Using an AI chatbot for health questions
An AI chatbot can help explain an unfamiliar term or prepare questions for an appointment. Its answer can still be incomplete, outdated or wrong, including when it sounds confident. Use reliable health guidance to check information that matters, and contact a healthcare professional for symptoms, treatment decisions or interpretation of your own results.
How a chatbot produces an answer
A large language model generates text from patterns learned during training and information provided in the conversation. Some systems also use search or other tools. These can improve an answer, but do not guarantee that its medical information or interpretation is correct.
The model has read many descriptions of chest pain, but it has not listened to your chest, taken your blood pressure, or looked at last year's notes. It does not know your age, medicines or family history unless you type them in, and it cannot check that what you typed is accurate.
A clinical assessment can bring together a detailed history, examination, test results and follow-up. A consumer chatbot cannot provide the same assessment or responsibility for care. Treat its output as information to check, rather than a diagnosis.
Tasks a chatbot can help with
A chatbot can assist with these tasks, provided you check important information against reliable sources.
- Explaining a term. A chatbot may help explain words such as HbA1c or LDL in a report. Confirm the explanation with a reliable source, and leave interpretation of your own result to the care team.
- Preparing for an appointment. Polyclinic consultations are short. Asking the chatbot to help draft a list of questions about a new medicine or a borderline result makes the ten minutes count for more.
- Understanding a report after the visit. A result explained in the clinic is easy to forget by the time the bus arrives. Typing the numbers in and asking what each line measures can fill the gaps.
- General health information. A chatbot may help explain concepts such as fibre or moderate exercise, but even familiar topics can contain mistakes. Check important statements against sources such as HealthHub, and open any references to make sure they support the answer.
In each case the chatbot is organising existing information rather than deciding what is wrong with a particular person.
Common limitations and errors
The failures are predictable once you understand the mechanism.
- Confident errors. The model writes in the same assured tone whether it is right or wrong. It can invent a drug interaction, misstate a dose, or cite a guideline that does not exist, without any change in voice. Nothing in the text itself signals the mistake.
- No examination. A rash, a lump, a swollen joint and a sore throat all look different in person. The chatbot has only what you noticed and chose to write down.
- No history. Two people with the same symptom can need different care depending on age, medicines and past illness. Most people do not type their full medical history.
- The common answer wins. Because the model produces the most likely text, it leans towards the most common explanation. Uncommon conditions, and common conditions that present in an unusual way, are where that habit misleads.
- It can reassure too easily. A polite question often draws a polite, reassuring reply, and reassurance from a program that cannot examine anyone is worth little.
The useful distinction is between explaining information and making a clinical decision. A chatbot may help with the first, but mistakes can occur in both. A specialised clinical AI system used within a healthcare service has a different purpose and oversight from an open consumer conversation.
Asking useful questions and protecting privacy
A few habits make a chatbot more useful and less risky.
- Describe rather than label. "I have had a dull pain under the right ribs for three days after meals" gives the model more to work with than "do I have gallstones".
- Ask it to draft questions for the doctor. A list of five things to ask at the polyclinic is a safer output than a verdict.
- Ask it to show the uncertainty. Prompting "what else could this be, and what would make each more or less likely" produces a more balanced list than a single answer.
- Check anything that matters. Confirm a dose, an interaction or a threshold on HealthHub or with a pharmacist before you act on it.
- Never change a medicine on its say-so. Stopping or adjusting a prescription is a conversation with the prescribing doctor.
Many chatbots process conversations on remote servers. Storage, training use and deletion rules vary by provider, account and settings. Leave out names, identification numbers and other identifying details, and avoid uploading a complete report unless you understand the service’s privacy arrangements. Singapore data-protection rules may apply, but their application depends on the organisation and circumstances.
When to seek care immediately
Call 995 for possible emergencies such as chest pain with sweating or breathlessness, stroke symptoms, severe breathing difficulty, uncontrolled bleeding or collapse. For suicidal thoughts, call 995 if you are in immediate danger. Otherwise talk to a person now: Samaritans of Singapore answers 1767 around the clock, and mindline answers 1771.
- Chest pain or pressure, especially with sweating, breathlessness, or pain spreading to the arm or jaw.
- A drooping face, weakness in one arm, or slurred speech, which are the signs of a stroke.
- A sudden severe headache, confusion, a seizure, or fainting.
- Difficulty breathing, lips turning blue, or swelling of the face and throat after food or a sting.
- Heavy bleeding that does not stop, or a serious injury.
- Thoughts of harming oneself. Samaritans of Singapore answers 1767 at any hour, Mindline.sg offers support online and on 1771, and an emergency department is open around the clock.
Short of an emergency, some situations still need an appointment rather than a chat. Persistent or worsening symptoms, unexplained weight loss, a new lump, unexpected bleeding or a change in bowel habit need assessment. Do not wait to see whether concerning symptoms settle on their own. Fever in a young baby, or fever with severe illness or confusion at any age, needs prompt medical advice.
Using AI around an appointment
A chatbot may help explain information and prepare for a consultation, but it can also make errors. Use it as an aid to a conversation with a healthcare professional, and leave diagnosing symptoms or changing treatment to them.
The sensible pattern is to use it before and after a visit, and to keep identifying details out of the conversation. If you have a symptom that is new, worsening or lasting, book a polyclinic or GP appointment, and if you have a red-flag symptom, call 995. Do not start, stop or change a medicine on the strength of a chatbot answer.
A practical next step
Ask for a plain-language explanation or a list of questions to bring to your clinician. Check the sources yourself and leave identifying details out. Do not use a chatbot to decide whether urgent symptoms can wait.
Sources and further reading
These references explain the guidance and research discussed above. For advice about your own health, speak to a qualified healthcare professional.
Read next
AI in Singapore's clinics: what is already in use
Where artificial intelligence sits in the system today
What a smartwatch actually measures, and how much to trust it
A watch reads a few signals well and estimates the rest