ChatGPT for health can help you prepare for an appointment: organize your symptoms, explain a term and draft questions to ask. The seven prompts below are for producing notes to check, without making a diagnosis, ruling out an emergency or changing treatment.
Misuse starts when a fluent answer becomes a medical decision. A language model produces probable text. It does not examine you, does not necessarily know your full medical history and can provide a nonexistent reference in a convincing tone.
The simple rule: use ChatGPT to prepare for the conversation with your healthcare professional, never to replace it.
Which ChatGPT health prompt should you use before your appointment?
Choose a specific task before copying your data:
| Your need | Prompt below | What to check yourself |
|---|---|---|
| Scattered symptoms | 1, timeline | Dates, facts and missing information |
| Notes that are too long | 2, one-page summary | No facts incorrectly added or removed |
| Questions forgotten at the appointment | 3, questions for your healthcare professional | Priorities suited to the reason for the visit |
| A report that is hard to read | 4, explanation | Consistency with the original document |
| Blood test results | 5, descriptive table | Numbers, units and ranges |
| Treatment list | 6, formatting | Names, doses and frequencies copied accurately |
| Medical claim | 7, sources | Links opened and content consistent |
For an appointment about persistent fatigue, the guide to causes to explore and fatigue assessment · FR helps clarify the reason for the visit. For glucose, prepare the questions and tests for insulin resistance, without asking AI to diagnose you.
What ChatGPT can do, and what it must not do
ChatGPT can help you structure what you already know. It is particularly useful for rewriting a report in plain French, sorting symptoms by date, condensing a treatment list or producing a preparation sheet before an appointment. You remain the source of the facts, and the professional remains responsible for their clinical interpretation.
It must not decide which disease you have, rule out an emergency, change treatment, calculate a dose on its own or claim that a laboratory result is “normal for you.” Two people with the same laboratory value may need different interpretations depending on their age, symptoms, possible pregnancy, medications and medical history.
Published evaluations make this caution necessary. A systematic review of 137 studies found considerable variation in methods and insufficient consideration of safety issues. A diagnostic meta-analysis found highly variable performance and a high risk of bias in most of the included studies.
The latest research does not validate “Dr. ChatGPT.” In July 2026, a study of 22 questions about adult type 1 diabetes obtained good scores for ChatGPT-5.0, but only on a questionnaire derived from guidelines and assessed by endocrinologists. Another study on short stature and growth delay found poorer answers on diagnosis and treatment, along with fewer recommendations to seek care in the most recent model. These results are specialized, depend on the version tested and cannot be generalized to all symptoms.
Data to anonymize before asking any question
Health data is not limited to your name. The CNIL includes test results, medical history, treatments and information that can reveal your health status. Removing your first name is therefore not always enough. A rare disease, an exact age, a hospitalization date and a small municipality can allow re-identification.
The CNIL also distinguishes pseudonymization from anonymization. Replacing a name with initials reduces the risk, but the data can still potentially be linked to a person. For a one-off question, use minimization instead: share only what the task requires.
Before copying a document into a general-purpose chatbot, remove the following:
- last name, first name, photo, signature and full date of birth
- address, phone number, email, employer and precise municipality
- French social security number, INS, patient record number and laboratory identifier
- login details, private portal link and password
- doctor’s name, facility, unnecessary exact dates, QR codes and barcodes
- information about a child, relative or anyone else without their consent
- raw genetic data file or complete medical record when a few lines are enough
However, keep the units, the laboratory’s reference ranges, the relevant approximate date, symptoms and medications useful to the question. Without these details, the explanation can become misleading.
Is ChatGPT Health available in France, and does it protect your data?
As of October 13, 2026, OpenAI’s current documentation indicates a gradual rollout of Health for eligible users aged 18 or over in the United States, on the web and iOS. It does not announce general access in France. This prompt guide can still be used in an ordinary conversation, but do not assume that your account has Health.
The documentation now distinguishes connected medical or Apple Health data, and conversations that use it, from other exchanges. The former are not used to train foundation models or target ads; conversations that do not use this data follow your usual data-control settings. A health question therefore does not automatically receive this special treatment.
Removing identifying details and limiting information remain useful. A connection, a privacy setting or a paid account does not turn an answer into a diagnosis. Check the service’s settings and policy when you use it.
Seven prompts to prepare for an appointment
Replace the brackets with minimal, already anonymized information. If the tool cannot answer without identifiable data, stop rather than add your entire record.
1. Turn notes into a timeline
Act as a medical preparation secretary, not a doctor. Using my anonymized notes, create a factual timeline with: approximate date, symptom, duration, intensity from 0 to 10, possible trigger and action already tried. Do not invent anything. Put any missing information in a section called “To clarify at the appointment.” Do not suggest a diagnosis or treatment.
Notes: [anonymized notes]
2. Produce a one-page summary
Condense the following information into a one-page appointment summary. Organize it into: main reason for the visit, onset and progression, impact on daily life, relevant medical history, treatments and supplements with known doses, allergies, tests already performed and three priorities to discuss. Flag contradictions without resolving them. No diagnosis.
Information: [anonymized text]
3. Prepare the right questions
Based on this reason for the appointment, suggest ten short questions to ask the professional. Sort them into three categories: causes to explore, tests or follow-up to discuss, signs that should prompt another consultation. Do not state that a test is necessary or recommend any medication.
Reason for the visit: [reason without identifying details]
4. Translate a report
Explain this anonymized report in plain French, sentence by sentence. For each technical term, give a neutral definition, then state what the document actually says and what it does not allow us to conclude. Do not fill in missing information or make a diagnosis.
Report: [relevant excerpt]
5. Read blood test results without overinterpreting them
Present these anonymized results in a table with: test, value, unit, laboratory range and general definition. Identify only values outside the provided range. Explain why an isolated value is not enough to make a diagnosis. End with five questions to bring to the doctor. Do not describe the situation as benign or serious.
Results: [values, units and ranges]
A full screenshot is rarely necessary. Copy the relevant lines after checking them, because automated image reading can also misread a digit, a decimal separator or a unit.
6. Make your medication list safer before the appointment
Format this list for a doctor or pharmacist: exact name, dose as stated, frequency, time taken, reported reason and approximate start date. Include supplements. Flag possible duplicate names as “to check,” without concluding that there is an interaction or advising any discontinuation or change.
List: [anonymized medications and supplements]
7. Require verifiable sources
Answer as a cautious medical information specialist. Separate: established facts, uncertainties, missing information and questions for a professional. For each important claim, provide a primary or institutional source with the exact title, organization or journal, date, DOI or PMID if available and direct URL. If you cannot verify a reference, write “unverified source” and do not invent one. Specify the population studied and why it is or is not comparable to my case.
General question: [non-urgent, anonymized question]
Can ChatGPT interpret blood test results?
It can copy and explain the results, but a value outside the range is not enough to establish a disease. Use prompt 5 for a descriptive reading. If the model mixes up g/L, mmol/L, dates or ranges, correct the notes using the laboratory report and bring the original document to your healthcare professional. A reassuring answer does not determine whether you can wait.
How to check a medical answer
A better prompt reduces some errors, but it does not turn the model into a certified medical database. The CNIL points out that a generative system follows a probabilistic approach and can produce inaccurate but plausible information.
Run every answer through this checklist before using it to prepare for your appointment:
- Traceability: does every important fact lead to an institutional page, a PubMed article or a DOI that actually exists?
- Consistency: do the source’s title, abstract and conclusion genuinely support the statement?
- Currency: is the recommendation still in effect and applicable in France?
- Population: does the study concern comparable adults, children, hospitalized patients or another group?
- Measure: are we talking about a symptom, a biomarker, a diagnosis or a clinical outcome?
- Contradiction: does a credible independent source present a different conclusion?
- Action: does the answer push you to start, stop or change treatment? If so, do not act without a doctor or pharmacist.
For blood test results, first compare the value with the original report, including its unit and that laboratory’s range. For a recommendation, trace it back to the HAS, Assurance Maladie, ANSM or a relevant medical society. A genuine PMID is not a universal seal of approval: you still need to check the methodology and applicability.
When to close ChatGPT and seek care quickly
Do not start by writing a prompt if rapid intervention may be needed. In France, call 15 for a medical emergency or 112. 114 is accessible, including by text message, chat or video, to people who are deaf, hard of hearing, deafblind or have aphasia.
Call immediately in the following situations:
- squeezing chest pain, feeling unwell or sudden breathlessness that could suggest a heart attack
- sudden weakness in an arm or leg, facial asymmetry, sudden problems with speech, vision or balance that could suggest a stroke
- sudden or rapidly worsening breathing difficulty
- loss of consciousness, seizures, significant bleeding, a severe allergic reaction or suspected poisoning
- overwhelming suicidal thoughts, intent or imminent risk of acting on them
For a new, persistent or worsening symptom, or in someone who is pregnant, a child, a vulnerable person or someone receiving treatment, seek professional advice even if there is no obvious emergency warning sign. ChatGPT can help you present a clearer account. It cannot guarantee that waiting is safe.
