Part 1 of 4 · By Elizabeth Beever
An educational blog series for the British Society of Lifestyle Medicine.
Series introduction
Patients are already asking AI tools about sleep, food, weight, stress and exercise. Lifestyle medicine needs a view on that, grounded in evidence rather than enthusiasm or alarm.
National policy points the same way. The Government’s 10 Year Health Plan for England sets out three shifts, from hospital to community, from analogue to digital and from sickness to prevention, and says it will make the NHS “the most AI-enabled care system in the world” (Department of Health and Social Care, 2025).
This four-part series explores what AI might contribute to prevention, from making health information more accessible to supporting behaviour change. It also examines the questions of safety, equity, evidence and accountability that need to be addressed before these tools can be trusted. The series focuses mainly on generative AI, meaning the chatbots and large language models people are already using, along with ambient voice tools that draft consultation notes.
The four articles cover:
- Article 1, Information is the easy part: how AI can make health information more accessible, and where it lets people down.
- Article 2, A good conversation is not behaviour change: what the evidence says about supporting behaviour change.
- Article 3, Who gets left behind?: whether these tools could narrow or widen health inequalities.
- Article 4, What good looks like: the questions to ask before recommending a tool, covering regulation, clinical safety and accountability.
References
Department of Health and Social Care (2025) 10 Year Health Plan for England: fit for the future. GOV.UK, 3 July (updated 30 July 2025). Available at: https://www.gov.uk/government/publications/10-year-health-plan-for-england-fit-for-the-future (Accessed: 1 November 2026).
Information is the easy part
AI can explain health information well. Whether it can do the harder things prevention needs is a separate question.
Picture someone leaving a ten-minute appointment with a leaflet on cholesterol. That evening they ask a chatbot to explain it again, more simply, and then ask what it might mean for Sunday lunch.
This is a real strength of AI, and it addresses something lifestyle medicine has always struggled with – explaining health information in language a particular person can understand and use. It is also the least demanding thing we could ask of the technology, and the case for it should be made on that basis.
What it does well
Someone preparing for an appointment can use a tool to turn a muddle of worries into three clear questions. A person on a tight budget can ask for meal ideas built around what is already in the cupboard. For people who feel awkward raising alcohol, weight or low mood in a consultation, exploring a subject privately may be valuable, and a conversational tool lets them revisit information at their own pace instead of remembering everything from a single appointment.
Clinicians can also use these tools to draft plain-language explanations and adapt patient information to a person’s language preferences. The bigger opportunity may be in the paperwork. Prevention is limited by time to listen, understand and build trust, and a good deal of that time goes on documentation. Ambient voice technology, which drafts consultation notes for the clinician to check, can reduce the time spent writing them. Most evaluations stop there, though, and it is not yet clear whether the minutes saved become more patient-facing care or are absorbed by an overloaded system. Early use can also add work, because every draft has to be checked (Shand and Morris, 2026). If the time does reach patients, AI may create more space for the conversations lifestyle medicine has always valued.
Where it lets people down
The trouble is that a fluent answer and a correct answer can look identical on the screen. A tool can state a supplement dose with total confidence and be wrong. It can present a contested finding in nutrition as settled, or leave out the one caveat that matters. A systematic review of 137 studies of chatbot health advice found that the way they were reported varied widely, which makes it hard to judge how reliable these tools are (Huo et al., 2025).
The pillars of lifestyle medicine show how much the risks can vary:
- Nutrition: Meal ideas can help someone plan affordable food. Advice for a person with kidney disease, an eating disorder or a complex medication regimen requires more care.
- Physical activity: Ideas for fitting movement into a working day carry little risk. Advice to increase exercise could be inappropriate for someone with undiagnosed exertional chest pain.
- Restorative sleep: General advice about routines may be useful. Loud snoring with daytime sleepiness could indicate sleep apnoea and needs assessment, not another sleep tip.
- Stress management: A tool may offer relaxation techniques but fail to recognise when someone is describing significant depression.
- Social connection: A tool can suggest ways to maintain contact, but it cannot provide the reciprocity of human relationships.
- Connection with nature: Ideas for time outdoors carry little risk, though what is realistic depends on where someone lives.
- Substances/behaviours: General information about alcohol or smoking can be useful. Telling someone who drinks heavily every day to stop could expose them to dangerous withdrawal.
A useful test follows from this. AI may be a reasonable source of general information when errors are unlikely to cause harm and the answer can be checked. It becomes a poor source as the stakes rise, or when an answer depends on a clinical picture the tool does not have. Patients and clinicians need to understand where that line falls, and the tools do not always make it clear.
Access is not the same as understanding
Making information easier to obtain does not make it easier to judge. People can be encouraged to ask where information comes from, look for uncertainty and bring anything that affects a health decision back to a healthcare professional. But responsibility for safe information cannot rest with the individual user. Developers and organisations need to demonstrate that their tools are reliable, communicate their limitations and provide routes to human support.
Information can also be technically correct and still unhelpful. Advice about cooking from scratch may not help a person without a kitchen, and suggestions to exercise outdoors may not suit someone with mobility difficulties or who feels unsafe in their neighbourhood.
AI could make health information more approachable and give people more confidence to ask questions. That is worth exploring, but it is not a clinical assessment and it cannot replace a conversation with a healthcare professional who understands the wider context.
References
Huo, B. et al. (2025) ‘Large language models for chatbot health advice studies: a systematic review’, JAMA Network Open, 8(2), e2457879. Available at: https://doi.org/10.1001/jamanetworkopen.2024.57879.
Shand, J. and Morris, S. (2026) Ambient voice technology in health care: what’s the evidence so far? Nuffield Trust, 4 February. Available at: https://www.nuffieldtrust.org.uk/news-item/ambient-voice-technology-in-health-care-what-s-the-evidence-so-far (Accessed: 1 November 2026).
About the author
My work focuses on digital health, responsible AI and healthcare innovation, with a particular interest in how emerging technologies can support prevention and more person-centred care. I recently completed the BiteLabs Digital Health, AI & Innovation Fellowship and am continuing to develop my knowledge of digital clinical safety. I am also committed to advancing AI literacy across the healthcare workforce, helping professionals develop the knowledge and confidence to engage critically and responsibly with emerging technologies.
I qualified from Newcastle University Dental School in 2013 and have worked across primary and secondary care in the UK and New Zealand. I am a member of the British Society of Lifestyle Medicine and work to promote its values and mission.
Please feel free to get in touch or connect with me on LinkedIn.
LinkedIn: https://uk.linkedin.com/in/elizabeth-beever
Website: https://elizabethbeever.com
