Gains:
- Ability to structure a clinical question (PICO) and use AI as a starting tool in evidence screening
- Ability to verify the authenticity, currency and level of evidence of AI references from independent sources
- Ability to interpret evidence according to the patient, context, and clinical expertise and avoid blind application
Modern physiotherapy is based on evidence-based practice (making decisions using the best current evidence, clinical expertise and patient values). But the sea of evidence is very large and the physiotherapist's time is very limited. Artificial intelligence (AI) is a powerful initial tool for structuring a clinical question, quickly summarizing the topic, and showing where to look. But there is a big trap here: AI can make up non-existent articles, fake authors and dates (hallucination). In this unit you will learn how to safely use AI in evidence screening, how to verify references and how to adapt evidence to the patient.
Configuring the question with PICO
A good evidence review starts with a good question. The PICO framework divides the question into four parts: P (Patient/Population) patient or population; I (Intervention) intervention; C (Comparison) comparison; It (Outcome) is the outcome measure. Example: “In adults with chronic low back pain (P), does motor control exercise (I) improve pain and function (O) more than general exercise (C)?” This structure both clarifies your thinking and gives a focused framework to AI. AI is very good at turning scattered curiosity into PICO.
Tip: Instead of asking the AI “what about so and so”, ask it in PICO format. Focused question brings focused and less hallucinatory response; It also makes clear what evidence you're looking for.
AI's most dangerous trap: fake reference
AI language models are trained to produce “real-sounding” text; not to know the truth. So he can make up a journal name, author, year, or even DOI number. These seem convincing but may not exist. No reference should be used without verification in an independent database (such as PubMed, PEDro, Cochrane, journal site). AI is a compass that shows you “where to look”; It is not the source itself.
Source of evidence
The role of AI
verification
Systematic review/meta-analysis
Summarizes the subject, suggests terms
Confirmation from real database
Randomized controlled trial
Indicates related work
Find and read the article, check if it is fake
clinical guide
Drafts the main message
Compare with current version of official guide
PEDro/Cochrane
Search term and question constructs
Verify score and result from source
Level of evidence and critical evaluation
Not all evidence is equal. In general, systematic reviews and meta-analyses are at the top, followed by randomized controlled trials, then observational studies, and expert opinion at the bottom. AI often does not say, or misstates, what level of evidence it relies on to support a claim. Your job is to find the source, determine the level of evidence, evaluate the quality of the study (sample, method, bias), and ask whether the result applies to your patient.
Caution: Even if a study says "it works," that study's patients may be very different from yours (age, stage, comorbidities). Applying evidence blindly is as wrong as not using evidence at all. Evidence + clinical expertise + patient values: all three decide together.
Step by step: AI-powered evidence screening
- Change the question to PICO. Have the AI help you, but you approve the final PICO.
- Request a plot summary. Get an overview of the field and key terms from the AI (don't rely on references yet).
- Verify references. Search each resource in the real database; throw away what is not found.
- Determine the level of evidence. Evaluate the design and quality of the actual source.
- Adapt to the patient. Does your patient match the study population? Filter with a clinical mind.
- Record the decision and source. Document in a traceable form what evidence you rely on.
Three mini cases (in numbers)
Case 1 — Fake reference. Physiotherapist asked AI for 5 sources for "eccentric exercise in achilles tendinopathy". The AI returned 5 references that looked real. In the PubMed check, 2 were missing (made-up author/year), and 1 had a changed title. Only 2 real sources left. Lesson: not a single reference can be used without verification.
Case 2 — PICO benefit. The vague question “what is good for neck pain” turned into this PICO with the help of AI: exercise + manual therapy (I) compared to exercise alone (C) and pain and function (O) in adults with mechanical neck pain (P). Focused questioning greatly accelerated screening and the physiotherapist achieved 3 actual systematic reviews.
Case 3 — Improper adaptation. AI summarized the result of a study for plantar fasciitis; but the study was on young athletes. The physiotherapist's patient was 70 years old, diabetic and sedentary. Rather than applying the result directly, the physiotherapist additionally sought evidence appropriate to the patient's profile and adapted the program. Lesson: do not apply blindly if the evidence does not fit the patient.
Weak prompt / Strong prompt
Weak: "Give 5 articles and evidence of the best exercise for lower back pain."
AI can generate fabricated references and present the claim without a level of evidence.
Strong: "Give a topic summary and possible key terms for the following PICO: [PICO]. If you reference, label each as 'VERIFIED' and indicate that it may not be factual; DOI/year is fictitious. When commenting on level of evidence, say it is conjecture. It's up to me; you give the starting point to the search."
Four copyable templates
Task: Convert my question to PICO format.Raw question: [clinical curiosity]I want: Parse into P, I, C, O + 5-8 search key terms (TR and EN).Just structure the question; do not claim evidence.
Task: Give a topic summary for the following PICO (I will not rely on references).PICO: [P/I/C/O]I Want: Overview of the field, controversial points, what types of sources should I look for (systematic reviews, manuals, etc.). If you provide references, label each one with "VERIFIED"; Don't give fake DOI/year.
Task: Translate this list of references into a verification plan.List: [AI-given sources]I want: For each source, write how to search in which database (PubMed/PEDro/Cochrane) + note "do not use until authenticity verified".
Task: Adapt this study summary to my patient.Study summary: [summary + population]My patient: [age, diagnosis, stage, comorbidity]I want: Differences between the study population and my patient + warnings about whether the result is directly applicable. It's up to me.
Clinical guidelines and trusted sources
The most practical source of evidence for the physiotherapist is often not individual articles but clinical practice guidelines and systematic reviews. Guidelines translate evidence in a field into actionable recommendations filtered by expert committees; This saves you time and prevents you from getting caught up in the misleading results of a single study. For physiotherapy-specific evidence, PEDro (database of physiotherapy randomized trials) and Cochrane reviews are reliable starting points. Use the AI as a bridge to these resources: ask “what current guides and systematic reviews should I look at on this topic,” then verify the resources the AI points to on their actual sites. This way, you both reduce the risk of hallucinations and work with a higher level of evidence.
Tip: Before adopting a claim, ask yourself “what does current guidance say about this?” The guidelines both summarize the evidence and highlight controversial points; It is much more solid ground than a single study.
Common mistakes
- Not verifying references. AI can produce fake sources that look real.
- Skipping the level of evidence. Not every claim carries the same weight; The design of the source is important.
- Asking unfocused questions. Questions like “what good” bring a scattered and hallucinatory response.
- Blindly applying the evidence. If the study population does not match the patient, the result may not be valid.
- Mistaking the AI summary as a source. The AI summary is the compass; It is not the source itself.
- Not checking for updates. An outdated guideline may conflict with current evidence.
In summary
AI is a valuable starting tool in evidence screening to structure the question (PICO), summarize the topic, and show where to look. But he can make up references; No source can be used without verification in an independent database. Determine the level of evidence yourself, evaluate the quality of the study, and adapt the conclusion to your patient with clinical wisdom. Evidence, clinical expertise and patient values decide together.
Application task
Choose a clinical interest. Convert to PICO with the first template, get a plot summary with the second template. If AI referenced, plan to verify with the third template and search for at least one reference in a real database (note if you find it). Finally, adapt a study summary to your own patient with the fourth template.
checklist
- [ ] I structured my question into PICO format.
- [ ] I verified the AI references against an independent database, eliminating fabrications.
- [ ] I evaluated the level and quality of evidence of each source.
- [ ] I checked the validity of the proof.
- [ ] I adapted the result to my patient's profile with clinical wisdom.
- [ ] I have recorded the decision and its source in a traceable form.