This study was approved by the Ethics Committee of Peking University Third Hospital (approval number M20241044). All participants provided informed consent prior to participation. All information collected during the study was treated as strictly confidential. Interview data were anonymized before analysis and stored in NVivo (version 15; Lumivero) on secure, password-protected institutional servers, with access restricted to authorized members of the research team.

In this study, AI-driven OMC services refer to OMC services in which AI technologies are used to support medication-related information provision, preliminary guidance, triage, or decision support. These services may include fully automated AI responses as well as AI-assisted models involving review by health care professionals. As this remains an emerging service model, the study explored participants’ perceptions of this broader service category rather than evaluating a single standardized platform. This study adopted a descriptive qualitative research design to comprehensively explore the factors influencing patients’ acceptance of and WTP for AI-driven OMC services. In-depth semistructured interviews were conducted to collect detailed information from participants.

Purposive sampling was used to recruit participants with diverse characteristics, including sex, age, geographic location, identity characteristics, prior OMC experience, and educational level. The research team approached potential participants from different cities in China through clinical, professional, and personal networks. Some participants also referred other eligible individuals to the research team. Recruitment continued until the sample included patients, caregivers, and health care professionals with varied backgrounds and experiences related to OMC services.

Eligible participants were individuals aged 18 years or older who were able to provide informed consent and participate in a Chinese-language interview. Patients, caregivers, and health care professionals were included to capture diverse perspectives on the factors influencing acceptance of and WTP for AI-driven OMC services. Participants were excluded if they were unable to provide informed consent, unable to complete the interview, or unwilling to be audio-recorded.

In-depth semistructured interviews were conducted in Chinese either face-to-face or remotely via telephone in a quiet and private setting between December 2024 and January 2025. The sample size was determined based on data saturation, with interviews discontinued when no new information emerged [21].

Before the interviews, participants were informed of the study’s purpose and provided written informed consent. Each interview lasted for 30-45 minutes and was audio-recorded with AI-assisted real-time transcription using a speech-to-text device. All transcripts were checked against the original audio recordings and verified against the recordings within 24 hours to ensure transcription accuracy. All interviews were led by QC, while SD served as the note-taker. Both researchers had backgrounds in pharmacy and qualitative research and were familiarized with the interview guide before data collection. A semistructured interview outline designed based on relevant literature and participant characteristics [22] was pilot tested with 3 volunteers and refined before the formal interviews. The outline consisted of 5 primary questions and was detailed in Textbox 1. Participants were not shown a standardized platform, workflow, use case, or pricing scenario before the interviews. Instead, the interviews used open-ended questions to explore participants’ perceptions, concerns, expectations, and WTP regarding AI-driven OMC services as a broad and emerging service model. Interviews began with open-ended prompts (eg, “what” and “how”) to elicit detailed responses, followed by probing questions for deeper exploration of emerging themes [23,24]. Plain and accessible language was used to encourage open communication. SD documented contextual information to complement the audio-recorded data, including facial expressions and body language in face-to-face interviews and tone, pauses, and emotional reactions in telephone interviews. Participants could withdraw at any time or refuse to answer any question.

The data analysis process was conducted concurrently with data collection, and further sampling decisions and interview questions were guided by the findings of the ongoing data analysis [24]. Interviews were conducted until ongoing coding indicated that no new codes or subthemes relevant to the study objectives emerged [21].

Transcripts were sequentially numbered and imported into NVivo (version 15) for coding. To minimize potential bias and enhance the credibility and consistency of the analysis, 2 researchers independently coded and analyzed the data, followed by cross-checking and discussion to resolve discrepancies. An initial codebook was developed through comparison of independently generated codes and was refined iteratively during the analysis.

The analysis followed Clarke and Braun’s 6-phase thematic analysis approach [25,26]. First, 2 researchers (SD and QC) independently read the transcripts to become familiar with the data. Second, each researcher independently generated initial codes across the dataset, focusing on participants’ perceptions of AI-driven OMC services without applying a fixed coding template. Third, the researchers independently organized related codes into preliminary themes. Initial coding remained open to concepts emerging from the data, while the theories of perceived risk and perceived benefit were used as a sensitizing framework to support theme organization and interpretation. Although the final themes were organized under the broad dimensions of perceived risk and perceived benefit, the subthemes were not predefined by these theories. Instead, they were developed and named by the research team based on recurring patterns in participants’ responses. The 2 sets of codes and preliminary themes were then compared. Differences in coding interpretations, theme boundaries, and theme names were discussed until consensus was reached. Where uncertainty remained, coding and thematic decisions were discussed with the wider research team. Analytic notes were used to document coding decisions, emerging interpretations, and points requiring further discussion. Fourth, the agreed themes were reviewed and refined to ensure internal consistency and clear boundaries. Fifth, the themes and subthemes were refined and named based on the coded data within the broader risk-benefit framework. Finally, the results of the analysis were compiled and presented in a structured report to support the interpretation of WTP and service acceptance.

The final set of codes, themes, and subthemes was jointly reviewed and confirmed by all team members. Representative quotations were selected from the Chinese transcripts to illustrate each theme, translated into English by bilingual researchers, and checked against the original text to preserve participants’ intended meanings.

A total of 20 participants were included in this study, comprising 5 male and 15 female participants, aged 22-68 years. Of the 20 interviews, 8 were conducted face-to-face and 12 were conducted remotely via telephone. The sample comprised 15 patients, 3 caregivers, and 2 health care professionals. In terms of geographic distribution, 11 participants lived in urban areas and 9 in rural areas. Regarding educational background, 5 participants held a master’s degree or higher, 9 held a bachelor’s degree, and 6 had completed high school education or below. Detailed participant characteristics are shown in Table 1.

The majority of participants assumed that AI-driven services should be free. Some believed that AI-driven online consultation services could be priced according to their level of human involvement. For example, fully automated AI-driven services that rely solely on algorithmic responses were expected to remain free or low-cost, as they primarily aggregate publicly available medical information. In contrast, AI-assisted services involving health care professionals could be charged, as expert validation and personalized advice make a paid model more justifiable. A tiered pricing model could balance accessibility and quality, offering free basic AI consultations while charging for expert-reviewed recommendations. Participants who preferred traditional offline consultations reported lower WTP, whereas those satisfied with prior OMC experiences showed higher WTP for such services. In summary, participants reported WTP values ranging from RMB 0 to RMB 200 (US $0-$27.28; RMB 1=US $0.1364 as of January 15, 2025), with some indicating a higher WTP for services reviewed or validated by senior professionals or tertiary hospitals.

This qualitative study explored perceived barriers, facilitators, expectations, and factors shaping public acceptance of and stated WTP for AI-driven OMC services. Through in-depth interviews with 20 participants from diverse backgrounds, perceived risks and perceived benefits were identified as 2 key dimensions influencing participants’ acceptance and WTP for these services. Regarding perceived risks, psychological, governance, social, functional, health, and financial risks were found to negatively affect acceptance, serving as potential barriers that may discourage engagement with AI-driven OMC services. For perceived benefits, convenience, diversity, reliability, efficiency, and educational benefits emerged as key motivating factors encouraging acceptance. Participants reported WTP values ranging from RMB 0 to RMB 200 (US $0-$27.28; RMB 1=US $0.1364 as of January 15, 2025), with participants’ views appearing to vary by their prior OMC experience and understanding of the service model. These findings provide valuable insights for the development and optimization of AI-driven OMC services, supporting targeted pricing strategies and personalized service design that address public preferences and concerns.

The theory of perceived risk provides a valuable framework for interpreting participants’ hesitancy toward AI-driven OMC services [27]. As novel health care technologies often face initial resistance due to unfamiliarity [28], psychological risk was reflected in participants’ lack of familiarity with AI-driven platforms, leading to skepticism about reliability and a preference for face-to-face interactions. The preference for in-person consultations reflects not only technological hesitancy but also established expectations regarding health care delivery models [29]. Governance risks emerged as major concerns, with participants expressing fears of data leakage and inadequate regulatory oversight. These findings align with previous studies that highlighted concerns about ethical dilemmas, unclear regulations, and data privacy in health care–related AI applications [30,31]. Lack of transparency remains a prevalent issue across most existing AI technologies [32]. Social risk was evident in participants’ discomfort with non–face-to-face communication, which they felt hindered the expression of complex health information. Deficiencies in empathetic responses and human-centered care make AI interactions less acceptable to some users [33]. Functional risks were associated with both clinical and technical limitations of the platforms, including the inability to prescribe medications, conduct tests, or deliver highly personalized advice, which reduced the perceived value of the service. Beyond clinical functionality, usability appears to be an important factor shaping trust and acceptance, as technical stability, ease of navigation, and platform reliability influence whether users feel confident relying on AI-driven OMC services. AI-driven consultations were considered more reliable for managing straightforward tasks and therefore more acceptable [34], although technical failures may compromise reliability and erode trust among health care professionals [32,35]. Health risks centered on urgent care availability and potential misinformation, while financial risks reflected uncertainty about whether AI-driven services would ultimately reduce or increase health care costs. These multifaceted risks collectively present substantial barriers to the adoption of and WTP for AI-driven OMC services, particularly among participants without prior OMC experience.

Perceived benefits emerged as key motivating factors for WTP for AI-driven OMC services. Importantly, these benefits refer to participants’ perceived benefits rather than demonstrated performance of the AI-driven OMC model, as this study did not evaluate a standardized platform or clinically validate the accuracy, safety, or reliability of medication advice. OMC may extend health care to marginalized populations through remote consultations, monitoring, and technology-assisted diagnosis [36], and AI is expected to provide more real-time predictions of patient outcomes in the future [37]. AI has expanded access to integrated health care by enhancing provider communication, facilitating interactions, and serving as a knowledge hub [38,39]. Such flexibility aligns with increasing consumer demand for on-demand health care [40,41]. Service diversity, including multiple communication modes and language support, created a more inclusive health care experience appealing to diverse user preferences and communication styles. Reliability was attributed to AI’s ability to process and integrate large datasets [42], although hybrid models with human oversight were considered more trustworthy. While user-friendly AI platforms are essential for implementation, integrating them into existing health care systems remains challenging and often requires substantial changes to established operational workflows [43]. Potential cost-effectiveness was perceived as another benefit, with participants expecting time or cost savings compared with traditional consultations. Moreover, participants perceived educational value in these services, particularly for improving medication knowledge, preparing for subsequent consultations, and supporting communication with health care professionals. However, whether these perceived benefits translate into improved clinical outcomes requires further evaluation.

Regarding WTP, our findings reveal that participants’ WTP was shaped by how they understood the service model and the level of human involvement. Many participants believed that fully automated AI-driven OMC services should be free or low-cost, whereas services involving review by health care professionals were considered more acceptable as paid services. These findings align with the broader digital health literature showing that WTP is highly context-dependent and varies by service modality, user characteristics, and the perceived value of the service. A systematic review of eHealth WTP studies identified substantial heterogeneity across studies: one-time WTP ranging from US $0.88 to US $191.84 and monthly WTP ranging from US $5.25 to US $45.64, with WTP differing by population characteristics and delivery modality [44]. Another study of primary care web portals similarly suggested that patients may be willing to pay modest fees for online services that offer direct practical value, such as physician messaging, medical record access, and medication refills [45]. More recent evidence from health apps in Hong Kong also showed that WTP was associated with prior app experience and perceived benefit, while unwillingness to pay was commonly linked to expectations that such services should be free, distrust, and limited awareness of benefits [46]. In line with previous studies, our participants appeared to attach value less to AI involvement itself than to clinically meaningful service attributes, particularly professional oversight, trustworthiness, convenience, and affordability. Since WTP was elicited through open-ended qualitative questions and participants discussed different service models, including AI-only and professionally reviewed services, this range should be interpreted as exploratory evidence of perceived value rather than as a pricing estimate. As highlighted by a previous study on cost-aware frameworks, health care AI development expenses—including data acquisition, computational resources, and developer labor—are frequently underestimated in conventional cost assessments [47]. Rather than supporting a specific payment model, our findings suggest that future payment models for AI-driven OMC services need to balance affordability, perceived value, professional oversight, and equitable access. The strategic design of AI payment models is therefore crucial to enhance WTP while optimizing financial efficiency and ensuring equitable access across populations [48]. In the future, insurance companies or health networks may leverage reimbursement policies and out-of-pocket costs to influence or restrict patients’ provider choices as part of cost-effectiveness–based payment strategies [49]. Prior OMC experience may have shaped how some participants described service value and payment conditions; however, this study did not formally compare WTP across experience groups. The experience-related observations, which refer to prior OMC experience with or without the involvement of AI, should be interpreted as exploratory and hypothesis-generating rather than as evidence of subgroup differences in WTP. Some participants with prior OMC experience provided more concrete descriptions of payment conditions, whereas participants without such experience more often expressed uncertainty or expected AI-only services to be free. Variations in preferred service models may also reflect personal characteristics, prior digital health experiences, perceived needs, and expectations [50]. These findings suggest that trial periods or freemium models may be worth further exploration to build initial trust, although future quantitative or experimental studies using standardized service descriptions and payment scenarios are needed before drawing conclusions about pricing models.

By identifying participants’ perceived barriers, facilitators, expectations, and payment considerations, our findings provide user-centered implications for the design, implementation, and future evaluation of AI-driven OMC services. This study used a qualitative approach, informed by the theories of perceived risk and perceived benefit, to identify perceived barriers, facilitators, expectations, and factors shaping public acceptance of and stated WTP for AI-driven OMC services. These findings suggest that service design may need to consider not only technical performance but also usability, trust, privacy protection, professional oversight, and affordability from the user perspective. For policymakers, establishing clear regulatory frameworks for AI-driven OMC services could help address privacy concerns and build public trust. Regulations should balance innovation with consumer protection. For telemedicine developers, our results highlight the importance of intuitive, user-friendly interfaces to overcome psychological barriers associated with unfamiliarity. Integrating AI-generated recommendations with human oversight could mitigate concerns about accuracy, personalization, and accountability. From a payment perspective, tiered service models that distinguish between basic AI-only support and professionally reviewed services may be worth further exploration, although future quantitative studies are needed to evaluate WTP and pricing models more precisely.

Several limitations should be considered when interpreting our findings. First, the sample size of 20 participants, while appropriate for qualitative research, limits the transferability of the results to broader populations and settings. Although patients, caregivers, and health care professionals were included to capture different stakeholder perspectives, the small numbers of caregivers and health care professionals limited the ability to compare views across groups or determine whether saturation was achieved within each stakeholder group. Therefore, subgroup-related findings should be interpreted as exploratory rather than representative of each stakeholder group.

Second, the study was conducted during a period when AI-driven OMC services were in the early stages of adoption in China, suggesting that perceptions may evolve as these services become more mainstream. Third, the focus on stated rather than revealed WTP may not fully capture actual payment behaviors in real-world settings. In addition, WTP was explored through open-ended qualitative questions rather than a formal contingent valuation approach, and participants were not asked to evaluate one standardized service model or payment scenario. Therefore, the reported WTP values should be interpreted as exploratory observations of perceived value rather than pricing estimates. Participants’ descriptions of AI as efficient, comprehensive, or reliable reflected perceived or expected attributes rather than empirically tested properties of a specific AI-driven OMC platform. Future studies should evaluate the clinical accuracy, safety, and reliability of AI-generated medical advice under standardized service scenarios. Fourth, because the interviews were conducted in Chinese and selected quotations were translated into English for reporting, some linguistic nuances may have been affected during translation, although translated quotations were checked against the original transcripts to preserve participants’ intended meanings. Although some findings may be relevant to countries with similar health care contexts, such as uneven medical resource distribution, growing telemedicine use, and early-stage AI integration, transferability should be considered cautiously. Health care financing, regulation, pharmacist roles, and public trust in AI may differ across settings. Future research should address these limitations through larger quantitative studies, cross-cultural comparisons, and experimental designs measuring actual payment behaviors rather than stated intentions.

Through semistructured in-depth interviews with 20 participants from diverse backgrounds, perceived risks (psychological, governance, social, functional, health, and financial risks) and perceived benefits (convenience, diversity, reliability, efficiency, and educational benefits) were identified as 2 key dimensions associated with participants’ acceptance and WTP for AI-driven OMC services. These findings provide exploratory insights into participants’ concerns, perceived value, and expectations, which may inform future service design and further research on payment models for AI-driven OMC services.