How can Patient-Facing Artificial Intelligence Reach People Facing Digital Barriers and Mistrust of Healthcare?

Bleik and colleagues’ pilot study of a human-like, non-generative artificial intelligence (AI) avatar for antidepressant education in primary care is a timely contribution to patient-facing digital health. It asks a practical question: can AI-delivered education be acceptable, credible and understandable when used alongside routine care? Participants’ responses were encouraging, with improved ratings for credibility, satisfaction and perceived understanding after iterative refinement. ¹

The study also highlights the central challenge for patient-facing AI. Most participants in the pilot study were highly educated and confident using digital tools. ¹ The harder test is whether such tools can reach people with limited or unreliable digital access, low digital confidence, limited health literacy, limited English proficiency, or mistrust of healthcare. Without deliberate attention to access and equity, digital interventions may disproportionately address needs of already advantaged groups and inadvertently widen inequalities. ²

Maternal immunisation is one area where this test matters. Vaccination against pertussis, influenza and respiratory syncytial virus (RSV) prevents serious illness in mothers and infants, yet uptake remains lower among some ethnic minorities and people living in more deprived areas.^3,4 In England, RSV vaccination among women who gave birth in November 2025 was 64.1%, but uptake was 53.4% in London and 34.7% among Black Caribbean women. ⁵ Pertussis coverage has improved, but substantial disparities remain: 47.1% among Black or Black British Caribbean women, and 65.6% in the most deprived decile of local areas, compared with 83.9% in the least deprived decile. ⁶ Influenza vaccine uptake in pregnancy also remains poor at 32.1% in 2023-24. ⁷ The consequences can be fatal: of 33 babies who died with confirmed pertussis in England between 2013 and June 2025, 27 were born to mothers who had not been vaccinated during pregnancy.^8,9

Barriers to maternal vaccination include limited knowledge of vaccine necessity, safety and effectiveness, as well as misinformation and mistrust.^3,4 Like Bleik and colleagues, who note that lengthy handouts may be ineffective, our patient and public involvement (PPI) group reported that vaccination leaflets can be too long, generic and poorly tailored to women’s concerns. ¹⁰ Many women prefer discussion with a trusted healthcare professional, but staff may lack the time, training or integrated records needed to provide detailed, consistent counselling. ¹¹

Conversational chatbots based on large language models may complement, but not replace, this care. Static leaflets and one-way videos are useful, but they cannot respond to a user’s specific questions, concerns or circumstances. A chatbot using retrieval from curated NHS-approved sources rather than unrestricted generation, developed with clinical and regulatory safeguards, could provide immediate, consistent, evidence-based answers from trusted sources, signpost local services and prompt escalation when clinician input is needed. Used well, it could make advice from midwives, general practitioners, pharmacists and obstetricians easier to reach.

Emerging evidence supports cautious optimism. In Argentina, a preregistered randomised trial involving 249,705 people found that a behaviourally informed chatbot more than tripled COVID-19 vaccine uptake compared with no message, although the absolute increase was 1.6% compared with the control group. ¹² In Hong Kong, a cluster-randomised trial among South Asian adults reported influenza vaccination uptake of 57.8% after a smartphone chatbot intervention versus 1.3% in controls immediately after the intervention, and 68.0% versus 2.0% at three months. ¹³

Our AI for Maternal Immunisation (AIMI) feasibility study is developing a co-designed, NHS-aligned chatbot with ethnically and socioeconomically diverse pregnant women and healthcare professionals. It will provide personalised information on vaccine benefits, risks, myths and access, and assess use, acceptability, trust, digital barriers, vaccination intention, chatbot interaction data, self-reported vaccination with medical record checks, recruitment, follow-up and staff views. Access to chatbot will be provided in different ways, including clinic quick response (QR) codes, email links and facilitated use through public computers in community centres or libraries. ¹⁴ These routes matter because patient-facing AI that works only for digitally confident users risks reproducing the inequalities it seeks to address. ²

Primary care and public health need standards for AI chatbots that go beyond satisfaction and engagement. Patient-facing systems should be evaluated as clinical and public health interventions, with attention to safety, accuracy, transparency, data governance, socioeconomic and ethnic inequalities, digital exclusion, workflow, workload and objective health outcomes.^15-17 For maternal immunisation, this means asking whether the chatbot is used by women with the lowest vaccine coverage, whether it improves trust and intention, and whether it supports rather than burdens staff.

Reaching women with greater mistrust will require more than accurate content. It will require co-design and co-production. NIHR defines public involvement as research carried out with or by members of the public, not to, about or for them; co-production goes further by sharing power and responsibility from the start to the end of a project.^18,19 Diverse PPI contributors should shape the questions a chatbot answers, its tone, languages and reading levels, including audio or translated formats where appropriate, the myths it addresses, its escalation points and the routes to vaccination it provides. For groups with good reason to mistrust health systems, trust cannot be designed by researchers alone; it must be built with the people whose trust is being sought.

Patient-facing AI should therefore be audited, transparent, co-designed and tested in the populations most likely to be left behind. Bleik and colleagues’ study is valuable because it places AI-supported education where care happens: language, trust, time, access and behaviour. The next step is to test whether interactive tools can improve decisions and outcomes where the need is greatest. Maternal immunisation is an ideal place to start.