Use of Artificial Intelligence in Public Health Education for Pandemic Preparedness and Response

Type of study

This research was conducted as a scoping review, following the methodological framework proposed by Arksey and O’Malley and further developed by the Joanna Briggs Institute (JBI) [7]. The protocol was structured to ensure transparency and reproducibility of the process, following the steps: (1) identification of the research question; (2) identification of relevant studies; (3) study selection; (4) data extraction; and (5) collection, synthesis, and reporting of results.

Accordingly, the research question and key search elements for this review were developed using the PCC strategy, a mnemonic that helps identify the core topics: problem, concept, and context. In this review, the problem was defined as the use of AI‑based technologies; the central concept was public health educational strategies; and the context was the preparedness, response, and recovery phases in the setting of pandemics, epidemics, and other emergencies. Thus, the guiding research question was: What scientific evidence exists on the use of AI‑based technologies in educational strategies targeting the preparedness, response, and recovery phases of public health emergencies, including pandemics, epidemics, and other events of public health relevance?

Study selection

The articles were stored and organized in the Zotero© reference manager. For study selection, search results were independently screened by two researchers using Google Forms© and Google Sheets©. Discrepancies were resolved by consensus or with the involvement of a third researcher for adjudication. The researchers compared the results of their searches, verified differences in findings, and consistently sought to include the largest possible number of eligible studies.

Search and data collection

Searches were performed in major bibliographic databases, PubMed/MEDLINE, Scopus, Web of Science, Embase, IEEE Xplore, and LILACS, and complemented by a targeted search of gray literature in Google Scholar, limited to the first pages ranked by relevance, as well as institutional documents cited in the included studies, such as those from the World Health Organization (WHO), the Pan American Health Organization (PAHO), and the Centers for Disease Control and Prevention (CDC). The strategies combined controlled descriptors and free‑text terms, tailored to each database, with Boolean operators, truncation, and restriction to title, abstract, and subject fields. Filters were applied for the period from 2010 onward and for publications in English, Portuguese, and Spanish. Complete search strategies for each database were archived to support reproducibility, and searches were last updated in October 2025.

Eligible studies included primary quantitative, qualitative, or mixed‑methods research, as well as methodological studies and implementation reports that described the explicit use of AI for educational purposes in public health in either real or simulated pandemic scenarios. Narrative and integrative reviews and bibliometric studies were also included when they provided syntheses applicable to educational practice. We excluded reports without an educational component (for example, purely technical surveillance or diagnostic applications), editorials and opinion pieces without methodological or implementation contributions, studies without full‑text access, and publications in languages other than English, Portuguese, and Spanish.

Record management included export in RIS and BibTeX formats and integration into a bibliographic manager, followed by two‑step deduplication: an automated software process (based on title, authors, year, and DOI) and manual inspection for residual cases. Study selection was conducted in two phases by two independent reviewers: title and abstract screening based on eligibility criteria and full‑text assessment of potentially relevant records. Disagreements were resolved by consensus and, when necessary, by a third reviewer. Records without full‑text access, after attempts to contact authors or institutions, were excluded with appropriate justification. The eligibility flow was documented in a PRISMA‑ScR diagram.

Data extraction was performed using a pilot‑tested form that captured: identification (author, year, country), public health emergency addressed, target population and channel/setting, AI modality and educational purpose, platform and presence of human curation/supervision, phase of the pandemic cycle, operational and scalability aspects, study type, and main reported limitations. When available, we also recorded educational outcomes, engagement and acceptability metrics, references to equity and language, costs/implementation, and ethical considerations. Missing information was explicitly classified as “not reported (NR),” avoiding unsupported imputations.

The synthesis combined descriptive and narrative approaches, organized into three complementary tables: (i) characterization of studies (identification, context, population, and design); (ii) AI interventions and strategies with their educational objectives, channels, and supervision; and (iii) functional synthesis linking the educational role of each intervention to pandemic cycle phases, AI‑mediated pedagogical mechanisms, scalability/operability conditions, level of evidence, and key gaps. The interpretive narrative integrated convergent findings and contextual variations, as well as recurrent absences of standardized metrics for educational and behavioral effects, economic evaluations, and equity analyses.

Given the inherent scope of mapping reviews, we did not conduct a formal risk‑of‑bias assessment. To strengthen interpretability, we explicitly reported the study type and limitations declared by the authors (for example, restricted samples, dependence on specific platforms, language bias, and absence of clinical outcomes) and discussed their impact on the generalizability and applicability of the findings. This strategy supports methodological transparency and coherence between the aim of mapping the field and the level of inference that is appropriate based on the available body of evidence.

Role of artificial intelligence in public health education across pandemic phases

This study mapped evidence on the use of AI‑based technologies in educational strategies aimed at planning the preparedness, response, and recovery stages of public health emergencies. The results of the studies assessed in this review reveal the consolidation of AI as a health education tool in the context examined, enabling risk identification, optimization of responses, and personalization of educational content for health professionals and the general public. These findings are corroborated by a previous study that explored lessons learned during the COVID‑19 pandemic and offered insights into the use of AI as support in crisis scenarios [1].

Public health emergencies constitute serious threats to population health, as they can result in widespread dissemination of infectious agents, high morbidity and mortality, and substantial impacts on the reorganization of health services [36]. In this context, AI emerges as a strategic technology by supporting rapid, data‑driven decision‑making, enhancing risk identification, and guiding timely interventions in epidemic and pandemic events. However, despite its considerable potential, the effectiveness of AI solutions depends on structural factors, such as global collaboration, robust governance, compliance with ethical principles, standardization and validation of information, interoperability between systems, and assurance of equity in access to and use of technologies [1, 37, 38].

The literature indicates that AI has substantial potential as an educational resource and as support for actions across all stages of pandemic and epidemic management. In the preparedness phase, algorithms applied to the analysis of large datasets can be used to forecast outbreaks, inform public policy formulation, and support vaccine development [1, 38]. During the response phase, AI can optimize healthcare delivery through efficient resource management, risk stratification, and reduction of health workers’ exposure to infectious agents [15, 38]. In the recovery phase, these technologies contribute to impact evaluation, monitoring of clinical outcomes, and the promotion of continuing health education [15].

Among the AI systems identified in this review, there was a predominance of chatbots and generative models (such as ChatGPT, Dress‑COV, and Watson Assistant), underscoring the consolidation of conversational AI as a strategic tool for risk communication, triage, and health education. In addition to expanding access to information, these solutions can provide personalized guidance for behavior change, tailoring communication to users’ levels of digital literacy and promoting greater engagement.

However, the studies analyzed emphasize that, for their potential to be fully realized, such technologies must be developed and continuously refined with a focus on equity, accountability, and cultural sensitivity, ensuring reliability and fairness in their application within public health contexts [14].

Adaptive learning systems and intelligent simulators demonstrate high potential, particularly in the preparedness phase, by enabling personalized training and continuous learning, thereby contributing to the development of essential digital competencies for working in emergency situations [1]. Evidence indicates that these technologies adjust content according to individual performance, simulate realistic clinical scenarios, and monitor users’ progress, making teaching more efficient, accessible, and learner‑centered, including at scale, with a positive impact on timely response capacity during health crises [1].

In addition, in the recovery phase, such tools can support ongoing workforce development, enabling feedback loops in learning processes and the incorporation of post‑event lessons. Nevertheless, despite the substantial benefits, close integration between AI solutions and human supervision remains essential to ensure pedagogical quality, equity, and adherence to ethical principles, especially in high‑pressure contexts and situations involving critical decision‑making [38].

Structural imbalance across pandemic phases and implications for health system resilience

The evidence analyzed in this review indicates that AI has been used predominantly in the response phase of public health emergencies, with more limited application in the preparedness and recovery phases. During health crises, the demand for rapid and accurate decisions drives the use of automated technologies for data generation and analysis, resource allocation, triage, and decision support [39, 40]. By contrast, the preparedness and recovery phases require long‑term planning, systems integration, and sustained investment, elements that have historically received lower priority and funding from decision‑makers [40]. Overcoming this imbalance is essential for strengthening resilient health systems capable of anticipating risks, reducing vulnerabilities, and incorporating post‑crisis learning [32, 40].

This concentration of AI‑enabled educational initiatives in the response phase represents a structural limitation for health system resilience. When educational uses of AI are primarily activated during crises, they tend to function as short‑term, reactive tools rather than as components of sustained capacity building. Insufficient integration of AI‑based education into the preparedness phase limits the development of digital literacy, critical appraisal skills, and institutional familiarity with these technologies before emergencies occur. Likewise, the relative neglect of the recovery phase constrains opportunities for systematic learning, workforce requalification, and the incorporation of lessons learned into future training cycles. As a result, health systems risk reproducing recurrent patterns of vulnerability, entering successive emergencies without consolidated educational infrastructures capable of supporting anticipatory, adaptive, and equitable responses.

Evidence gaps, equity, and governance challenges in AI‑mediated education

With regard to educational evidence, the included studies exhibit considerable methodological heterogeneity and, in most cases, lack robust indicators for assessing the impact of AI on learning, behavior change, or improvement in clinical practice. A critical gap identified across the reviewed studies is the absence of standardized and validated metrics to assess educational outcomes of AI‑mediated interventions. Most initiatives rely on proxy indicators such as engagement, user satisfaction, or self‑reported knowledge gains, which limits comparability across settings and impedes cumulative evidence generation. Without agreed‑upon frameworks to measure competencies, skill acquisition, or changes in professional practice, it remains difficult to determine whether AI‑based educational tools produce meaningful and sustained impacts beyond immediate crisis communication [1, 22, 41, 39].

Equally underexplored are dimensions related to equity, cost‑effectiveness, and long‑term sustainability. Few studies explicitly examined whether AI‑enabled educational strategies reduce or exacerbate existing social and digital inequalities, particularly in low‑resource or marginalized contexts. Moreover, economic evaluations were rarely reported, and operational sustainability beyond emergency funding cycles remained unclear. This lack of analysis constrains the translation of promising pilot initiatives into scalable and durable public health education policies [17, 18, 22, 42, 43].

The review also identified a scarcity of validated instruments specifically designed to measure the effect of AI‑mediated educational interventions in pandemic contexts [41]. In this context, the literature recommends the development of metrics that capture not only engagement or knowledge gains, but also competency acquisition, patient safety, and health outcomes [1].

Finally, important gaps persist in governance models and ethical validation processes for AI‑mediated educational content. Although several studies acknowledged the need for human oversight, detailed descriptions of accountability mechanisms, validation workflows, data governance, and bias mitigation strategies were uncommon. The limited transparency surrounding these processes raises concerns regarding reliability, trust, and ethical compliance, particularly when educational interventions are deployed at scale during high‑stakes public health emergencies [3, 14, 18, 37, 39].

The studies analyzed also revealed a strong dependence on commercial platforms and limited transparency regarding ethical aspects and human validation processes for AI‑mediated educational content, underscoring the need for coordinated action at technical, institutional, regulatory, and societal levels to ensure governance, reliability, and accountability in the application of educational algorithms [39]. Although AI‑mediated education has considerable potential to mitigate digital inequalities during pandemics and other public health emergencies, by expanding access to knowledge and supporting the inclusion of diverse audiences, its implementation is still constrained by structural barriers such as insufficient technological infrastructure, shortages of qualified professionals, data limitations, and ethical challenges related to privacy and information security [42].

In this regard, the responsible advancement of these technologies requires recognition and mitigation of inequalities that may be reproduced or exacerbated by the inappropriate use of AI. For digital solutions to be genuinely inclusive, it is essential to incorporate digital determinants of health, such as usability, accessibility, interactivity, digital literacy, and technological availability, into their design, in close alignment with the social determinants of health, which shape opportunities for access, learning, and engagement with digital technologies [43]. Thus, the integration of these determinants is a fundamental requirement for promoting equity in access to digital health and strengthening health justice across different populations.

The establishment of strategic alliances among academic institutions, health services, the technology industry, and regulatory bodies demands ethical governance, system interoperability, open infrastructure, ongoing professional training, and collaborative research [3, 38]. Such efforts are essential to ensure that AI applied to health education is accessible, safe, and capable of responding appropriately to the global challenges posed by public health emergencies.

Limitations

This scoping review has some limitations that should be acknowledged. As a mapping review, it aimed to describe the breadth and characteristics of the available evidence rather than to assess effectiveness or causal relationships, and no formal risk‑of‑bias assessment was conducted. The heterogeneity of study designs, interventions, and reported outcomes limited direct comparison across studies. In addition, the reliance on published literature and selected gray literature may have resulted in the omission of unpublished initiatives or rapidly evolving implementations, particularly those developed during acute phases of public health emergencies. Finally, most included studies focused on high‑ and middle‑income settings, which may limit the generalizability of the findings to low‑resource contexts.