You’re already asking AI about your health. Here’s how to ask better

Dr Suwanna Suwannaphong has been practising medicine for more than 20 years. A cardiologist and internal-medicine specialist, she is program director at VitalLife Scientific Wellness Center in Bangkok, Thailand, and also practises at Bumrungrad International Hospital, a leading private hospital in the Thai capital, with a particular focus on preventive medicine and preventive cardiology.

VitalLife brings a different kind of credibility to the conversation. Founded in 2001, the centre has spent more than two decades working in preventive and longevity medicine, with access to Bumrungrad International Hospital’s wider clinical infrastructure.

The question for doctors now is what to do with the information patients bring in. Some see AI as a useful way for people to educate themselves before an appointment. Others worry about patients arriving convinced by a diagnosis or treatment recommendation that may be based on incomplete information, weak evidence or an answer delivered with more certainty than medicine allows.

Suwannaphong sits firmly in the first camp—with conditions. She isn’t telling patients to stop using LLMs. What she wants them to do is use them better, and her first question is a good place to start: “how do you prompt the AI?”

See also: Inside Asia’s biohacking landscape: the science, trends and scientists to know

Suwannaphong’s first recommendation is to tell the LLM what kinds of sources you want it to use.

If you ask whether a treatment works, it may draw from research papers, news articles, clinic websites and marketing material. Her advice is to narrow the field: ask for original research and published studies, and ask how closely that research relates to the question you are asking.

“Any LLM with web access will be able to find the ads, commercials or any content,” she says. “If you prompt the LLM to eliminate those and go into the original research, that would narrow down a lot of things.”

Then ask what kind of evidence you are looking at. Was the research done in humans? How many people were involved? Are there several studies pointing in the same direction, or one small study that has been repeated online until it sounds established?

Not all LLMs perform at the same level.

For health research, Suwannaphong recommends using a more capable model if you have access to one. In practice, that may mean a paid version rather than the free model.

The difference can show up in how well the model searches, handles longer or more complicated questions and works through research.

Her point is simple: if you are using AI to investigate something as important as your health, the quality of the tool matters too.

The more general the question, the more general the answer.

Suwannaphong gives a simple example: research relevant to a 45-year-old Southeast Asian woman may not tell you much about a 35-year-old German man.

Age, sex, ethnicity, medical history and other medications can all change how useful a piece of research is.

An LLM can help search for smaller, more specific studies that might otherwise be difficult to find. That does not make the answer personalised medical advice, but it can make the research you bring into a consultation more relevant.

One instruction Suwannaphong recommends is particularly useful: tell the AI to say when it doesn’t know.

LLMs have a tendency to keep answering. In health, that can be a problem.

Ask it to separate what is well established from what is early, conflicting or based on limited evidence. Ask whether there are important gaps in the research. And if there is not enough information to answer the question properly, ask it to say so.

This is where Suwannaphong thinks AI can resemble some wellness influencers: both can sound extremely sure.

A doctor may sound more cautious because medicine involves trade-offs, contraindications and incomplete information. “There is no hundred per cent in medicine,” she says. “We tend to believe in AI” because of the confidence with which it presents an answer.

That confidence should never be the reason you trust it.

Suwannaphong is not sceptical of AI itself. She believes doctors will increasingly need it to keep up with the volume of medical research. For now, though, the most useful role for a general LLM may be much simpler.

Ask it to help you understand the evidence. Ask it to show you what may apply to you. Ask it what it is unsure about.

Then take those questions to someone who can see the rest of the picture.

NOW READ

Everyone wants natural beauty, but what does ‘natural-looking’ really mean?

Has self-care become too much work?

The unspectacular habits high-achieving women protect to keep them performing at their best