Slow down and think before asking health questions to an AI chatbot

They might sound like a person, but they work like a machine

This guide is for anyone who might use an online artificial intelligence (AI) chatbot to ask questions about their health.

AI chatbots are things like ChatGPT, Google Gemini, and Microsoft Copilot. These are computer systems that have been 'trained’ with lots of information and data. This might be from the internet, books, published research articles, and many other sources. When we say that AI has been ‘trained’, we mean that it has learned the patterns between words in these data. It uses these patterns to create its responses.

The technical name for an AI chatbot is ‘large language model’, or LLM. Read more about how LLMs work.1

You may have already used an AI chatbot without realising it. If you use Google to search the internet, you may have noticed an 'AI overview’ at the top of the page. This uses the same technology as the Google Gemini chatbot. If you use a smartphone, it might have a built-in AI assistant which works like a chatbot.

Most online AI chatbots have websites. For example:

There are lots of other chatbots too, including Claude, Grok, DeepSeek, Vibe, Perplexity, Meta AI, and more. This guide does not endorse any particular AI chatbot.

You can use AI chatbots by visiting their website, or by downloading their app onto your smartphone or other personal device.

You can interact with AI chatbots by typing messages, using your voice, or uploading files or photographs.

You can use many AI chatbots for free. In most cases you need to register for an account first. If you want to use the latest and most powerful versions of the chatbot, you may need to pay a subscription fee.

AI chatbots and humans are not the same. Humans have brains, and we can think, learn and reason. AI is very good at spotting patterns, but it does not work in the same way as humans. When an AI chatbot writes a message, all it is doing is predicting what word might come next in a sequence. It does this by breaking down sentences into smaller chunks called ‘tokens’. In many cases, tokens are individual words.

For example:

Bears are...

... large

... mammals

... hairy

... dangerous

... powerful

Any of these options would make sense. The chatbot might be more likely to pick a particular option based on what else you have asked, or what the company who made it prioritises.

... green

... is

... flying

... !

... birds

On the other hand, these options would not make much sense.

Chatbots choose the word that fits the pattern best. In many cases, this means that the final sentence is true. But not always. Sometimes the chatbot will pick a word which fits the pattern, but the final sentence is not correct.

The newest types of AI chatbots have other features, where they can check what they have written makes sense before showing it to you. Even though this makes them more accurate, they can still be wrong.

Lots of people worldwide have started using AI chatbots for all sorts of things.

Some people use them for work – for example, to help write letters, or shorten complex reports. Other people use them in their personal life – for example, to plan a holiday, or find a recipe for dinner.

AI chatbots are often called 'general-purpose'. This means they can be useful for lots of different things. Some companies have now launched chatbots made specifically for health questions. Some examples of these include ChatGPT Health, Claude for Healthcare, and Copilot Health. Being built for health questions does not automatically make a chatbot safe or accurate.

Lots of people have started asking AI chatbots for help with their health. Researchers at King’s College London found that about 1 in 7 adults in the UK have used an AI chatbot instead of contacting their GP or other NHS service.2

People use AI chatbots for different things when it comes to their health. Below are some examples based on research published by Microsoft:3

  • Helping to understand health problems or diseases
  • Asking about symptoms they have
  • Getting fitter and more active
  • Help with emotions and mental health
  • Completing forms and other health paperwork

Convenience was the most common reason people said they use AI chatbots for health. More than half of those who use them said it had been good for their physical and/or mental health.4

If two different people ask an AI chatbot the same question, they may get very different replies. There are a few reasons for this:

  • Chatbots don't give the exact same answer every time because there is a bit of chance built into how they pick their words. Some chatbots can also search the internet for answers, but they might not always look at the same websites. This means that even the same person asking twice may not get the exact same reply.
  • Different people might use different chatbots, or they might use different versions of the same chatbot.
  • Many chatbots have ‘memory’, which stores things you told them before, and feeds this into later answers. Over time, chatbots may adapt their responses in ways that you seem to prefer.5 Read more about memory.

This means that your experience of using an AI chatbot might be very different to someone else's. You shouldn't assume that you're getting the same answers as other people.

In addition, different chatbots work in different ways. Each has strengths and weaknesses. The companies that make them also have different priorities, which change the way they reply.

AI chatbots are not perfect, and they can get things wrong. This means that they could give you information which is not true. If they encourage you to change the way you live, or give you incorrect advice about the medications you take or any health problems you have, this could cause you harm.6

Researchers are particularly worried about the risks of people using AI chatbots when they are experiencing a mental health crisis. There is evidence that they could worsen some people's condition, rather than helping.7

Recommendation:

If you are feeling low or struggling to cope, you can call Samaritans free on 116 123 or text SHOUT to 85258 at any time of day or night. If you or someone else is in immediate danger, or you feel unable to keep yourself or someone else safe, call 999 or go to A&E as soon as possible. If you are not sure how urgent it is, call NHS 111, which is free and open 24 hours a day.

If you are outside of the UK, you can use this website to find help:
https://findahelpline.com

Researchers at the University of Oxford tested this.8 They gave about 1300 people a set of realistic medical scenarios. They asked them to try and work out what was wrong with the patient in the scenario.

Some people were allowed to use AI chatbots for this. Others were just allowed to search the internet or use their own experience.

The researchers found that the people who used AI chatbots were no better at identifying what was wrong with the patient, or what type of help they needed.

When they looked into why this happened, they found that people often gave incomplete information to AI chatbots, and the chatbots often misinterpreted what they said. Even when the chatbot gave the right answer, people sometimes didn't agree with it.

Many health questions will be best answered by a health professional. Depending on the question, this might be a doctor, nurse, pharmacist, dentist, physiotherapist or someone else.

For some questions though, we might be able to get a good answer by ourselves if we use a chatbot carefully.

They are available 24/7 via your computer or smartphone – no appointment needed. They can access lots of information including published scientific research. They don’t get hungry, tired or stressed. If we use them well, they can empower us to improve our health and wellbeing.

We have built this guide to help you get the most from these tools. If you Ask SMART, you can:

  • Write clear questions that get you better responses,
  • Think about how much personal information you share,
  • Check if the information you’re getting is true,
  • Know when it’s best to speak to a human health professional instead.