only
Competent physiotherapist examination and approval is mandatory
Be fast in the green zone. Use AI in the yellow zone but confirm every clinical claim, count and contraindication. In the red zone, AI never has the final say; It is merely an aid that speeds up the expert's work.
Multi-layer verification discipline
Verification is not something to be postponed thinking "I'll check it out later"; is part of the workflow. A solid verification consists of five layers:
- Return to examination. If the AI made a recommendation, compare it with the patient's actual exam findings (pain, range of motion, strength, comorbidities).
- Clinical mind. Is the recommendation appropriate to the stage, the healing biology of the tissue, and the patient's goals? Are there any contraindications?
- Test the evidence. Confirm numerical or clinical claims with current evidence and clinical guidelines.
- Expert eye. If the decision is in the field of another discipline (physician, orthopedist, neurologist), seek consultation.
- Leave your mark. Record which output was validated and how; Let it be checked later.
Hint: "AI said" is not a justification. The justification for a physiotherapy decision is always examination findings, clinical reasoning, evidence and, where necessary, expert opinion. AI helps you prepare these justifications, it does not replace them.
Privacy, KVKK and ethics
Patient data is sensitive and special personal data. A patient's name, diagnosis, photo, video, contact information; All of these are within the scope of KVKK (Personal Data Protection Law) and are subject to professional secrecy liability. Before giving data to a cloud-based AI tool, ask three questions: Is this data really necessary? Can it be anonymized? Does the tool I use use or store data in training? The safest way is to provide only the clinical context ("45-year-old female patient, chronic low back pain for 3 months") anonymously, without sharing any personal data.
Ethics is broader than privacy. AI output may be biased (trained with insufficient data for certain populations), you need to transparently inform the patient that you are using AI, and you cannot make automated decisions without the patient's consent. Responsible use means protecting the patient as the subject of the process.
Step by step: Introducing AI safely into a task
- Place the task in the region. Green, yellow or red?
- Anonymize context. Clear real name, contact and identity data.
- Give a clear brief. Clearly write the diagnosis, goal, constraints, and format.
- Consider the output a draft. Never use it like the final product.
- Apply layers of verification. Examination, clinical reasoning, evidence, expert, trace.
- Record the decision and its justification and inform the patient.
Four copyable templates
Role: You are a physiotherapy assistant (no clinical decision making).Task: Produce a DRAFT for [task].Context: Anonymous patient profile [age, gender, diagnosis, stage, goal, restrictions].Rule: Label "CONFIRMATION REQUIRED" for each clinical claim/number you are unsure of.Rule: Warn possible contraindications in a separate heading.Format: Pointed, reasoned, plain language.
Task: Mark each clinical/medical claim in the text below.Text: [Previous output of AI]I want: For each claim, (1) the claim, (2) the type of source that needs to be verified,(3) whether there is a red flag. Don't suggest a decision, just mark it.
Task: Anonymise the following patient text.Text: [text containing personal data]Rule: Replace identifiers such as name, contact, identity, institution, date with [TAG].Preserve clinical context. The output is only anonymized text.
Role: Critical reviewer with the eye of a senior physiotherapist.Task: Find risks in the AI output below.Output: [text to review]I want: Every point that is wrong/missing/risky to security + why it is risky+ recommended verification step. Don't praise, just show the risk.
Weak prompt / Strong prompt
Weak: "Write an exercise program for lower back pain."
This prompt does not include the patient, stage, red flags, and constraints; AI produces a generic and potentially risky list.
Güçlü: "Produce a DRAFT home exercise program for a 42-year-old patient who has mechanical chronic low back pain for 3 months, has no red flags, and works sitting down. There is no acute exacerbation. Write 'CONFIRMATION REQUIRED' for every load you are not sure about, warn possible contraindications separately. In plain language, in justified items. The clinical decision is mine."
Strong prompt context makes clear the security constraint and the role of AI; The output is a verifiable draft.
Common mistakes
- Mistaking AI output for clinical opinion. The output is a draft; It is not a substitute for examination and clinical wisdom.
- Leaving the red flags to AI. Differential diagnosis and danger signs are human responsibility.
- Pasting personal data. Sharing name, ID, contact, identifiable photo/video is a violation of KVKK.
- Not verifying the source. AI can make up references and numbers; Every claim must be verified.
- Not informing the patient. It is an ethical requirement to transparently state that you are using AI.
- One prompt, one output. Good results usually come with several rounds of critical exchange.
In summary
AI is a powerful accelerator in physiotherapy: it produces drafts, text, abstracts and layouts. But diagnosis, clinical decision, tactile evaluation and responsibility belong to the competent physiotherapist. Place every task in the green/yellow/red risk zone, anonymize personal data, five-layer verification of every output, and inform the patient transparently. Be the professional who oversees AI; not the one who submits to it.
Application task
Choose an anonymous patient scenario from your own practice. Using the first and second templates in this unit, first produce an exercise program outline, then have the clinical claims marked on the same printout. For each claim, write in one sentence how you would verify it. Find at least two "VERIFICATION REQUIRED" points and indicate the actual source of verification.
checklist
- [ ] I placed the task in the green/yellow/red zone.
- [ ] I anonymized personal data, I never shared unnecessary data.
- [ ] I treated the AI output as a draft, not clinical opinion.
- [ ] I implemented the five-layer verification (examination, clinical reason, evidence, expert, trace).
- [ ] I considered red flags and contraindications as human.
- [ ] I informed the patient about the use of AI and made the decision.