User Guide to our Evidence-based Medical AI

PulseLife gives healthcare professionals access to its evidence-based medical AI. It is not a substitute for clinical judgement: its role is to speed up access to medical knowledge and support your clinical thinking. 

This guide is designed to help you: 

  • Write effective queries (“prompts”); 
  • Interpret responses critically; 
  • Use AI within a compliant and ethical framework. 

💬 Have a question? Contact us here. 

Ask your Medical AI Expert a question

1. Principles of responsible use 

1.1 Confidentiality and data protection 

  • PulseLife's technology filters out individual-level information entered by mistake that the system could otherwise capture. 
  • You remain responsible for: 
  • Never providing any data into PulseLife that could identify an individual (name, date of birth, patient record number, address, etc.) 
  • Providing only the clinical information necessary to answer the question asked 
  • The AI Medical Chatbot runs on independently certified, GDPR-aligned infrastructure 

1.2 Clinical responsibility 

PulseLife's AI Expert provides a view of the current state of medical evidence; it is not a decision-making device. Clinical responsibility remains entirely with the healthcare professional. 

2. Understanding the “medical prompt” 

2.1 Definition 

A prompt is how you phrase your request to the AI system: the more precise your instruction, the more relevant the response. 

2.2 Structure of an effective prompt 

A good prompt follows this logic: Clinical Context + Question OR Objective + Constraints 

Clinical context (where relevant): provide the clinical information needed to answer the question in context (sex, age, medical history, observations, symptoms, etc.) 

  • “pregnant woman” 
  • “history of myocardial infarction; moderate renal impairment” 

Question OR expected objective: write a precise question OR clarify your objective (relating to the context, if one is given) 

  • “… what are the possible oral anticoagulation strategies” 
  • “… does the literature recommend gestational diabetes screening” 

Additional instructions (where relevant): 

  • “… according to the latest RCOG guidance” 
  • “… according to the latest NICE guidelines” 
  • “… show the benefits/risks” 

 

Further examples: 

Clinical context + Question + Constraints: 

  • 64-year-old patient, exertional dyspnoea, lower-limb oedema, elevated BNP, echocardiogram showing LVEF of 35%. What is the current treatment strategy according to NICE guidance? 

Clinical context + Question: 

  • 28-year-old man, sleep disturbance, anxiety, occasional dark thoughts. What criteria help distinguish major depression from an anxiety disorder? 

Simple question: 

  • What are the contraindications for methylphenidate (e.g. Ritalin)? 
  • What investigations are indicated when endometriosis is suspected? 
  • What are the indications for active surveillance in prostate cancer? 

 

2.3 Where to start 

  1. Start with a subject you know well; build your confidence and test the tool's limits. 
  2. Try a complex question you're confident in. 

32-year-old woman presenting with severe dysmenorrhoea and chronic pelvic pain resistant to NSAIDs. Regular cycle, but pain worsens premenstrually; deep dyspareunia. On examination: tenderness in the posterior fornix, tender nodules on the uterosacral ligament. Pelvic ultrasound shows a 3 cm left ovarian endometrioma, suggestive of endometriosis. How can the diagnosis of endometriosis be confirmed? 

2.4 Common mistakes 

  • No question or objective: “64-year-old patient, exertional dyspnoea, lower-limb oedema, elevated BNP, echocardiogram showing LVEF of 35%” 
  • A question or objective that's too vague: “What treatment for breast cancer?”, “Explain diabetes” 
  • Too much irrelevant information: “56-year-old patient, diabetic on metformin for 5 years, works as a teacher, persistent cough for 5 days with no fever: what breast cancer screening is recommended?” 
  • Several questions at once: asking one question at a time gets you a precise answer to each of your queries 

3. Best practice by type of need 

3.1 Exploring clinical options 

Aim: explore the evidence 

Sample prompt: 

“For an adult patient with BRAF-positive metastatic melanoma, what are the first-line options according to the latest phase III studies?” 

Tips: 

  • Set out the clinical context clearly (age, symptoms, comorbidities, stage, etc.) 
  • Frame the question clearly in relation to that clinical context 

3.2 Literature review or scientific update 

Aim: access a clear, well-sourced summary 

Sample prompt: 

“Summarise the main international recommendations (2023–2025) on the management of type 2 diabetes, citing sources.” 

Tips: 

  • Set out clearly what you're looking for 
  • Be specific about your instructions (which guidelines, which date range, etc.) 

4. Compared examples: “poor” vs “good” prompt 

Objective 

Poor prompt 

Good prompt 

Therapeutic options 

“What's the treatment for diabetes?" 

“In a 65-year-old patient with poorly controlled type 2 diabetes on metformin, no renal impairment, what are the current recommendations for second-line treatment according to NICE guidance (NG28)?" 

Literature review 

"What's new on lung cancer?" 

“Cite the main recommendations published since 2023 on metastatic NSCLC with KRAS mutation and summarise their findings in 5 points." 

5. Verification and clinical judgement 

5.1 Assessing the reliability of a response 

Ask yourself these three questions every time: 

  1. Is the response clearly based on one or more legible reference(s)? 
  2. Are the references appropriate (region, source, date)? 
  3. Is the reasoning consistent with current clinical practice and my own experience? 

5.2 Rephrasing to improve quality 

If the response is incomplete or too general: 

  • Add context (“in the case of moderate renal impairment…”) 
  • Add instructions (“base this only on international guidelines”) 
  • Rephrase if needed, clarifying the question or objective “as if you were speaking to a colleague” 

6. Security, compliance and PulseLife best practice 

6.1 Secure environment 

PulseLife operates on infrastructure that is: 

  • Independently certified to HDS, the French national standard for secure health-data hosting; 
  • Protected under GDPR-aligned safeguards; 

6.2 Limitation of liability 

Suggestions produced by the AI system do not constitute a prescription or a recommendation. The healthcare professional remains solely responsible for the final decision and its outcome. 

6.3 Training and supervision 

Users must be trained in the critical reading of AI-generated responses. They must always supervise the tool. 

7. Conclusion: AI as a partner in clinical reasoning 

Artificial intelligence replaces neither the knowledge, the intuition, nor the responsibility of the healthcare professional. 

Used rigorously, it becomes a partner in clinical reasoning: it structures your thinking and speeds up access to knowledge. 

Prompting well means caring well: phrase your queries with precision and check the answers with judgement. 

❓ Have a question?
Contact us here.

PulseLife – Trusted medical intelligence 

Secure. Human. Clinical.