Utilisation of AI-driven chatbots for perioperative health information seeking: a descriptive qualitative study of orthopaedic patients and family members

Abstract

Objective

This study aimed to explore orthopaedic patients’ and families’ experiences with artificial intelligence (AI)-driven chatbots for perioperative health information, focusing on usability, effectiveness and perceptions.

Design

A descriptive qualitative design was employed.

Setting

This study was conducted at a regional care centre for orthopaedics.

Participants

We recruited 13 participants (patients undergoing orthopaedic surgeries and family members) through purposive sampling. Face-to-face semistructured interviews were conducted to capture participants’ experiences and insights. Data collection was concluded when data saturation was achieved. All interviews were audio recorded and transcribed verbatim within 24 hours. Transcripts were verified and analysed using the Colaizzi’s data analysis method.

Results

Four themes emerged from interviews, including: (1) preference of AI chatbots over search engines; (2) improved accessibility and quality of information; (3) preference of AI over human interactions and (4) importance of effective prompting.

Conclusions

AI-driven chatbots offer a promising adjunct to perioperative patient education by delivering immediate, tailored guidance that overcomes the limitations of conventional search engines and busy clinical settings. Study participants valued chatbots’ efficient, context-sensitive retrieval, professional-level advice and non-judgmental interactions, which fostered trust and reduced anxiety. Effective prompting emerged as a key user skill, directly shaping response relevance and accuracy. Chatbot-generated health information should be regularly reviewed for accuracy. Structured tutorials may be offered for user capacity building.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The study employs a descriptive qualitative approach well-suited for capturing rich, detailed insights into the real-world experiences of patients and families using artificial intelligence (AI)-driven chatbots.

• The use of purposive sampling ensures the inclusion of participants with relevant experiences and provides a focused and indepth exploration of AI chatbot use in perioperative care.

• The study’s focus on Chinese participants may limit the applicability of findings to other cultural or linguistic contexts.

Introduction

Perioperative health information is critical for orthopaedic patients and their families to navigate the surgical journey effectively. It helps reduce anxiety, improve preparedness and enhance recovery outcomes.^{1 2} However, patients and families often encounter challenges in accessing accurate, reliable and timely information during this period. Traditional sources, such as healthcare professionals or printed materials, may not always be readily available or tailored to individual needs, leaving information gaps that can negatively impact decision-making and overall experience.

AI-driven chatbots are transforming the healthcare landscape by leveraging natural language processing and machine learning to provide conversational, user-friendly interfaces for information and support.^{3 4} These chatbots are increasingly adopted to address gaps in patient education and healthcare delivery, offering personalised, evidence-based and easily accessible information tailored to individual needs.5,7 By providing 24/7 availability and the ability to interact dynamically, AI chatbots can help patients better understand their conditions, manage symptoms and navigate care pathways.

Studies have demonstrated that AI chatbots, such as ChatGPT, can rapidly generate educational materials for complex medical conditions, streamlining content creation and making information more accessible, especially in resource-limited settings.⁸ Comparative analyses of large language models (LLMs) such as ChatGPT and Google Gemini reveal that both can provide reliable and concise information on surgical conditions, though readability and required education level may vary between platforms.⁹ AI chatbots have also been shown to improve the readability of patient-facing content, particularly in cancer education, by simplifying language and reducing the literacy barrier without compromising content quality.¹⁰ In chronic disease management, AI tools using natural language processing can tailor patient support and education to individual needs, with clinicians perceiving these tools as highly useful for patient empowerment and care efficiency.^{11 12} Furthermore, research indicates that individuals’ intentions to seek health information from AI are influenced by attitudes and perceived control, rather than trust or risk perceptions.¹³

Existing perioperative patient education tools often lack interactivity, personalisation and scalability, limiting their ability to meet patients’ diverse needs effectively. While LLMs and AI-driven chatbots show promise in addressing these limitations, there is limited evidence on how orthopaedic patients and their families perceive and use these tools during the perioperative period. Understanding user experiences and identifying factors that influence adoption, usability and effectiveness of AI-driven chatbots is essential to optimise their potential.

This study aims to explore the user experiences of orthopaedic patients and their families with commercially available AI-driven chatbots for perioperative health information seeking. Employing a descriptive qualitative approach, it seeks to capture rich, indepth insights into how these tools are perceived, used and integrated into the perioperative care journey. The findings will provide valuable guidance for designing, implementing and improving AI-driven health information tools, ensuring they meet the needs of patients and caregivers effectively.

Methods

Study setting and research team

This study took place at the Third People’s Hospital of Chengdu, a leading care and research centre for complex and severe orthopaedic surgeries that manages over 1200 perioperative cases annually, including many patients aged under 18 years who may rely on parents or close relatives for support and medical information, including use of AI-driven chatbots during the perioperative period. This patient mix enabled us to explore chatbot use both by younger patients and by the caregivers who often act as primary information seekers on patients’ behalf.

The research team comprised healthcare professionals actively involved in various aspects of perioperative care, such as preoperative assessment, symptom management, patient education and postoperative follow-up. To ensure methodological rigour and consistency, all team members completed training in qualitative research methods, semistructured interviewing techniques and Colaizzi’s thematic analysis prior to data collection and analysis.

Study design

This is a descriptive qualitative study, chosen to explore and understand the user experiences of orthopaedic patients and their families with commercially available AI-driven chatbots for perioperative health information seeking. A descriptive qualitative approach is particularly suited for studies aiming to capture rich, detailed insights into real-world experiences and perceptions,¹⁴ making it ideal for examining how individuals interact with and perceive the utility of innovative technologies such as AI chatbots in healthcare.

Sampling strategy

We employed purposive sampling to recruit orthopaedic patients and their family members who had used AI-driven chatbots for perioperative health information seeking.^{15 16} To capture a wide range of experiences, we deliberately diversified the sample across participant role (patient vs caregiver), age, gender, educational level and occupation. This approach ensured inclusion of individuals who had directly used AI chatbots or used them for information seeking on behalf of patients during the perioperative period, while maximising variation in demographic and contextual factors that could potentially influence attitudes and behaviours. By sampling for both relevance and diversity, we aimed to generate rich, detailed insights.

Inclusion and exclusion criteria

Patients were eligible if they were between 18 years and 60 years of age, had undergone orthopaedic surgery over the last 6 months before the time of study, had no diagnosed psychological or cognitive condition that could impair their ability to participate, had used at least one AI-driven chatbot to seek health-related information during the perioperative period and were fluent in Mandarin Chinese. Any patient who did not meet all the criteria was excluded.

Family members were eligible if they were between 30 years and 70 years of age, had played an active role in caring for or making medical decisions on behalf of a patient during the perioperative period, had no diagnosed psychological or cognitive condition that could impair their ability to participate, had used at least one AI-driven chatbot to seek health-related information on the patient’s behalf during that period and were fluent in Mandarin Chinese. Any family member who did not meet all the criteria was excluded.

We chose to include family members because many younger patients rely on parents or close relatives to gather and interpret health information. Including caregiver perspectives allowed us to capture both the direct experiences of patients and the experiences of those who serve as the primary users of AI tools on behalf of patients.

Data collection

The demographic and clinical data of patient participants were retrieved from their medical records. The demographic data of family member participants were collected at the beginning of the interviews. Data for thematic analysis were collected through face-to-face, semistructured interviews conducted in a private and quiet meeting room, free from disturbances.^{17 18}

A semistructured interview guide was developed, featuring open-ended questions to explore various aspects of the experiences of patients and their family members through face-to-face semistructured interviewing. The interview guide was pilot-tested and refined using feedback from the first two participants, whose responses were excluded from the thematic analysis. See online supplemental appendix 1 for the final interview guide.¹⁹

To maintain consistency, all interviews were conducted by the principal investigator (PI), who is also the first author of this study. No more than three interviews were scheduled per day to minimise interviewer fatigue and ensure the quality of data collection.

Procedure

The PI identified and selected participants during one of their perioperative care encounters based on the inclusion and exclusion criteria. Eligible individuals were invited to participate after the PI explained the study objectives, methods, intended use of findings and measures to protect privacy and confidentiality.

Participants who provided consent were immediately invited to take part in a 30 min face-to-face interview. During the interviews, the PI used techniques such as active listening, probing, reflective mirroring and strategic pauses to encourage participants to share their experiences in depth. Field notes were also taken to document significant non-verbal cues, including notable changes in facial expressions, body postures, prolonged pauses and other emotional displays that provided additional context to participants’ verbal responses.

Data saturation and transcription

Data saturation was considered achieved when two consecutive interviews failed to yield new meaningful information or themes.²⁰ At this point, data collection was concluded.

All interviews were audio-recorded with participants’ consent and transcribed verbatim by another researcher within 24 hours of each interview. To ensure accuracy, the transcripts were reviewed and verified by a separate researcher.^{17 21}

Data analysis

Colaizzi’s thematic analysis method was used to analyse the interview transcripts in this study.²² The analysis was led by the PI and an additional researcher. The process began with a thorough reading and re-reading of all transcripts by the research team to develop a comprehensive understanding of the content, forming the basis for the initial analysis. Significant statements directly related to the participants’ use of AI-driven chatbots for perioperative health information seeking were then identified.

Each significant statement was examined to extract underlying meanings. The research team engaged in detailed discussions to compare and refine these interpretations. These extracted meanings were subsequently grouped into thematic categories by organising similar insights, facilitating the structuring of data into coherent segments that accurately represented key aspects of the participants’ experiences.

Once thematic categories were established, the themes were integrated and refined to construct a comprehensive narrative. This narrative was consistently reviewed against the original data to ensure fidelity to the participants’ experiences. A detailed description of the themes, incorporating both textural and structural elements, was developed to effectively articulate and contextualise the user experiences.

To validate findings and enhance credibility, member checks were conducted by returning the interpreted results to participants for their feedback and confirmation.

Verification, rigour and trustworthiness

To ensure the rigour and trustworthiness of the study, several strategies were employed. In addition to member checking, the research team was carefully selected after a thorough review of their academic qualifications, relevant research experience and philosophical orientations.^{14 23} Team members disclosed any personal experiences that could influence their perspectives. Rather than excluding members with such experiences, the team used reflective practices to harness these perspectives, enriching the analysis with greater compassion and nuance.

Data triangulation was achieved by synthesising information from interviews, demographic data and observed non-verbal cues. This minimised potential biases and strengthened the findings. Regular peer debriefing sessions were conducted within the research team and with an external qualitative researcher who was not directly involved in the study to ensure methodological rigour.

Data confidentiality and privacy protection

Strict measures were taken to protect participant confidentiality and privacy. Personal identifiers were anonymised using index numbers (eg, PX for patients and FX for family members, where X represents sequential numbers). Interviews were recorded only after obtaining explicit consent, and participants were informed about the confidentiality of their responses and their right to withdraw at any time without impacting their care.

All collected data, including audio recordings and transcripts, were stored on a password-protected flash drive and secured in a locked drawer in the PI’s office. Physical documents, such as demographic forms and field notes, were also stored securely and were accessible only to authorised members of the research team.

Use of LLM

We used ChatGPT (OpenAI, San Francisco, USA) to translate and refine English-language text for participant quotes, the initial manuscript in Chinese and the subsequent revisions. None of the researchers are native English speakers or are proficient in academic writing in English. We selected the ChatGPT after comparing several LLM AI tools for its favourable accuracy and fluency in English language. All ChatGPT outputs underwent careful review by two bilingual team members proficient in reading academic literature in Chinese and English. A bilingual proofreader then conducted a final review of the manuscript and the subsequent revisions. The research team approved the final version before submission.

Results

Demographics

13 participants, including six patients and seven family members, were interviewed between 17 June and 16 July 2024. Participants’ ages ranged from 19–61 years (average age 37.2 years). Gender distribution was balanced, with seven female and six male participants. Education levels varied, as eight participants held at least a college degree while five had completed or were attending a senior high school or below. Occupational backgrounds showed diverse socioeconomic characteristics, including five students, two public servants, two housewives and several other occupations. This purposively selected sample allowed for comprehensive exploration of experiences across diverse roles, ages, educational backgrounds and professional contexts for enriched details and insights. (table 1)

Table 1. Demographic characteristics of participants (n=13).

#	Age, year	Gender	Education level	Occupation
P1	18–30	Male	College or over	Student
P2	18–30	Male	Senior high or below	Student
P3	30–40	Male	College or over	Public servant
F4	>50	Female	Senior high or below	Retired
P5	18–30	Female	College or over	Small business owner
F6	30–40	Female	College or over	Housewife
F7	>50	Male	College or over	Public servant
F8	30–40	Female	College or over	Student
P9	18–30	Male	Senior high or below	Student
P10	30–40	Male	College or over	Freelancer
F11	30–40	Female	College or over	Office worker
F12	18–30	Female	Senior high or below	Student
F13	18–30	Female	College or over	Housewife

F, family member; P, patient.

Main themes

Thematic analysis of the interview data revealed four key themes reflecting participants’ experiences with commercially available AI-driven chatbots for perioperative health information seeking, including the enhanced accessibility and quality of information, the importance of effective prompting skills, the transition from traditional search engines and the challenges posed by specificity and accuracy in chatbot responses (online supplemental appendix 2).

Theme 1: preference of AI chatbots over search engines

Participants overwhelmingly favoured AI chatbots over conventional search engines for perioperative health inquiries. They emphasised that chatbots eliminated the tedious process of sifting through irrelevant links, ads and conflicting advice typical of search engines. Instead, AI chatbots provided immediate, concise, practical answers tailored to each user’s specific surgical context.

Subtheme 1: efficient, targeted retrieval

Participants described traditional search as finding a ‘needle-in-a-haystack’, requiring many clicks to locate relevant content amid ads and noise. In contrast, chatbots returned focused answers in a single exchange. This efficiency reduced frustration and effort, allowing users to obtain needed information, such as recovery timelines or wound-care instructions, without browsing multiple pages.

I can just ask it directly and the AI chatbot gives me a very accurate answer. I don’t have to fish around everywhere and analyse things myself. It’s such a hassle. I often read a lot and still end up confused. AI is so straightforward: I ask, and it answers right away. (P1)

Subtheme 2: personalised, context-sensitive advice

Users reported that AI chatbots adapted recommendations to their individual surgical details, whereas web articles only provided generic or even contradictory guidance. Chatbots considered factors such as surgery type, patient age and recovery stage to deliver customised plans, dietary advice, activity schedules or warning signs. This tailoring fostered trust in the relevance and applicability of the information, which participants found more valuable than broad, one-size-fits-all resources on search engines.

The AI can give me tailored advice based on my child’s situation. For example, I tell it what surgery my daughter had, then ask, ‘How should she recover after surgery?’ It replies, ‘Based on your surgery type, I recommend you do X and Y during the first week….’. (F6)

Subtheme 3: clear, practical guidance

Rather than lengthy articles, participants valued AI chatbots’ concise, step-by-step instructions. Guidance on wound care, medication timing and diet came formatted as clear lists and schedules. Users appreciated that chatbots highlighted key precautions, sterilisation steps, signs of infection or when to contact a doctor, without extraneous detail. This practical clarity saved time and reduced confusion, enabling users to understand and follow recommendations.

It gives me the answer right away and even reminds me what to watch out for. It’s so reassuring. The responses are direct and practical. For instance, when I asked, ‘How soon after surgery can I start walking?’ it told me, ‘In your case, you can usually try walking around three days post-op.’ (P3)

Theme 2: improved accessibility and quality of information

Participants valued AI chatbots for delivering expert-level health advice without the constraints of clinic visits or the inefficiencies of web searches. Chatbots provided immediate, reliable answers anytime, reducing time spent filtering online results or waiting for appointments. This seamless access and enhanced information quality addressed common frustrations, such as long waiting times and inconsistent web content.

Subtheme 1: 24/7, seamless access

Chatbots eliminated logistical barriers such as clinic hours, appointment queues and slow doctor responses. Users could ask questions at any time, even late at night, and receive instantaneous advice. This on-demand availability ensured that frequent, simple queries required no scheduling, making perioperative information seeking considerably more convenient.

One night I suddenly remembered a question, whether my son could eat fruit after his surgery, so I immediately asked the AI and got an answer right away. (F11)

Participants contrasted AI chatbot interactions with time-consuming hospital visits. Rather than enduring long waits to see a doctor, this convenience reduced stress with traditional information seeking.

At the hospital, you have to book on your phone, wait in line to register, and then wait forever just to ask one question. With the AI (chatbot), I can ask anytime, anywhere. (P5)

Subtheme 2: professional-level, detailed guidance

AI chatbots’ responses included step-by-step, easy-to-understand instructions and contextualised recommendations. Users appreciated the clarity and comprehensiveness, which surpassed the generic or fragmented, even flawed information from traditional sources.

It quickly told me, ‘Avoid showering in the first week after surgery to prevent infection. If your wound heals well, you can take a short shower after one week, but be careful not to get the incision wet.’ (P3)

When I searched online for post-op recovery schedules, the first three pages were ads and the next few were articles I couldn’t make sense of. Asking AI, it laid out the whole timetable professionally. (F11)

Theme 3: preference of AI over human interactions

Participants contrasted AI chatbots with busy clinicians, highlighting chatbots’ unwavering attentiveness, thoroughness and emotional safety. Chatbots listened without impatience or judgement, responded consistently to both simple and complex queries and removed the need to ‘bother’ doctors. Users often felt anxious about approaching busy doctors but found chatbots patiently receptive. The AI interface encouraged open dialogue, allowing participants to voice any concern, no matter how trivial, without fear of annoyance or dismissal.

I feel like the AI chatbot is better than a person. It doesn’t think my questions are stupid or get annoyed if I ask basic things. It never brushes me off, even if I ask a lot. When doctors seem unhappy, I hesitate to ask more. AI has none of that. I can ask anything! (F13)

Chatbots removed the emotional hurdle of ‘bothering’ clinicians. On-demand access meant no waiting rooms or rushed consultations. The convenience of querying the AI at any hour, without worrying about imposing on medical staff, encouraged users to actively seek information.

I’m actually a bit afraid to ask doctors at the hospital… the (AI) chatbot is different. It listens patiently and gives very detailed answers, which makes me feel at ease. (P1)

Theme 4: importance of effective prompting

Many participants emphasised that ’how I ask questions’ directly shaped the usefulness of chatbot responses. Broad or vague prompts elicited generic advice, whereas detailed, context-rich queries produced specific, actionable guidance. Users learnt to refine their inquiries with increased specificity and context.

At first, I asked, ‘How do I recover after a lower-leg fracture surgery?’ It gave a broad answer… Later, I clarified my son’s condition and surgery, then asked, ‘When can my son start practicing with crutches?’ It responded, ‘Based on his situation, he can usually try walking with support around three days post-op….’ (P1)

Some participants described an iterative approach of questioning, first posing an initial, broad question and then following up with focused queries based on the AI’s first response. This back-and-forth allowed them to drill down in depth and breadth.

First I asked, ‘How do I care for my surgical wound?’ it said, ‘Keep the wound dry and other generic stuffs’. Then, ‘What if the incision becomes red and swollen?’ It replied, ‘If you notice redness, swelling or heat, contact your doctor right away…’ I kept asking detailed questions, like what medicine to take and when, before or after meals. (P10)

Discussion

Participants overwhelmingly preferred AI chatbots over traditional search engines for perioperative health inquiries, which is supported by emerging research indicating both the efficiency and personalisation advantages of chatbots. Studies show that users value the ability of AI chatbots to provide immediate, concise and context-specific answers, reducing the frustration associated with sifting through irrelevant links, advertisements and conflicting information typical of search engines.^{12 24} This aligns with our finding on efficient, targeted retrieval, as users report that chatbots streamline the information-seeking process, delivering focused responses in a single exchange and minimising the effort required to locate relevant content.²⁵

Additionally, the personalised, context-sensitive advice offered by chatbots resonates with findings that users appreciate tailored recommendations based on individual health details, which fosters greater trust and perceived value compared with the generic guidance found on web articles.^{12 26} However, while satisfaction and perceived utility are high, other research also notes that users remain cautious about the accuracy and actionability of chatbot-generated information, often cross-checking advice with healthcare professionals or other sources.^{27 28} Furthermore, concerns persist regarding the complexity and readability of chatbot responses, which may limit accessibility for some patients.28,30 This observation, however, was not prominent in our participant group. Overall, our findings confirm the potential of AI chatbots to enhance the health information-seeking experience through efficiency and personalization.

The finding that AI chatbots improve both accessibility and the quality of health information is strongly supported by recent research. Studies consistently show that chatbots provide 24/7, on-demand access to health advice, eliminating barriers such as clinic hours, appointment wait times and the inefficiencies of traditional web searches.^{31 32} This immediate availability is especially valued for routine or urgent questions, allowing users to receive timely, stress-reducing guidance without the need for scheduling or travel.^{33 34}

In terms of information quality, AI chatbots are recognised for delivering clear, step-by-step and contextually relevant instructions that often surpass the fragmented or overly complex content found through conventional search engines. For example, chatbots have demonstrated high accuracy and reliability in handling medical queries, with some models achieving precision and recall scores above 95% and providing detailed, actionable recommendations tailored to individual needs.³⁵ However, while the quality of information is generally high, some studies note that chatbot responses can still be written at a reading level above what is recommended for the general public, indicating a need for further improvements in readability and actionability.^{10 28 30}

Participants’ preference for AI chatbots over human interactions for health information, citing chatbots’ patience, non-judgmental responses and emotional safety, reflects a growing trend documented in recent research. Studies show that AI chatbots are often rated as more empathetic and thorough than human clinicians in online health forums, with one large-scale evaluation finding that chatbot responses were preferred over physician responses in nearly 79% of cases and were rated significantly higher for both quality and empathy.³⁶ Similarly, in mental health support, AI chatbots have demonstrated greater perspective-taking and empathic concern compared with human therapists, and users report feeling more comfortable disclosing sensitive or ‘trivial’ concerns to chatbots, free from fear of judgement or impatience.^{37 38} This emotional safety and consistent attentiveness encourage users to seek information more openly and frequently, reducing the anxiety associated with ‘bothering’ busy clinicians.^{12 37} However, while chatbots excel in providing supportive, non-judgmental environments and detailed responses, some research notes that users may still harbour concerns about trust, accuracy and the lack of nuanced human understanding, especially for complex or emotionally charged issues.^{5 39}

According to the participants, the effectiveness of AI chatbot responses in health contexts is highly dependent on the specificity and clarity of user prompts. This is consistent with recent studies that structured, context-rich prompts significantly enhance the accuracy, relevance and completeness of chatbot-generated health information. For example, the use of the SMART prompt structure, incorporating details about the seeker, mission, AI role, register and targeted question, led to notably higher quality scores in clinical scenarios and patient inquiries compared with unstructured prompts.⁴⁰ This process aligns with participants’ experiences of refining their questions iteratively to obtain more actionable and tailored advice. Research also indicates that users often struggle with vague or poorly constructed prompts, resulting in generic or less useful responses and that providing users with prompting guidelines or frameworks can improve both the quality of AI responses and user satisfaction.⁴¹ Furthermore, effective prompting not only improves informational accuracy but also supports more empathetic and contextually appropriate chatbot interactions, which is especially important in sensitive health domains.^{42 43} As a result, user education and prompt engineering may play a vital role in maximising the value of AI chatbots for patient education and health information seeking.

Implications for practice

Healthcare organisations should embed AI-driven chatbots into perioperative care workflows to provide patients with personalised guidance outside clinic hours. Suggestively, AI chatbots should be integrated into admission and discharge protocols and patient portals to provide consistent access to up-to-date recommendations, reducing the risk of information gaps. Clinicians and health educators should develop in-app tutorials on crafting context-rich prompts to enhance response precision and support patients in obtaining actionable advice. In ensuring quality of chatbot-generated information, regular review committees composed of multidisciplinary stakeholders should audit chatbot content against evidence-based guidelines to maintain clinical accuracy and mitigate potential biases. To address varying health literacy levels, chatbot designers should implement adaptive interfaces capable of adjusting language complexity and offering multimedia formats. Additionally, organisations should establish clear policies defining the scope of chatbot deployment, delineating when to escalate complex or emergency queries to human providers. Structured tutorials may be provided on effective prompting to enhance health information seeking.

Limitations

This study has several limitations. First, conducted among Chinese participants, the results may not fully reflect experiences in other cultural or healthcare contexts. Second, reliance on self-reported data introduces the potential for recall or social desirability bias, as participants may overemphasise positive experiences with AI chatbots. Finally, the study focused solely on orthopaedic patients and their families, which may not capture the perspectives of other patient populations or healthcare providers. Future research should explore broader and more diverse samples to enhance the transferability and applicability of the findings.

Conclusions

AI-driven chatbots offer a promising adjunct to perioperative patient education by delivering immediate, tailored guidance that overcomes the limitations of conventional search engines and busy clinical settings. Study participants valued chatbots’ efficient, context-sensitive retrieval, professional-level advice and non-judgmental interactions, which fostered trust and reduced anxiety. Effective prompting emerged as a key user skill, directly shaping response relevance and accuracy. Chatbot-generated health information should be regularly reviewed for accuracy. Structured tutorials may be offered for user capacity building.

Data availability statement

Data are available upon reasonable request.