The development and use of chatbots in enhancing health care access for underserved and vulnerable populations: a scoping review

Abstract

Background

Chatbots, as innovative conversational agents, leverage their widespread accessibility to enhance healthcare service delivery, and in recent years have attracted considerable attention for their capacity to serve underserved and vulnerable populations. This study aims to conduct a scoping review to examine the development and application of chatbots for these populations, highlighting their role in addressing healthcare disparities.

Methods

This scoping review was conducted using Arksey and O’Malley’s framework and reported according to PRISMA-ScR guidelines. Six databases, including PubMed, Scopus, Web of Science, Embase, ProQuest, and EBSCO, were searched without time restrictions for English-language studies. Eligible articles focused on the development and/or evaluation of chatbots designed to address healthcare access barriers among underserved and vulnerable populations. Data were extracted, charted, and synthesized to identify key themes, patterns, and research gaps.

Results

This review identified 18 studies, most of which were conducted after 2020 (n = 15) and primarily in the United States (n = 5). The main target groups included adolescents (n = 4) and women (n = 3), with a focus on domains such as mental health (n = 6) and sexual and reproductive health (n = 6). Chatbots were predominantly rule-based (n = 13) and mostly delivered via social media (n = 6)or web platforms (n = 6). Nine studies conducted evaluations, reporting outcomes such as acceptability and feasibility.

Conclusion

Although the development and use of chatbots for underserved and vulnerable populations have grown in recent years, research in this area remains limited across countries. Chatbots, as interactive communication tools, hold notable potential to improve access to healthcare services. Based on the findings of this review, researchers can better understand the capabilities of chatbots in this context and are encouraged to conduct further studies focusing on these populations.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-25814-2.

Introduction

Equitable access to healthcare services is a fundamental right for all people throughout life, playing a crucial role in maintaining health, improving well-being, and enhancing quality of life [1]. Access to healthcare services is a complex concept, with multiple interpretations presented in scientific literature [2]. According to Parker's definition, healthcare accessibility refers to a community's ability to reach, obtain, and afford healthcare services [3]. This encompasses various dimensions, including the availability of health facilities, geographic accessibility, and Financial affordability [4]. Approximately 400 million individuals worldwide lack access to healthcare services [5]. By 2024, an estimated 3.5 million people had lost their lives due to the lack of access to healthcare services [6]. From a geographical perspective, 8.9% of the global population (equivalent to 646 million people) are unable to reach healthcare facilities within one hour, even with access to motorized transportation [7]. Among these populations, underserved and vulnerable populations account for a notable proportion [8, 9]. Underserved populations are defined as groups that encounter considerable obstacles in obtaining healthcare services relative to the general population [10]. These disparities stem from factors including geographical, economic, cultural, and linguistic issues, as well as unfamiliarity with the healthcare delivery system and low health literacy [11]. The term “underserved” is often used alongside “vulnerable,” and although these two are conceptually related, they are not interchangeable [12]. Multiple factors contribute to limiting access to healthcare services for vulnerable populations. These factors can be categorized into three main domains: the physical domain (e.g., high-risk mothers and infants), the psychological domain (e.g., those with chronic depression), and the social domain (e.g., refugees) [13, 14]. Every individual may fall into one or both categories of underserved or vulnerable populations. Numerous studies have highlighted the lack of access to healthcare services among underserved and vulnerable populations, emphasizing the need to address and mitigate these barriers [15–18]. The significance of including underserved and vulnerable populations in research is thoroughly elucidated in the study by Mehl et al. The authors consider such inclusion not only a necessary undertaking but also an ethical imperative, as the active participation of these groups in the research process can play a pivotal role in reducing existing healthcare disparities and advancing health equity [19, 20].

Digital technology is increasingly recognized as a vital tool for promoting social justice in the modern era [21–23]. In today's digital landscape, chatbots have emerged as accessible, cost-effective, and scalable technological interventions [24]. Chatbots, as conversational agents, play a pivotal role in facilitating human-technology interactions [25]. These computer programs are designed to simulate human conversation by responding to user queries through text, voice, visuals, or a combination thereof [26]. Despite challenges such as limited internet access in underserved regions, the mobile phone and personal computer markets have experienced notable growth in these areas [27]. This expanding opportunity highlights the potential of digital tools, such as chatbots, to create new avenues for delivering healthcare services in underserved and vulnerable communities [28]. These tools can provide effective support for assessment, diagnosis, information provision, treatment, and management of various health conditions, particularly for underserved and vulnerable individuals with limited resources and access to healthcare services [29–31]. Chatbots enable flexible conversational interactions by allowing communication at any time and from any location. This accessibility helps create a safe environment for patients, reducing stigma and thereby encouraging greater engagement [32]. In addition, chatbots can simplify complex medical concepts and present them in clear, conversational language, which supports individuals with low health literacy in understanding essential health information and participating more effectively in their care [33].

The design of chatbots varies depending on factors such as the target population, intended functionality, desired interaction type, and the development platform employed [34]. For instance, Han et al. [35] developed a conversational agent named PTSDialogue to support vulnerable individuals with post-traumatic stress disorder (PTSD) who face notable barriers to accessing essential treatment resources. This chatbot provided psychoeducation, assessment tools, and symptom management strategies, with findings indicating its feasibility and acceptance as a supportive tool for individuals with PTSD. Similarly, Manole et al. [36] proposed an AI-based chatbot, built on the ChatGPT framework, to deliver personalized interventions for anxiety patients, particularly in low-resource settings. The chatbot offered mindfulness exercises, cognitive restructuring, and breathing techniques, achieving high user satisfaction due to its accessibility and personalized approach.

To the best of our knowledge, the existing literature lacks a scoping review on the development and use of chatbots specifically designed for underserved and vulnerable populations. This article represents the first comprehensive review in this field. Given the substantial population lacking access to healthcare services, particularly underserved and vulnerable populations and the remarkable advancements in chatbot technology in overcoming barriers to healthcare access and enhancing service delivery, this study aimed to review the applications of chatbots, to identify the target populations and health domains in which they have been employed, and to examine how these chatbots have been developed and evaluated. Such a review could establish a conceptual framework to better understand the capabilities, limitations, and challenges of chatbots for underserved and vulnerable populations, identify existing research gaps, and provide meaningful insights for researchers, technology developers, and healthcare policymakers to design and implement effective and equitable chatbot-based interventions.

Methods

Protocol and registration

The study protocol was developed and subsequently registered on the Open Science Framework (OSF) after data collection to enhance transparency and ensure alignment with the PRISMA-ScR checklist. The protocol is available at 10.17605/OSF.IO/WT3B2.

Framework

We implemented the scoping review method based on Arksey and O’Malley’s methodological framework [37] and the PRISMA standard [38]. Arksey and O’Malley’s methodological framework is a highly cited approach focuses that a scoping review should done because of these reasons: 1) examine existing literature in a given field 2) determine the Feasibility and necessity of conducting a systematic review 3) synthesizing the research findings 4) identify research gaps in the existing literature and it consists of five steps: 1) identification of the research question, 2) identification of the relevant studies, 3) selection of included studies, 4) charting the data 5) Collating, summarizing, and reporting results.

The PRISMA reporting guidelines are designed to improve the reporting of systematic reviews. This study used PRISMA-ScR, a PRISMA extension for scoping review. The checklist contains 20 essential reporting items and 2 optional items.

Identifying the research question

What are the different applications of chatbots for target populations and health domains in underserved and vulnerable communities, and how have these chatbots been developed and evaluated.

Identifying the relevant studies

Search

The search strategy was developed in consultation with a doctoral student specializing in library sciences. This ensured the inclusion of all relevant keywords, synonyms, and MeSH terms.

Keywords from two distinct domains were used in the search strategy: 1) underserved and vulnerable population, 2) chatbot.

For underserved and vulnerable population, the terms such as Medically Underserved Population, underserved communities, Rural Population, Developing Countries, Low- and Middle-Income Countries, Resource Limited Areas were used. For chatbot, the terms involved chatterbot, interactive agent, conversational agent. Additionally, a set of MeSH terms were applied in PubMed. The complete search strategy, including all search strings and MeSH terms for each database, has been provided in Supplementary Table 1.

Information sources

An extensive search was conducted up to 18 February 2025 in 6 bibliographic databases, namely PubMed, Scopus, Web of science, Embase, ProQuest and EBSCO with no time restrictions. Searches were limited to English studies. The results were imported into reference management tool, and duplicate references were removed from the results.

Selection studies

Eligibility criteria

This review encompasses original studies published in peer-reviewed journals that meet specific inclusion criteria. Eligible studies were required to focus on the development, with or without evaluation, of at least one chatbot designed specifically for underserved and vulnerable populations. Additionally, these studies had to address specific barriers to healthcare access (e.g., linguistic, geographical, or other relevant obstacles) faced by these populations. Studies were excluded if the chatbot was not specifically designed to address barriers to healthcare access for underserved and vulnerable populations. Additionally, studies were excluded if they primarily explored aspects such as user engagement, satisfaction, ethical considerations, potential benefits, limitations, or acceptance, without directly focusing on the development of chatbots tailored to mitigate specific healthcare access barriers for these populations.

Selection of sources of evidence

After removing duplicate studies, all retrieved records were imported into Rayyan, a free web/mobile application for the screening process. Two authors independently (blind mode on) evaluated the titles, abstracts, and full texts of the retrieved articles. Studies that did not meet the predefined criteria were excluded. In cases of disagreement between the two reviewers, the first author acted as a tiebreaker and made the final decision after discussion to reach consensus. Finally, the first author double-checked all included articles to ensure consistency with the inclusion criteria before proceeding to data extraction.

Charting the data

Data charting process

Key data items for analyzing and synthesizing the qualitative data were determined based on preliminary reviews of articles and then the consensus meeting with all authors. Two authors separately extracted these data, and the first author then examined and, if feasible, harmonized the extracted data.

Data items

For each eligible article, the following data items were extracted: authors; year of publication; title; location of study (country); study design; target users (population); barriers to healthcare access; chatbot name; chatbot features, platform, approach; method of input by the user; description of how the chatbot address the condition; outcome measure; Conclusion; evaluation criteria; evaluation methods.

Collating, summarizing, and reporting results

Synthesis of results

The data were organized and summarized using Microsoft Excel to create structured tables of extracted information. Charts were generated to illustrate various target populations, health domains, and intervention types, facilitating the identification of patterns and comparisons across studies. These visualizations enabled a clearer understanding of research trends, thematic areas, and gaps within the literature.

Results

Literature search

A total of 676 articles were identified from electronic databases using the specified keywords. After removing 384 duplicates, 265 articles were excluded following the review of titles and abstracts (Fig. 1). An additional 12 articles were excluded following full-text analysis as they did not meet the inclusion criteria. Ultimately, 18 articles were included for qualitative analysis. Full details of the final articles are given in (Table 1).

Fig. 1.

The PRISMA diagram

Table 1.

Summary of selected studies

Authors/ Year of publication	Title	Country	Study Design	Target users(population)	Barriers to healthcare access	Chatbot name	Chatbot Features, Platform, approach	Method of input by user	How does the chatbot address the condition	Outcome measure	Conclusion	Evaluation Criteria	Evaluation Methods
Jack et al./2015 [39]	Reducing Preconception Risks among African American Women with Conversational Agent Technology	USA	Randomized controlled trial (RCT)	African American women aged 18–34, non pregnant, English-speaking	-Limited health literacy -racial disparities in healthcare access, -lack of culturally sensitive care	Gabby	An online interactive conversational agent (animated character with nonverbal communication) Platform: web-based (They received email) AI OR Rule base: Rule-based	predefined options and free-text input	AI chatbot gives personalized preconception advice and tracks health goals. It adapts to user needs and creates custom health plans Counseling/support, treatment support	Reducing Preconception health risks (e.g., nutrition deficiencies, STI risk, lack of folic acid supplementation)	Using the Gabby chatbot had a positive impact on reducing health risk factors among participants. Women who interacted with Gabby spent more time engaging with the system, found it easy and understandable, showed greater reductions in both the number and proportion of health risks, and many used the information to improve their health	-	-
King et al. (2017) [40]	Testing the comparative effects of physical activity advice by humans vs. computers in underserved populations: The COMPASS trial design, methods, and baseline characteristics	USA	Cluster-Randomized Controlled Trial	Latino adults 50 and older with limited access to health services	- Racial/Ethnic minority populations - language literacy levels - Low-income populations with limited healthcare access - Lack of culturally adapted programs for Latino adults	Carmen	The ECA is an interactive animated touchscreen tool delivering bilingual (Spanish/English) designed for accessibility without keyboards Platform: dedicated computer (supplied by the study) located in a private, secure area at each designated community center receiving that intervention AI OR Rule-base: Rule based	predefined conversation boxes	The chatbot offers customized exercise guidance(advice), helping users start walking programs. It provides personalized feedback and goal-setting tools to address barriers like social anxiety counseling	-Changes in weekly walking minutes (primary) -Physical function, BMI, and well-being (secondary)	It would be potentially low-cost, readily accessible option that could be broadly disseminated across a range of community settings It has substantial potential to reduce the health disparities gap by influencing a key health behavior in underserved populations	-	-
Daley et al./ 2020 [41]	Preliminary Evaluation of the Engagement and Effectiveness of a Mental Health Chatbot	Brazil	prospective observational study	was installed voluntarily by members of the general population who were Portuguese speakers located in Brazil, over 18 years of age, and with internet access	-large mental health treatment gaps - scarce resources available	Vitalk	Platform: instant messenger platform (mobile application) AI OR Rule-base: mix	predefined response options, with some ability to input free text	The chatbot delivers CBT-based mental health support through interactive tools and coping strategies. It provides symptom tracking, relaxation exercises, and crisis intervention when necessary Treatment support	reducing stress, anxiety and depression	Vitalk effectively reduced symptoms of anxiety, depression, and stress, with higher user engagement linked to greater improvement in anxiety and depression	-	-
Miraj, F., et al./2021 [42]	Development and feasibility testing of an artificially intelligent chatbot to answer immunization-related queries of caregivers in Pakistan: A mixed-methods study	Pakistan	Mixed-methods study (quantitative and qualitative)	Caregivers of children under 2 years visiting immunization centers in low-resource settings	-limited-resource setting -Low literacy levels -Limited access to reliable immunization information -low-income communities	Bablibot (Babybot)	supports Roman Urdu (Urdu written in English script), interoperability with Electronic Immunization Registries (EIRs) Platform: WhatsApp (they received SMS) AI OR Rule-base: mix	Free text	The chatbot delivers instant vaccination guidance via SMS/WhatsApp, using AI to help caregivers track schedules and side effects. It escalates to human experts when needed, ensuring reliability in low-resource areas Health monitoring Treatment plan	-Improved access to immunization information (e.g., vaccination schedules, side-effect management) -High user satisfaction (90% based on exit surveys)	It was suggested that a local language AI-based chatbot is a feasible and acceptable intervention for providing immunization information to caregivers in limited-resource and low-literacy settings	Feasibility -Acceptability -Technological Performance	Quantitative: Tracked 1877 messages across 874 conversations and analyzed user engagement metrics Qualitative: in-depth interviews
Nelekar et al.,2021 [43]	Effectiveness of embodied conversational agents for managing academic stress at an Indian University (ARU) during COVID-19	India	Mixed-methods study (quantitative and qualitative)	61 undergraduate students at an Indian university	-High academic stress and competition -Low awareness of mental disorders -Stigma around mental health discussions (social stigma) -the scarcity of mental health professionals -COVID-19 restrictions reducing social interaction	ARU	Embodied conversational agent (ECA), supports Hindi language, TTS (text to speech) voice, Verbal and not verbal communication, COVID-19 tailored dialogue, based Platform: web based AI OR Rule-base: Rule based	Predefined values with multiple-choice options	The chatbot offers customized stress-management tips for students, adapting advice to cultural beliefs and goals education	-Significant stress reduction -Increased intentions to adopt stress-reducing behaviors -High trust and working alliance scores	The ECA technology has shown promise in terms of stress reduction amongst Indian students	-	-
Ntinga et al., 2022 [44]	The Feasibility and Acceptability of an mHealth Conversational Agent Designed to Support HIV Self-testing in South Africa: Cross-sectional Study	South Africa	Cross-sectional pilot study	Adults aged ≥ 18 years residing in rural Vulindlela subdistrict in KwaZulu-Natal province or neighboring communities, able to use a smartphone, and with prior HIV testing experience	-Stigma, -discrimination from health workers, -lack of privacy in clinic-based testing - distance to health facilities, - costs,	Nolwazi_bot	Built on SnatchBot platform, supports isiZulu and English,,offers four counselor personalities Platform: Telegram AI OR Rule-base: Rule based	predefined options and some simple free-text input	The chatbot guides users through HIV self-testing with culturally sensitive instructions and counseling, while connecting positive cases to care and offering PrEP information to negatives all via judgment-free conversations Counseling education Treatment support	-HIV testing uptake, -linkage to care for HIV-positive individuals	Most participants indicated that their HIV testing experience with a chatbot was much better than that with a human counselor	Feasibility acceptability	Quantitative: Posttest structured survey, Qualitative: asked open-ended questions into the interview
Santa-Cruz et al., 2022[45]	Mobilizing digital technology to implement a population-based psychological support response during the COVID-19 pandemic in Lima, Peru	Peru	Retrospective cohort study	Adults in Lima, Peru, with a focus on vulnerable populations affected by COVID-19, including those experiencing psychological distress	-High levels of psychological distress in the population -Treatment gap in Latin America -Limited access to mental health services due to COVID-19 restrictions -Economic and social challenges exacerbating mental health issues	ChatBot-Juntos	Provided multimedia elements (maps, images, sounds) Platform: WhatsApp, Facebook, Web AI OR Rule-base: Rule-based	Predefined yes/no responses questions with some text input	The chatbot screens for distress using SRQ, delivers PFA/grief support(counseling)in spanish, and refers to specialists via WhatsApp/Facebook, with follow-up to track progress Screening Counseling Treatment support	-Reduction in psychological distress (measured by SRQ scores), -referral rates to mental health services	Significant reduction in SRQ scores was observed 3 months after PFA [median SRQ score changed from 9 to 5 after PFA plus referral to mental health services)median SRQ score changed from 11 to 6(	-	-
Wang et al., 2022[46]	An Artificial Intelligence Chatbot for Young People’s Sexual and Reproductive Health in India (SnehAI): Instrumental Case Study	India	Instrumental case study	Adolescents and young adults in India, particularly those from vulnerable and hard-to-reach populations, focusing on sexual and reproductive health (SRH) education	-Taboo nature of SRH topics in Indian society -Limited access to accurate SRH information -Gender disparities in digital literacy and device ownership	SnehAI	Supports Hinglish (Hindi + English), Offers multimedia content (videos, GIFs, emojis) Platform: facebook AI OR Rule-base: Mix	Free text input with some predefined values	The chatbot creates a safe space for sensitive SRH discussions, delivering AI-powered education and myth correction while linking users to national helplines counseling	Identification of gender disparities in chatbot usage High user engagement Improved access to SRH information	Overall, SnehAI successfully presented itself as a trusted friend and mentor; the curated content was both entertaining and educational, and the natural language processing system worked effectively to personalize the chatbot response and optimize user experience	-	.-
Escobar-Viera et al./2023 [47]	A chatbot-delivered intervention for optimizing social media use and reducing perceived isolation among rural-living LGBTQ + youth: Development, acceptability, usability, satisfaction, and utility	United States	exploratory pilot study with Single-group pretest–posttest	Rural-living LGBTQ + youth aged 14–19 in the U.S., screened positive for social isolation	-Geographic isolation, -lack of LGBTQ + -specific mental health resources in rural areas -lack of information and social support in rural areas	REALbot	Platform: facebook AI OR Rule-base: Rulebased	Predefined options	a chatbot intended to deliver an educational program and teaching social media to increase social media self-efficacy and reduce perceived isolation among rural living LGBTQ + youth education	-improve social media efficacy, -reduce perceived isolation, -bolster connections	REALbot deployment was found to be feasible and acceptable, with good usability and user satisfaction scores	Acceptability usability satisfaction	Quantitative: Questionnaires (UEQ–S, CUQ, PSSUQ, CSQ-8) qualitative feedback: open-ended questions on ‘likes’ and ‘dislikes’ about the intervention
Matheson et al./2023 [48]	Using Chatbot Technology to Improve Brazilian Adolescents’ Body Image and Mental Health at Scale: Randomized Controlled Trial	Brazil	Randomized Controlled Trial (RCT)	Brazilian adolescents aged 13–18 years, geographically and socioeconomically diverse, who spoke Brazilian Portuguese with access to Facebook Messenger	-disparities between mental health needs and services -Limited access to mental health interventions and resources, -high prevalence of body image concerns, -political,economic and research funding disparities in LMCI	Toptty	uses gamification, with avatar (Dandara or Gabriel) Platform: facebook AI OR Rule-base: Rulebased	predefined responses	The chatbot addresses body image concerns via CBT-based microinterventions in portuguese, media literacy tools, and gamified skill-building activities counseling	Improvements in body image, affect, and self-efficacy	intervention was associated with small to moderate improvements in body image among adolescents, as well as significant enhancements in affect and self-efficacy	-	-
McMahon et al./2023 [49]	Perils and promise providing information on sexual and reproductive health via the Nurse Nisa WhatsApp chatbot in the Democratic Republic of the Congo	Democratic Republic of the Congo (DRC)	Descriptive study (lessons learned from development and scaling)	Women and adolescents in the DRC, particularly those facing barriers to SRH services	-Lack of knowledge about where to obtain safe services -stigma around SRH topics, -high rates of sexual violence censorship on social media, -under-resourced health systems )Limited health infrastructure(,	Nurse Nisa	content and navigation were created simultaneously in English, Swahili, Hindi, and French Platform: Whatsapp AI OR Rule-base: Rulebased	predefined responses and menu navigation (e.g., acronyms like "VBG" for SGBV content)	The chatbot delivers confidential SRH information on the stigmatised topics of abortion, contraception, emergency contraception, and gender-based violence via evidence-based, self-guided interactions education	Improved knowledge on SRH topics (abortion, contraception, gender-based violence)	Chatbots, despite their limitations, offer a low-cost and private solution for delivering reliable, evidence-based information on sexual and reproductive health (SRH)	-	-
Nkabane-Nkholongo et al., 2023[50]	Adaptation of the Gabby conversational agent system to improve the sexual and reproductive health of young women in Lesotho	Lesotho	Qualitative study	Young women aged 18–28 in rural Lesotho, with a focus on those facing barriers to sexual and reproductive health (SRH) education	-limited SRH knowledge, -health professionals face constraints related to time and resources, -judgmental attitudes from healthcare providers, -Cultural stigma, -lack of privacy,	Nthabi	Embodied Conversational Agent (ECA) with animations, gestures, and non-verbal cues Tool: Mobile based Platform: mobile application AI OR Rule-base:Rule based	Predefined values	The chatbot delivers culturally attuned SRH education via an embodied AI agent, using relatable gestures and local language (Sesotho) to teach HIV prevention and reproductive health all through judgment-free interactions education	Improved access to SRH information, reduced stigma, increased knowledge of HIV prevention and family planning		Acceptability	Qualitative: Focus group discussions key informant interviews: with interview guide with open-ended questions
Mauricio et al./2024 [51]	AyudaMujer: A Mobile Application for the Treatment of Violence Against Women in Peru	Peru	Experimentalstudy (3-week intervention)	women residing in Lima, Peru, who have experienced physical, sexual, or psychological violence, including five university students, eight with completed school studies, and seven without completed secondary education	-Insufficient treatment of abused women -Shame, -lack of awareness, -fear of blame, -limited access to emergency centers	AyudaMujer	Platform: mobile application AI OR Rule-base: Mix	Free text input,	A mobile application is proposed to support victims of violence against women. The type of violence and risk level are identified by a chatbot, a specialist is assigned for remote care, nearby WECs are located for in-person help, news is provided, and specialists can give feedback on the content Screening & treatment support	Treatment of Violence against women (physical, psychological, sexual)	risk of violence is reduced by an average of 19.43% after three weeks of use. The results show that this tool can contribute to the treatment of abused women	-	-
Mohanty et al./2022 [52]	A Digital Intervention to Reduce Disparities in Well-Child and Immunization Completion in Community Health	USA	Interventional pilot study	Caregivers(parents) of children aged 0–17 years, primarily racial/ethnic minorities (82%) and low-income families in Chicago	-Stigma, -economic hardship, -Transportation limitations, -language barriers -COVID-19 disruptions	CHEC-UP	bilingual (English/Spanish), appointment scheduling, Tool: smartphones or tablets Platform: web based accessible via linked send with sms and email AI OR Rule-base: AI	Predefined values	The chatbot sends personalized well-child visit reminders, provides anticipatory guidance, and enables appointment booking via multilingual messaging to improve preventive care access Health monitoring Treatment plan	-Well-child visit completion, -immunization uptake,	Engaging patients with chatbots improved vaccination and well-child uptake. Patients were highly satisfied with chatbot engagement	Satisfaction	survey (quantitative and qualitative) interviews(qualitative)
Nkabane-Nkholongo et al./2024 [53]	Usability and Acceptability of a Conversational Agent Health Education App (Nthabi) for Young Women in Lesotho: Quantitative Study	Lesotho (adapted gabby)	Descriptive quantitative study	Young women aged 18–28 years in rural districts of Lesotho	-Limited human resources for health, -limited internet access in rural areas	Nthabi	Platform: mobile application AI OR Rule-base:Rule-based	predefined multiple-choice inputs	The chatbot provides culturally tailored SRH education via interactive dialogues and motivational interviewing, offering offline resources to promote health behavior change education	Improved sexual and reproductive health education (focus on HIV, family planning, tuberculosis, healthy eating, folic acid use)	the app was effective in helping them make decisions” and “could quickly improve health education and counselling	-Usability -acceptability	22-item Likert scale survey. (quantitative)
Wen et al./2024 [54]	Chatbot-interfaced and cognitive-affective barrier-driven messages to improve colposcopy adherence after abnormal Pap test results in underserved urban women: A feasibility pilot study	USA	Feasibility pilot study	Underserved urban women (aged 21–65) with abnormal Pap results, primarily low-income, non-Hispanic White (52%) and African American (37%)	-racial/ethnic minority -knowledge gaps (HPV/cervical cancer knowledge) -cultural (fear of HPV transmission, body image, procedural anxiety), -Linguistic (non-English speakers), economic (cost/insurance concerns),	Cervix Chat	bilingual (English/Spanish) Platform: Web-based (: accessible via linked send with text message) AI OR Rule-base: Rule based	Predefined values	The chatbot improves colposcopy adherence via C-SHIP-based counseling, addressing HPV knowledge gaps and procedural anxiety through personalized texts, education, reminders, and coping tools counseling	-Improved colposcopy adherence, -reduced psychosocial barriers	Post-intervention evaluations documented high satisfaction and perceived usefulness, with recommendations for incorporating additional practical and educational content	Feasibility satisfaction	Feasibility: Follow up Surveys (quantitative), Satisfaction: 5-point Likert scale statements (quantitative) telephone interview(qualitative)
de Graaff et al./2025 [55]	Evaluation of a Guided Chatbot Intervention for Young People in Jordan: Feasibility Randomized Controlled Trial	Jordan	Feasibility randomized controlled trial	Young people aged 18–21 years in Jordan with elevated psychological distress	-few evidence-based psychological interventions available for this age group, especially in LMIC, -stigma, -Lack of mental health services, -limited resources in LMIC	STARS	Platform:web based AI OR Rule-base: Rule based	predefined response options	The chatbot delivers CBT-based mental health support including psychoeducation, emotion regulation, behavioral activation and problem-solving tools via interactive lessons, with optional e-helper calls for added guidance counseling	Reducing symptoms of depression and anxiety	This study demonstrated the feasibility and acceptability of the STARS intervention and research procedures	Feasibility	Quantitative & Qualitative: Semi structured interview
Mathur et al., 2025[56]	Let’s chat!" Piloting a digital chatbot for HIV prevention among cisgender women and transgender men in Nigeria	Nigeria	Mixed-methods pilot study (quantitative surveys + qualitative interviews)	Cisgender women and transgender men (assigned female at birth) in Lagos, Nigeria, with limited access to HIV prevention services	-Low PrEP (pre-exposure prophylaxis) awareness among underserved populations -Stigma in healthcare settings -Geographic/economic access limitations	Let’s Chat	Platform: web based AI OR Rule-base:Rule based	Predefined values with multiple-choice or true/false responses	The chatbot provides confidential HIV prevention counseling, prompts users to evaluate their own and their partner’s behaviors that may increase HIV risk, and offers PrEP guidance. Using interactive risk assessments, it empowers users with personalized insights and facilitates discussions with healthcare providers counseling	-Increased PrEP awareness -Reduced stigma in HIV prevention discussions	Participants valuing the tailored responses they received. The chatbot was easy to understand, good way to understand health information, learnt from it	Feasibility acceptability	Exit surveys (quantitative) In-depth interviews using semi-structured interview guide (qualitative)

Characteristics of included studies

All included studies were conducted between 2015 and 2025. Of the 18 eligible studies, the majority (n = 15, 83%) were published after 2020. Although the included studies encompassed data from 10 different countries, the majority were conducted in the United States (n = 5, 28%), followed by Brazil (n = 2), India (n = 2), Peru (n = 2), Lesotho (n = 2), and one study each in Pakistan, South Africa, Congo, Jordan, and Nigeria. The studies varied in research design, with the most common designs being pilot studies (n = 4, 22%), randomized controlled trials (n = 4, 22%), and mixed-methods studies (n = 3, 17%).

Target population and health domain

The review of studies revealed that chatbots have been implemented across diverse target populations and health domains. Chatbots designed for adolescents and young adults [43, 46, 48, 55] primarily focused on key themes, including mental health [43, 48, 55] and sexual and reproductive health [46]. Studies targeting women [39, 51, 54] predominantly addressed sexual and reproductive health [39, 54] and violence against women [51]. Among young women [49, 50, 53] sexual and reproductive health [49, 50, 53] was a particularly prominent focus. Studies involving ethnic and racial minorities [47, 56] primarily addressed mental health [47] and HIV prevention [56]. Chatbots developed for caregivers of children [42, 52] were utilized in the domains of immunization awareness and Uptake. Among other groups, primarily general adult populations residing in underserved areas [40, 41, 44, 45] the topics examined included mental health [41, 45] promotion of healthy lifestyles [40] and HIV prevention [44]. Figure 2 summarizes the target populations and health domains of chatbots developed for underserved and vulnerable populations (Table 2).

Fig. 2.

Summary of diverse populations & Health Domains targeted by chatbots in underserved and vulnerable communities

Table 2.

Target populations and Health domains

Target population	Health Domain
Adolescents and Young Adults	Mental health
	Sexual and reproductive health
Women	Sexual and reproductive health
	Violence against women
Young Women	Sexual and reproductive health
Ethnic and Racial Minorities	Mental health
	HIV prevention
Children caregivers	Immunization Awareness and Uptake
Other	Mental health
	Lifestyle promotion
	HIV prevention

Type of intervention

Figure 3 illustrates the types of interventions implemented by chatbots for diverse populations within underserved and vulnerable communities across various health domains, as categorized in the current literature. Although there is no clear demarcation between types of interventions focusing on different health domains, the chart provides a general classification based on the primary focus of the chatbots.

Fig. 3.

Summary of intervention types targeted by chatbots in underserved and vulnerable communities

Sexual and reproductive

Various chatbots [39, 46, 49, 50, 53, 54] have been developed to enhance knowledge and promote healthy behaviors related to sexual and reproductive health. These tools provided personalized and accessible information through education [49, 50, 53] and counseling [39, 46, 54] to adolescents, women, and young women facing barriers such as cultural taboos, limited health literacy, and restricted access to information. For instance, SnehAI [46] utilizing “Hinglish” (a blend of Hindi and English), created a safe space for discussing sensitive sexual and reproductive health topics and addressing misconceptions. through interactive education. Similarly, the Cervix Chat chatbot in the United States [54] primarily delivered counseling through text-based interactive conversations. By incorporating cognitive-emotional assessments, it identified psychological and informational barriers among patients following abnormal Pap smear results. Through simulating empathetic dialogues and providing accurate information about colposcopy procedures, the chatbot helped reduce anxiety, addressed common questions, and enhanced patients’ awareness, confidence, and adherence to follow-up care.

Mental health

Chatbots have played a notable role in delivering mental health support [41, 43, 45, 47, 48, 55] to underserved and vulnerable populations through interventions such as education [43, 47], counseling [48, 55], treatment support [41], and combined services including counseling, screening, and treatment support [45].

For example, ARU in India [47] during COVID-19 restrictions and STARS in Jordan [55] through education and counseling deliver personalized stress management strategies to alleviate psychological distress among students and young adults. Additionally, in Peru, ChatBot-Juntos [45] leveraged social messaging platforms to identify and screen individuals with psychological distress, provide Psychological First Aid (PFA), offer grief support, and facilitate referrals to specialists. These tools collectively demonstrated their effectiveness in addressing gaps in mental health services for underserved and vulnerable populations.

Immunization awareness and uptake

In this domain, chatbots such as Bablibot in Pakistan [42] and CHEC-UP in the United States [52] were developed to address linguistic, geographical, and informational barriers for children caregivers of underserved and vulnerable population, through health monitoring [42, 52] including automated vaccination schedule reminders, provision of information on vaccine side effects, and treatment support [42, 52] via referrals to human specialists. These innovations have notably increased vaccination rates and improved child health outcomes in various regions.

HIV prevention

The Let’s Chat chatbot [56] primarily provided counseling guidance and targeted cisgender women and transgender men. It employed interactive risk assessments that encouraged users to reflect on their own and their partners’ behaviors that may increase HIV risk. Additionally, the chatbot provided confidential guidance on pre-exposure prophylaxis (PrEP), supporting informed decision-making and preventive behavior change. Similarly, the Nolwazi bot [44] focused on facilitating HIV self-testing by offering culturally sensitive instructions and supportive counseling. Through ongoing interactions, this chatbot strengthened prevention efforts by linking individuals with positive test results to care and providing PrEP-related information to those with negative results, thus contributing to early diagnosis and secondary prevention.

Violence against women

The AyudaMujer chatbot in Peru [51] was designed for women who are victims of physical, sexual, or psychological violence. Through screening to identify the type and severity of violence, as well as providing treatment support and connecting users to specialists, this chatbot successfully contributed to reducing violence.

Healthy lifestyle promotion

The chatbot Carmen in the United States [40] primarily through counseling and continuous interactions and providing personalized feedback to encourage regular physical activity increased user motivation and prevented barriers such as social anxiety.

Chatbot design

The review identified 18 unique chatbots with distinct names, such as Gabby, Nurse Nisa, and CervixChat, primarily distributed across social media platforms [42, 44, 46–49] (n = 6), including Facebook (n = 3), WhatsApp (n = 2), and Telegram (n = 1), as well as web-based platforms [39, 43, 52, 54–56] (n = 6), with three chatbots [41, 51, 53] (n = 3) accessible via mobile applications, one chatbot [40] available through software installed on dedicated center computers, and one chatbot [45] designed as a hybrid, accessible simultaneously through web, social media, and mobile application platforms.

All chatbots utilized text-based input for user interaction. The majority [40, 43, 47–50, 52–56] (n = 11) employed predefined response options as the primary input method. Two chatbots [42, 51] (n = 2) utilized free-text input, while five chatbots [39, 41, 44–46] (n = 5) used a hybrid approach combining predefined options with free-text capabilities. Furthermore, five chatbots integrated Embodied Conversational Agents (ECA) alongside text-based interfaces. Gabby [39] featured an animated character with nonverbal cues, while Carmen [40] used an animated character with simple speech and also nonverbal behaviors on a touchscreen, supporting Spanish and English. ARU [43] employed an animated avatar with Hindi support and gestures, and Nthabi [50] an adaptation of Gabby for Lesotho, included nonverbal cues. Toptty [48] offered users a choice between male and female avatars, enhancing interaction through gamification.

Among the 18 chatbots examined [39, 40, 43–45, 47–50, 53–56] (n = 13) delivered responses based on rule-based systems, four [41, 42, 46, 51], were designed as hybrid systems combining rule-based and Natural language processing(NLP) approaches, and only one [52] exclusively utilized NLP for its development.

The chatbots supported multiple languages beyond English, including Spanish [40, 52, 54], Hindi [43], isiZulu [44], and a combination of Swahili, Hindi, and French [49]. Additionally, some chatbots were tailored to specific regions, supporting local languages such as Roman Urdu [42] and Hinglish [46].

Evaluations of chatbot

Among the reviewed studies, only nine [42, 44, 47, 50, 52–56], evaluated chatbots. Six of these [42, 44, 47, 53, 54, 56] reported multiple outcomes, with the most frequently evaluated being acceptability (n = 6) and feasibility (n = 5), followed by satisfaction (n = 3), usability (n = 2), and technological performance (n = 1). The majority of studies (7 out of 9) employed mixed quantitative and qualitative approaches to evaluate chatbot design. One study utilized a purely qualitative focus group method to evaluate acceptability, while another used a fully quantitative 22-item Likert-scale questionnaire to evaluate usability and acceptability. Quantitative evaluation metrics predominantly consisted of surveys (5 out of 8 studies). For qualitative approaches, interviews were the most common method (5 out of 8 studies), followed by open-ended questions in surveys or questionnaires (2 out of 8 studies), with only one study employing focus group discussions.

Discussion

Although the use of chatbots in healthcare, particularly for underserved and vulnerable populations, is still in its emerging stages, we identified 18 studies from 2015 to 2025, with 15 of these conducted after 2020, indicating a notable increase in research in this field in recent years. This scoping review aimed to report key findings from prior studies on the development and use of chatbots for underserved and vulnerable populations. The reviewed studies demonstrate that the chatbots employed varied in terms of targeted populations, health domains, healthcare delivery methods, chatbot design, and evaluation approaches. The target populations primarily included adolescents and young adults, women, young women, ethnic and racial minorities, and caregivers of children. The majority of the reviewed studies focused on adolescents and young adults, and women, reflecting the specific needs, vulnerabilities, and gaps in traditional healthcare systems for these groups. Adolescence and young adulthood represent critical periods of psychological, physical, and social development [57]. Many health risks, such as anxiety, depression [58], and high-risk sexual behaviors [59] often emerge during these years. However, adolescents frequently encounter challenges in accessing information from healthcare systems [60]. In this context, chatbots can partially bridge the communication gap between adolescents and healthcare systems by providing reliable and accessible information [61]. Similarly, women, as a vulnerable group, face challenges including cultural taboos (social stigma) [62], low health literacy [63], and limited access to culturally appropriate care [64]. By offering a private and non-judgmental environment, chatbots have enhanced women’s access to accurate and personalized information in areas such as contraception, safe abortion, gender-based violence, and mental health, utilizing commonly used platforms (e.g., WhatsApp, Facebook, Telegram). Numerous chatbots have been implemented within social media platforms, such as Facebook Messenger and WhatsApp. These platforms, owing to users’ prior familiarity and minimal training requirements, rank among the most widely utilized and have garnered considerable attention across various domains, including healthcare. Social media platforms, particularly Facebook, have emerged as primary sources of public information, including health-related content, especially for adolescents and young adults [65]. Similarly, a study investigating the application of Twitter and Facebook in university-based healthcare education concluded that these platforms are regarded as potentially effective supplementary tools in higher education for healthcare training [66]. WhatsApp is broadly accessible to users in the contemporary digital landscape. It is compatible not only with smartphones but also with basic mobile devices, rendering it an inclusive tool for diverse societal groups [67]. Moreover, compared to other social media platforms, WhatsApp requires lower data consumption [68]. This feature enables users to engage in seamless and continuous communication without concerns regarding excessive data usage [49].

An examination of healthcare delivery methods implemented across various health domains revealed that counseling and education constituted the largest proportion. This finding is consistent with prior studies, as enhancing awareness and knowledge in health-related areas, particularly sexual and reproductive health and mental health, is recognized as a fundamental principle in correcting misconceptions, promoting behavioral change, and strengthening self-care [69, 70]. The significance of this lies in the fact that many health issues and disorders arise from insufficient awareness, persistent erroneous beliefs, or unhealthy behaviors [71], particularly among underserved and vulnerable populations who face numerous barriers to accessing health information [72]. Furthermore, interventions such as treatment support, health monitoring, and screening are among the critical and valuable approaches designed to be delivered through chatbots for these populations.

All reviewed chatbots utilized text-based interaction as the primary mode of user engagement. Among these, the use of predefined options was the most prevalent method for data input. In several studies, alongside predefined options, the capability for free-text input was also supported, primarily employed for capturing patients' demographic information. This approach is particularly justified in the context of healthcare for underserved and vulnerable populations, as the selection of predefined options not only simplifies chatbot interaction for users but also mitigates common barriers such as limited internet access, low digital or health literacy, and challenges associated with typing on smart devices.

Based on the evidence, of the 18 chatbots examined, 13 were designed using rule-based systems. Four others operated as hybrid systems, leveraging both predefined rules and NLP, while only one chatbot was entirely NLP-based. The predominance of rule-based chatbots indicates that, in the healthcare domain, particularly when addressing underserved and vulnerable populations, simpler and more structured approaches remain preferred. This preference is attributed to their greater compatibility with weak communication infrastructure (e.g., limited internet access), higher reliability, and lower likelihood of errors in data collection or the provision of initial counseling. Notably, the number of studies utilizing hybrid approaches exceeded those relying solely on AI-based methods. One possible explanation for this trend is the need to maintain simplicity for specific, structured tasks where pre-formulated responses suffice, eliminating the need for more complex interactions that demand higher literacy levels or intricate analyses dependent on advanced technical infrastructure. In contrast, AI capabilities are typically employed as a complementary approach to manage unpredictable scenarios and more complex interactions. However, the limited adoption of AI-based chatbots reflects a notable challenge. While the development of such chatbots could facilitate more natural and personalized interactions, factors such as high development costs, technical complexities, and concerns regarding data privacy and security have hindered their widespread implementation.

Support for local languages alongside English underscores the necessity of adapting chatbots to the linguistic needs of target users. This is especially important, as low literacy levels and limited proficiency in English represent major barriers to accessing healthcare services for underserved and vulnerable populations. Consequently, incorporating local language support, constitutes a social, cultural, and ethical imperative. Such an approach enhances user acceptance and trust, thereby improving the effectiveness of health interventions and promoting health equity within communities.

Among the studies included in this review, only nine addressed the evaluation of chatbots. This highlights the limited evidence available regarding the evaluation of chatbots designed and developed for underserved populations and vulnerable groups. It is noteworthy that evaluation not only contributes to improving the design and development of chatbots but also facilitates the measurement of their effectiveness in achieving health-related objectives [73]. In many instances, there is no need to design and develop an entirely new chatbot for a target population, as customizing existing chatbots and conducting evaluations can determine their suitability and effectiveness for similar populations. An example of this approach is the adaptation of the Gabby chatbot in the United States, which was localized as Nthabi in Lesotho. This case demonstrates that scientific evaluation of customized versions not only enables resource-efficient development but also confirms their effectiveness and appropriateness for similar populations in different regions. The majority of studies focused on evaluating user acceptance and the feasibility of chatbots. Furthermore, most of these studies examined multiple outcomes, highlighting the importance of evaluating chatbots from different perspectives. Evaluation methods for chatbots have varied due to differences in evaluation criteria. Nevertheless, quantitative methods (primarily through questionnaires), qualitative methods (mainly via interviews), and mixed methods have been identified as the three predominant and common approaches in chatbot evaluations.

Most studies have been conducted in the United States, followed by Brazil, India, and Lesotho; however, the number of studies in these countries remains relatively small, the absence of research in other regions, particularly in low- and middle-income countries and in the Middle East, is especially striking. Given the potential benefits of chatbots, further studies are recommended to enhance access to healthcare for underserved and vulnerable populations.

Limitations and strengths

To the best of our knowledge, this review is the first to provide an overview of chatbots utilized to enhance access to healthcare among underserved and vulnerable populations. The review presents inventories of chatbots employed in this domain, categorized based on their features and objectives. It also delineates the evaluation approaches for chatbots and the challenges associated with their applications. Our review not only examines the current state and applications of chatbots but also contributes to analyzing their multifaceted implications, including their impact on health outcomes. However, this review did not consult healthcare professionals or other key stakeholders, which may have limited the comprehensiveness of the findings. Additionally, the long-term impact of chatbots on underserved and vulnerable populations remains unclear. Some systemic barriers, nuanced user needs, and contextual challenges may not have been captured, representing potential limitations of this review. Due to the absence of a formal classification of existing chatbots, the categorization of their features and objectives was primarily based on a narrative analysis of the studies.

Conclusion

This review identified 18 studies on the development and use of chatbots for underserved and vulnerable populations, targeting six health domains related to youth and adolescents, women, young women, ethnic and minority populations, and caregivers of children. The focus of these chatbots on key topics such as mental health and sexual and reproductive health represents an a well-designed intervention to the previously unmet needs of these groups. The review demonstrates that chatbots, as an innovative tool in the health sector, particularly for underserved and vulnerable groups, have considerable potential to improve access to healthcare services. The use of diverse platforms, combined with features like simple language, immediate responsiveness, and low cost, has made chatbots an effective and popular option among these populations. Although this field is still in its early stages, preliminary studies have reported positive outcomes in terms of feasibility, acceptability, and enhanced access to healthcare services. However, only a limited number of studies have employed clear criteria to evaluate chatbot performance. This highlights the necessity of a comprehensive and multidimensional evaluation framework to guide the development and use of chatbots for underserved populations, improve access to healthcare, and ultimately promote health equity.

Authors’ contributions

SH: Conceptualization, Methodology, Formal Analysis, Investigation, Visualization, Validation, Project administration, Writing – Original Draft Preparation, Writing – review & editing. ML: Methodology, Formal analysis, Visualization, Writing – original draft. SB: Data curation, Formal Analysis, Investigation, Writing – Original Draft Preparation. HVL: Data curation, Formal analysis, Writing – Original Draft Preparation. AH: Conceptualization, Search Strategy Development, Methodology, investigation SAFA: Conceptualization, Methodology, Formal Analysis, Supervision, Project administration, Writing – Original Draft Preparation, Writing – Review & Editing. The authors read and approved the final version of the manuscript.

Funding

There is no fund.

Data availability

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

This study was reviewed and approved by the review board and the ethics committee of Iran University of Medical Sciences (IR.IUMS.REC.1404.596).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.