Prominent narratives of generative artificial intelligence (AI) highlight these tools’ cost-saving potential for health organizations via more efficient communication. However, even if a particular tool were to perform adequately at one point in time, evaluating models across updates and upgrades is complex. There are few examples from industry that are feasible enough and few examples from academia robust enough for health organizations. To help fill this gap, we examined the performance of ChatGPT to identify the presence or absence of 6 evidence-based practices from the Centers for Disease Control and Prevention (CDC) Clear Communication Index (CCI) in short-form texts. First, we carried out a prompt engineering process to identify optimal approaches for prompting ChatGPT to correctly label the clear communication practices in a test dataset of 27 tweets, from simulated front-end user and back-end developer perspectives. Using expert labels as comparison points, we found front-end users and back-end developers would reach different conclusions about how to best prompt ChatGPT. Then, we investigated how accurately different versions of ChatGPT’s models could label these clear communication practices in a data set of 260 tweets. Using expert labels as comparison points, we found stable performance across two dated GPT-3.5 model versions, albeit with wide confidence intervals. Even so, we found that GPT-4 models could actually perform significantly worse on some CDC CCI labeling tasks. Our results imply the need for increased investment in health communication experts to evaluate generative AI performance over time, which may be less practical than investing in such experts to communicate with patients and engage communities directly.
While artificial intelligence (AI) chatbots provide quick answers to health questions, they raise equity concerns about what information is visible, who controls it, and how it is presented. This perspective examines AI as an information intermediary and highlights areas for improvement. We introduce Equitable Visibility Optimization (EVO), a public health framework that organizes the distinct responsibilities of health organizations, AI developers, information consumers, clinicians, and educators to improve the visibility of credible health information in AI systems. We recommend that health organizations conduct visibility audits to ensure their content is discoverable by AI systems; that health information users are trained in AI literacy; that AI developers adapt models to support multilingual users; and that AI developers incorporate fact-checking training loops that systematically weight verified health data to reduce misinformation exposure. Ultimately, EVO promotes shared responsibility across sectors to strengthen accountability, improve discoverability, and center equity as AI technologies evolve.
Despite experiencing higher obesity-related mortality than women, men remain underrepresented in behavioral weight-loss studies. This scoping review examined recruitment strategies and intervention characteristics used to engage men in behavioral weight-loss interventions and identified common features among studies that successfully recruited men compared with those that encountered challenges. Thirty-four studies met the inclusion criteria and varied in setting, design, and participant characteristics. Recruitment was most successful when interventions were developed through formative research, aligned with men's identities, values, and priorities, and also clearly communicated these characteristics through recruitment materials. Common design principles included framing participation around health, strength, physical performance, and family responsibility; incorporating meaningful social identities such as sports fandom, fatherhood, or cultural identity; partnering with trusted community organizations; using credible interventionists; fostering social support and accountability; and providing flexible intervention formats. No single recruitment channel consistently outperformed others. Instead, channel effectiveness depended on the target population and intervention context. Passive recruitment methods were generally effective in settings with established social networks, whereas active community outreach was often necessary to engage populations historically underrepresented in behavioral weight-loss research. These findings suggest that improving male representation requires attention to both recruitment approaches and intervention design. Researchers should consider these processes together when developing behavioral weight-loss interventions for men and improve reporting of recruitment methods and outcomes to strengthen future evidence.
Cardiovascular disease (CVD) is the leading cause of death in the United States and disproportionately affects racial/ethnic minoritized populations, low-income communities, older adults, and rural residents. Behavioral interventions can reduce cardiovascular risk, yet traditional programs face scalability barriers, including high costs and limited reach. This narrative review synthesizes evidence on the efficacy and equity of patient-directed AI tools that support health education and behavior change for cardiovascular health. Using PRISMA-ScR elements, we searched 10 databases (January 2010 to December 2025), yielding 19 empirical studies evaluating Generation 2.0 AI systems aligned with the American Heart Association’s Life’s Essential 8 (LE8) framework. Of these, five were randomized controlled trials (RCTs), and 14 were nonrandomized (quasi-experimental, observational, or early-stage) studies. AI-enabled interventions such as conversational agents, machine-learning models, and reinforcement-learning systems addressed hypertension, diabetes, heart failure, and related conditions. Among the RCTs, one demonstrated superiority of an AI-powered program over usual care (AI-delivered Diabetes Prevention Program), whereas the remaining four reported outcomes equivalent or non-inferior to standard behavioral care. Effects were most apparent in high-intensity interventions that paired machine learning–based personalization with structured behavioral curricula, although the multi-component nature of these programs limits inference about AI-specific contributions. Across nonrandomized studies, greater engagement intensity was consistently associated with improved cardiometabolic outcomes, although causal attribution to AI-specific mechanisms remains limited. Most studies had small samples, short follow-up, and limited reporting on safety and equity. Technology-intensive interventions were concentrated in higher-resourced settings, and populations with limited health literacy showed reduced engagement. Overall, evidence supports cautious optimism that AI can meaningfully scale personalized cardiovascular prevention; however, realizing that equitable and clinically meaningful impact will require rigorous trials with longer follow-up, explicit behavioral science integration, safety-conscious governance, and equity-centered strategies addressing digital determinants of health.
Individual characteristics and social drivers of health (SDOH) likely interact to shape 24-hr movement behaviors (24HMB), yet most research has focused on single factors. We identified intersectional subgroups and examined disparities in meeting physical activity (PA), sedentary behavior (SB), sleep, and combined 24HMB recommendations. We performed a cross-sectional latent class analysis of 2022-2023 National Survey of Children's Health data (n = 68,000, ages 6-17) using sex, race/ethnicity, caregiver education, household language, and food sufficiency. Survey-weighted logistic regression estimated class differences. Five classes emerged, with labels emphasizing the indicators that most differentiated profiles: Class 1 (52% of the population; high education, non-Hispanic [NH] White), Class 2 (10%; high education, NH Asian), Class 3 (5%; low education, NH White), Class 4 (20%; low education, Hispanic/Black), and Class 5 (12%; non-English, low education, Hispanic). The prevalence of meeting PA recommendations (19.48%) was lower than meeting SB recommendations (49.93%) and sleep recommendations (64.52%). Compared with Class 1, odds were lower in Classes 2 and 5 (ORs = 0.56-0.64) for meeting PA recommendations, lower in Classes 3-5 (ORs = 0.56-0.72) for meeting SB recommendations, and lower in all other classes (ORs = 0.49-0.76) for meeting sleep recommendations. Relative to Class 1, disadvantaged classes had higher odds of meeting none (ORs = 1.39-2.05) and lower odds of meeting all three (ORs = 0.48-0.63) recommendations. Findings highlight the need to consider combinations of individual characteristics and SDOH rather than one-size-fits-all approaches to improve 24HMB adherence across diverse groups.
