Public health decisions about intervention planning, resource allocation, emergency response, and long-term policy formulation are usually made by governmental public health experts, be they in local/tribal, state/territorial or federal positions. However, in some cases, public health-related decisions are made not at the expert level, but by an elected official such as a mayor or governor. Regardless of decision-maker or context, high-quality information and analytical capabilities for identifying, accessing, and processing data to support these decisions is necessary (1, 2). Artificial intelligence (AI) may be a valuable tool that public health officials and elected leaders can incorporate as one input to decisions. AI and machine learning can facilitate access to a broader set of data, more quickly, and may be helpful for an expert or leader to see a more complete picture of the available data and to understand tradeoffs inherent to a decision.

We define AI as a branch of computer science that focuses on building tools that represent characteristics of the human mind and have the capacity to complete intellectual tasks that humans can perform (3, 4). Machine learning (ML), natural language processing, and image and signal processing are all AI subfields that may be used individually or within a single application (5). Broadly, AI has shown potential to help public health experts and elected officials navigate complex decision-making environments and promote technological growth, foster sustainability, and improve quality of life (6, 7). Digital health interventions, such as telemedicine, wearable devices, electronic medical records, and mHealth, have already led to significant advances in public health research and practice, including affordable and accessible health education modalities and chronic disease self-management (8, 9). Digital health technologies generate massive amounts of data that AI-driven analysis may be used to better understand real-time symptom tracking, health services demand, and intervention efficacy (10, 11). AI may aid in analyzing large datasets for disease surveillance and prediction, contact tracing, and other activities that inform government programs, policies, and outbreak responses (12). Although numerous potential applications have been proposed in the literature, the use of AI in public health decision-making remains limited. While limited, AI has been used in a government context to predict child maltreatment (13), hospital patient loads, and forest fires (14). Potential further applications of AI in government and public services include using predictive analytics to plan for resource allocation (e.g. emergency responses to extreme weather events), improving public safety, and forecasting economic and market trends (15, 16). However, it is important to note that empirical documentation of actual implementation by governmental decision makers is sparse.

As was demonstrated clearly during the COVID-19 pandemic, public health crises can precipitate a shift of responsibility for some high-consequence public health decisions to elected leaders such as mayors and governors. Sometimes this shift in decision-making authority occurs because there is a complex set of non-public health factors (e.g. social, economic, political) that must be considered carefully. Sometimes this shift happens because of the intense public scrutiny that comes with emergency response- elected leaders may want to have more control over decisions when they will be held personally responsible for the outcomes (17). In these emergency situations, public health experts often provide support for decision-making as part of an advisory body. However, especially in urgent and complex emergencies, there may be a need for additional decision-support resources and tools for elected leaders, who may not have public health training or previous experience with response to a public health emergency. AI has significant potential as a tool for elected officials in making these public health decisions, especially when paired with a diverse set of expert advisors.

In this work, we present findings from a scoping review to assess current literature on AI applications for public health professionals and elected officials in both routine public health decisions and crisis decisions during an emergency response. Our focus is on elected officials, which we define as individuals in politically accountable positions with direct authority over policy decisions, while recognizing that these officials often work alongside public health professionals in the decision-making process. For this review, we include studies where AI tools were used to inform decisions ultimately made by elected officials or governing bodies. This review helps to contextualize current literature on how AI may be used to support public health decision making efforts, provides recommendations, and identifies areas for future research. The literature review aims to address the following research questions:

For this review, we defined “governmental public health decision-making” broadly to encompass decisions made by individuals with statutory or delegated authority to implement public health interventions, allocate resources for public health programs, or establish public health policies. This includes:

Using the MeSH database (19) as a starting point, we developed a keyword list (Table 1) to represent key concepts for the scoping review, namely “Artificial Intelligence,” “Government,” and “Public Health” in the PubMed Database. Given the diversity of government contexts that AI may be deployed in, we considered keywords related to local, state, and federal government along with government agencies.

The OAISTER database required that searches be no longer than 30 words, so we used the following adjusted query to meet these guidelines:

(“Artificial Intelligence” OR “Computer Reasoning” OR “AI (Artificial Intelligence)” OR “AI” OR “Machine Intelligence” OR “Machine Learning” OR “Deep Learning” OR “Supervised Machine Learning” OR “Unsupervised Machine Learning” OR “Data Science”) AND (“Federal Government” OR “National Government” OR “Government Agency” OR “Municipal Government” OR “State Government” OR “State Governments” OR “Provincial Government” OR “Local Government” OR “City Government” OR “Government, County” OR “Metropolitan Government” OR “Official” OR “Decision Maker” OR “Decision Making” OR “Government”) AND (“Public Health” OR “Community Health” OR “Environment, Preventive Medicine and Public Health” OR “Outbreak” OR “Pandemic” OR “Epidemic” OR “Outbreak Response” OR “Emergency”).

All literature identified through the searches underwent title and abstract screening, conducted by 2 separate reviewers, to determine whether each met the criteria for full-text review. A full-text review of applicable articles was conducted in Covidence (20) by 2 separate reviewers to assess if they merited inclusion in the final analysis. Data from the articles that merited inclusion was extracted into a form by 2 separate reviewers. At each stage of the analysis process, any disagreements between the reviewers were resolved by discussion and consensus among the 2 initial reviewers or, if consensus could not be reached, via adjudication by a third reviewer. The data extracted included the study period, country or location of study, AI intervention, study design, governmental decision making that the AI intervention supported, and limitations. Because scoping reviews aim to map the breadth of available evidence rather than assess study quality or synthesize effectiveness, we did not conduct formal quality appraisal of included studies.

Based on the limited evidence identified in this review, several critical questions remain unanswered. None of the reviewed studies included comparative analyses or outcome evaluations, preventing conclusions about whether AI-supported governmental decisions led to better public health outcomes, improved decision quality (accuracy, comprehensiveness, timeliness), or offered advantages over existing analytical methods and expert judgment. The studies describe AI tool development and deployment but do not assess impact on population health metrics, epidemic control, or the actual influence of AI outputs on policy decisions. Additionally, we lack evidence on adoption patterns. This includes rates of sustained operational use, factors influencing governmental officials' choices to utilize AI tools, and the pathway from AI-generated insights to enacted policies.

All seven included studies focused exclusively on COVID-19 pandemic response, preventing assessment of whether these applications would transfer effectively to routine public health decision-making, other types of emergencies, or different governmental contexts. The unique characteristics of the COVID-19 pandemic (unprecedented data availability, intense public attention, rapid technological mobilization) may limit generalizability. Furthermore, while equity concerns are discussed theoretically in the broader literature, no studies empirically assessed whether AI use affected health equity, exacerbated disparities, or distributed benefits and burdens equitably across populations. These substantial gaps highlight the nascent state of evidence in this field. While AI tools have been developed and deployed during the COVID-19 pandemic, rigorous evaluation of their role, effectiveness, and impact in governmental public health decision-making has not yet been conducted.

The challenges and opportunities discussed below synthesize insights from the included studies alongside broader literature on AI governance and public health decision-making. Many of these considerations extend beyond what was explicitly documented in the seven included studies and represent areas warranting attention in future research and practice. The identified gaps in the literature are a catalyst for distinguishing future areas for growth. Artificial intelligence presents a strategic opportunity to enhance public health decision-making and particularly public health crisis response practice. First, advanced surveillance systems can analyze diverse data sources simultaneously to detect emerging disease patterns before they become apparent through traditional monitoring approaches (33, 34). Second, customized health communication becomes achievable as artificial intelligence generates messages precisely tailored to audience characteristics, thereby significantly improving information accessibility across diverse populations and cultural contexts (35). Third, workflow optimization through intelligent automation could liberate public health professionals from administrative burdens, enabling greater focus on community engagement and strategic initiatives (36). Fourth, data-driven decision support becomes more robust as AI models synthesize complex epidemiological, environmental, and social data to forecast trends and evaluate interventions (37, 38). These capabilities may provide public health leaders with unprecedented tools to strengthen early warning systems, communication effectiveness, operational efficiency, and evidence-based practice, which are all critical components for addressing complex population health challenges in resource-constrained environments.

Despite opportunities created through AI to transform public health, the perceived benefits are accompanied by substantial challenges that demand immediate attention that future work may investigate. The rapid evolution of artificial intelligence has outpaced regulatory frameworks, creating a governance gap that leaves both individuals and institutions vulnerable (36, 39). Traditional de-identification protocols are increasingly insufficient against sophisticated re-identification algorithms, threatening the confidentiality of sensitive health information (39). Similarly, artificial intelligence systems with built-in prejudices threaten to exacerbate health disparities particularly for underrepresented communities and could potentially lead to distorted disease forecasting and inequitable resource allocation (36, 37). Operational challenges such as lack of integration into public health systems, workforce training gaps, and inadequate infrastructure, further complicate the effective implementation of artificial intelligence in public health (34).

In this review, there were several strengths and limitations. Our findings are strengthened by the comprehensive scope and size of this review. The literature review included 13,239 papers at onset and each phase of the review had at least 2 reviewers. Next, our review investigated a thorough and well-defined research question that has not yet been explored in the literature. However, although the review screened over 13,000 papers, only 7 were included in the final extraction process of the review. The included studies only discussed AI/ML interventions in the context of COVID-19, so there is little diversity, and most studies were descriptive studies. The review process also removed literature reviews and conference proceedings which might have contained additional examples. This also means that there has been minimal highly rigorous evaluation of AI intervention in public health decision making. Another important limitation is the primarily descriptive nature of included studies, with limited evaluative rigor or empirical assessment of outcomes. Most studies documented AI use or described AI outputs but did not rigorously evaluate whether AI improved decision quality, timeliness, or public health outcomes. This constrains our ability to draw conclusions about AI's actual impact on governmental decision-making effectiveness.

Furthermore, methodological constraints may have introduced bias and limited the scope of our findings. First, our restriction to English-language publications may have resulted in geographic and cultural bias. Countries with substantial AI investment and governmental AI applications that publish primarily in languages other than English may have been underrepresented in our review. This language restriction may have caused us to miss important implementations and lessons learned from diverse governmental contexts.

Although we included Governmental and non-Governmental reports and white papers in our search criteria, we may have excluded important gray literature including conference proceedings. Conference proceedings, in particular, often report early-stage implementations and pilot studies that may represent important real-world applications. The exclusion of these sources, while necessary for maintaining study quality standards, may mean our review captures primarily academic research rather than operational practice. Future work should endeavor to improve on these limitations and examine future AI/ML interventions in more expansive public health decision making processes and contexts.

Finally, the challenges identified in our discussion of AI implementation drew from both the primary studies in our review and the broader adjacent literature. Given that only seven studies met our inclusion criteria and all focused narrowly on COVID-19, supplementing with relevant literature from AI ethics, public health informatics, and governmental technology implementation was necessary to provide comprehensive context. However, this means some challenges discussed may not have been explicitly identified in studies specifically examining AI use by governmental decision-makers. We have attempted to clearly distinguish between findings from our primary review and insights from adjacent literature throughout the Discussion.

The small number of included studies (n = 7), their exclusive focus on a single public health crisis (COVID-19), and their predominantly descriptive rather than evaluative nature severely limits our ability to draw general conclusions about AI's role in governmental public health decision-making. The current evidence base describes AI applications but does not evaluate their effectiveness, impact, or appropriate role in public health governance.