When Asa greets a patient on WhatsApp with "Let's book you in," it might feel like helpful urgency to a health administrator—but to the person on the other side of the screen, it can feel like pressure. That distinction, small as it seems, reveals something crucial about the future of health care: how AI communicates matters as much as what it does.
Researchers at the University of Surrey studied how patients in North London responded to Asa, a generative AI receptionist designed to help people book cervical screening appointments. Through interviews with patients at a local GP surgery and a survey of 300 people eligible for NHS cervical screening, Dr. Doris Dippold and her team analyzed what builds genuine rapport between patients and health chatbots—and what destroys it.
The findings, published in the journal Lingua, reveal a paradox at the heart of health AI design. Patients responded warmly to Asa's friendliness and the choice it offered them. They appreciated its female-presenting voice, which some found easier for disclosing sensitive information—like needing to cancel an appointment because of menstruation—compared with speaking to a male receptionist. The chatbot's integration into their existing routines felt natural and helpful.
But follow-up messages sent within 24 hours felt intrusive. Imperative phrasing—"Let's book you in"—landed as aggressive rather than supportive. And for patients managing mental health challenges, neurodivergent conditions, or heavy caring responsibilities, the pressure to respond quickly felt unfair and burdensome.
The deepest friction emerged around trust. Many patients expressed worries about data security, impersonation, and the blurring of human and AI boundaries. When Asa stated that users could "chat to me as if I am a real person," it backfired. Rather than reassuring, patients read it as suspicious—even deceptive.
Dr. Dippold observed that this points to a fundamental misunderstanding about anthropomorphism in health care. "Our analysis shows that anthropomorphism is not universally positive," she noted. "Human-like features can build rapport—but when they clash with patients' expectations for transparency in a health care setting, they undermine exactly the trust the chatbot is trying to build."
This research arrives at a critical moment. Cervical screening uptake across the U.K. fell 5.3% in 2023–24, with ethnic minority groups consistently underrepresented in screening programs. The GP surgery where Asa was tested serves a highly diverse, socioeconomically deprived community in Islington. Here, accessible and equitable communication isn't a nice-to-have feature—it's the foundation of whether people engage with preventive health care at all.
The study suggests that health care chatbots should prioritize transparency about how they work, give patients genuine control over decisions, and treat them with consistent respect. Crucially, the research frames emotional support not as a luxury enhancement but as essential infrastructure. "Feeling seen, appreciated and emotionally supported is not a luxury feature in health AI," Dr. Dippold said. "If patients disengage because a chatbot feels pushy or untrustworthy, the health service loses them entirely."
As NHS services increasingly turn to AI to manage scheduling and outreach, this work offers a roadmap for building technology that actually serves the communities it's meant to help—by listening to what those communities say they need.