Data for this study were collected from Instagram, a Metaverse photo-sharing social media platform. To source Instagram posts, our team utilized CrowdTangle Search, a public insights and third-party web-scraping tool with access to Metaverse data. We leveraged CrowdTangle to filter through all publicly accessible English-language posts that contained one of 2 queries (similar to a library search): (1) “mpox,” the updated CDC-supported name for the virus, or (2) “monkeypox,” the former name of the virus. Data were collected for posts dated from May 5, 2022, through January 17, 2023. English-language posts were identified based on caption language. This study used publicly available social media data. No private or identifiable user data were collected or reported. All data were analyzed in aggregate, further described below, and no individual accounts are identifiable in this study. A total of 18,616 posts met these inclusion criteria and were downloaded for use in 2 iterative pilot coding rounds, and for the final randomized data sample, as specified below. As of August 14, 2024, CrowdTangle has been shut down and replaced by the Meta Content Library, which has more limited access.

Our study employed a manual qualitative deductive coding procedure to review the data. “Deductive coding” refers to a typology of qualitative research where a researcher begins with a theoretical framework, sets of hypotheses, or previously established coding schemas. For our study, we relied on a pretested codebook that had been applied to previous Instagram-related research studies, including pre-exposure prophylaxis for HIV prevention and skin cancer [13,14]. We adapted this codebook to reflect discourse pertaining to mpox, which was further stratified along several key dimensions, including source characteristics; mpox risk factors; prevention methods; treatment methods, benefits, and barriers; stigma; and citation attributions. These dimensions reflected our coding schema. Each post included for analysis was compared against our rubric. Prior to analyzing the final dataset, Cohen κ statistics were analyzed for interrater reliability among our human coders using RStudio (Posit, PBC, formerly known as RStudio, Inc). The full codebook with associated Cohen κ values can be found in Multimedia Appendix 1. For the final analysis, IBM SPSS 26 was used to analyze descriptive statistics. For RQ3, specifically, tests for statistical differences (ie, Wald H0 differences in proportion) between independent sample proportions were run between the top 10% most-liked posts in the sample and those in the bottom 90% of liked posts. The “top 10% most-liked posts” were defined as the subset of posts in our analytic sample (N=1000) with the highest number of likes. Specifically, posts were ranked by total like count, and the top decile (n=100 posts) was classified as “most-liked,” while the remaining 90% (n=900 posts) served as the comparison group. This approach allowed us to examine differences in content characteristics between highly visible posts and the broader corpus. Wald tests were used due to a large sample size and suitability for comparing independent proportions. Sparse categories were interpreted cautiously. Mean differences (MDs) and associated P values are reported.

Two human coders examined profile source and message content characteristics. The 2 coders and a moderator conducted 2 iterative pilot rounds using the most chronologically recent posts in the downloaded CrowdTangle dataset, where both coders coded the same set of posts and were assessed for interrater reliability. After the second round, among 72 variables and 63 out of 70 coded posts that met inclusion criteria, Cohen κ statistics ranged from −0.17 to 1.0, with a mean κ of 0.70 and median κ of 0.83, indicating substantial to perfect agreement across the majority of variables. Variables with none to slight agreement after the second round of pilot coding included posts’ cues to action, portrayal of mpox risk factors, diagnosis- and treatment-related content, and the extent to which posts conveyed the seriousness of mpox. Variables with low interrater reliability (eg, cues to action, risk factors, and seriousness) were excluded to preserve analytic validity. Prevention-related variables were retained and serve as proxies for risk-related content.

Following the 2 pilot rounds, a total of 1200 Instagram posts were randomized and split between the 2 independent human coders with the intention of achieving a final dataset of 1000 for analyses. A sample of 1000 posts was selected to ensure adequate representation across content features, and this sample size is consistent with prior Instagram-based content analyses using manual (human) coding approaches [13,14]. Coder 1 coded 499 posts and Coder 2 coded 501 posts that ultimately met the inclusion criteria. We excluded posts that did not mention mpox or were not written in English. Coded variables included, but were not limited to, source characteristics (ie, organization type [eg, business and nonprofit] and individual type [eg, physician and journalist]) as well as mention of mpox causes, prevention methods, and treatment. The determination of the type of individual or organization behind the profile was based on self-described information in the account biography and/or post content (eg, credentials provided such as MD, journalist, or organizational affiliation), consistent with prior Instagram content analyses [13,14]. Photographic content was coded based on visible features within the image (eg, mpox lesion depiction, presence of individuals, and contextual clues), following predefined codebook criteria. Hashtags were extracted from post captions and analyzed descriptively. Frequencies of hashtag usage were computed, and the top 10 most common hashtags were identified. The full codebook can be found in Multimedia Appendix 1.

This study utilized publicly available Instagram data and did not involve direct interaction or intervention with human participants. The study was reviewed by the Institutional Review Board of Indiana University-Bloomington, affiliated with the last author and was determined not to constitute human subjects research.

Regarding solution framing, just over one-third (384/1000, 38.4%) of posts promoted at least one prevention method, most commonly the mpox vaccine (263/384, 68.5%), followed by avoiding close contact (56/384, 14.5%), avoiding contact with objects (32/384, 8.3%), and abstaining from sexual activity (29/384, 7.6%) and condom use (5/384, 1.3%). A sizable number of prevention-related posts highlighted barriers to prevention methods (111/384, 28.9%).

A very small percentage of the sample mentioned mpox treatments (25/1000, 2.5% mentioned traditional biomedical treatments, such as antiviral medication; 2/1000, 0.2% mentioned alternative treatment methods). Of those posts that were coded as having depicted traditional biomedical treatments for mpox (n=25), about one-third mentioned the benefits of treating it (eg, “antiviral drugs licensed for smallpox use are effective and can be used against monkeypox”). Among these same posts, 28% (7/25) mentioned barriers to receiving treatment (eg, “WHO does not recommend vaccinating against mpox” and “there is currently a limited supply of JYNNEOS, also known as Immune or Imvanex”).

Due to poor interrater reliability regarding depictions of mpox causes on Instagram, this element of the research question was not directly addressed. However, the prevention variables do capture aspects of risk, such as close exposure to infected persons (eg, skin-to-skin and exchange of bodily fluids) and objects used by infected persons.

Statistically significant differences were identified between the top 10% most-liked posts and the rest of the posts in the sample. One of the most notable and expected differences is that the top 10% most-liked posts primarily originated from public figures or celebrities (MD=−0.591; P<.001) if they originated from an individual’s account, compared to the rest of the posts. Among organizational accounts, the top 10% most-liked posts were more likely to be coded as a business (MD=−0.299; P<.001) or news/media organization (MD=−0.350; P<.001). Among the bottom 90% liked posts, posts were more likely to originate from official governmental accounts (MD=0.222; P<.001), nonprofit organizations (MD=0.129; P=.006), and health care organizations (MD=0.102; P=.005).

A statistically significant higher proportion of the top 10% most-liked posts included a nonmoving photo/image in their posts, as opposed to the rest of the posts (MD=−0.119; P=.024), included visual depictions of mpox on body parts (MD=−0.081; P=.035), mentioned prevalence of mpox (MD=−0.18; P<.001), and cited sources (MD=−0.162; P=.001).

This study aimed to evaluate the content and source characteristics of mpox-related posts on Instagram. Broadly, our results highlight the quantity and diversity of health-related content that can be found online, including the range of content types, sources, and message framing within mpox-related Instagram posts. We contextualize our findings in relation to the existing academic literature below.

Social media maintains an unfortunate reputation as a harbinger of misinformation, social anxiety, and political divide [15]. Numerous studies have highlighted that increased polarization in the United States, and the rise of misinformation can largely be attributed to excessive social media use and poor content moderation [16]. However, other studies have highlighted that, in a health communication context, social media can potentially transcend numerous cultural, structural, and systemic barriers to care to promote knowledge and improve health outcomes [17]. This observation has been previously made in computational and qualitative studies pertaining to addiction and recovery [18], pre-exposure prophylaxis for HIV prevention [19], reproductive decision-making, and vaccine communication [20]. Collectively, these studies have observed that helpful content can offset poorly vetted or low-quality information. However, certain criteria must be met, including evidence of a post originating from a verified account, culturally adaptive messaging, and the presence of online, real-time moderators to mitigate the risk of misinformation.

Our data largely contain evidence of accurate and insightful health promotion along the aforementioned criteria. For instance, within the top 10% of posts, much of our sample was coded as originating from a business entity or news organization. These posts were interpreted as part of information dissemination efforts during the early stages of the global mpox outbreak. In addition, we observed a substantive minority of posts related to treatment and prevention methods, most of which aligned with the standards set by the CDC and other similar entities. Importantly, our sample did not include hateful, vitriolic, or deliberately misinformed content about mpox. While this does not negate the potential existence of such content on Instagram, our findings suggest that, among highly engaged content, the information appears to be generally useful for individuals seeking answers about mpox risk, infection, spread, and treatment. Future research should consider expanding the scope to a larger sample of posts, which could help in identifying more generalizable findings or conducting a deeper qualitative review for information quality.

Broadly, our findings align with those of previous research illustrating the potential benefits of social media for cost-effective health communication and information sharing. For example, Jiang et al [21] highlighted that, despite known risks of social media and information quality, social media should play a continued and growing role in health communication for health promotion. As Valdez and Patterson [22] contended, the applicability of social media for health promotion will depend on the role of content moderation. On Reddit, for example, posts pertaining to health are highly moderated by trained professionals, which result in constructive and helpful messaging pertaining to myriad health behaviors, including those pertaining to sexual and reproductive health. Although it is unclear the extent to which Instagram is moderated by such professionals, our results at least illustrate that the platform, and likely the Metaverse more broadly, may be creating such avenues for informed health content. However, we acknowledge that caution must be undertaken given documented research citing Instagram’s photo-sharing capability as a venue for body image concerns.

Empirical work examining mpox on Instagram remains scarce. One study analyzing posts under #monkeypox documented early themes and misinformation risks, noting that stigmatizing and inaccurate narratives coexisted with public health messaging on the platform [22]. Compared with that work, our analysis uses a substantially larger randomized sample, a validated multidomain codebook adapted from prior Instagram health studies, and a direct comparison of the top 10% most liked posts to the remaining corpus, enabling inference about engagement-linked content characteristics (eg, source types, image features, prevalence mentions, citations).

Findings from adjacent platforms converge on several patterns that our study also observes. On TikTok, for example, multiple analyses reported variable information quality and prominent engagement with visually compelling content; videos created by physicians or authorities were relatively limited; and misinformation or conspiracy content appeared early in the outbreak, especially in trending feeds [23,24]. On X, formerly known as Twitter, studies have documented high volumes of mpox discourse with substantial misinformation and stigma toward LGBTQ+ communities, alongside network dynamics where news outlets and a small number of influencers shaped attention [11,25]. The analyses of Reddit have emphasized community problem-solving and information-seeking among gay and bisexual- and men who have sex with men–focused subreddits, often moderated to reduce misinformation and toxicity, with themes evolving as the outbreak and public health declarations unfolded [26]. Cross-platform work comparing Reddit and X found topic and psycholinguistic differences that map onto platform cultures—more community support and practical guidance on Reddit versus faster, more event-driven discourse on X [27]. Broader social media studies during the 2022 mpox emergency also traced toxicity dynamics and the need for targeted mitigation strategies, reinforcing the public health value of verified source cues and active moderation [28].

Against that backdrop, our Instagram results add 3 contributions: information regarding source composition, content features associated with engagement, and tone and quality signals. First, we show that highly engaged Instagram posts disproportionately originate from public figures/celebrities and news/business accounts, while government, nonprofit, and health care organizations populate the long tail—mirroring influence concentrations on X but documented here for Instagram with a larger coded sample. Second, nonmoving images (vs videos), visible depictions of lesions, explicit prevalence mentions, and the presence of citations all characterize the top-liked posts; prior TikTok and X studies did not simultaneously test this combination of features on Instagram. Third, unlike some cross-platform studies that reported on misinformation and stigma, our sample of highly engaged Instagram posts contained comparatively little hateful or deliberately misleading content, suggesting that—at least during the window we studied—Instagram’s most amplified mpox content skewed toward broadly informative material. This aligns with municipal and public health agency experiences that consistent, branded messaging can perform effectively on real-time platforms when paired with clear visuals and timely updates [29].

This study is not devoid of limitations. First, we were only able to collect information from publicly available Instagram accounts. Second, this study intentionally focused on the initial outbreak period (May 2022 to January 2023). Accordingly, an area for future research might incorporate more recent data. Third, our inclusion criteria may have limited the generalizability of our results. For example, we only included Instagram content written in English. Fourth, we acknowledge that likes are influenced by account size and reach. While engagement rates were not available via CrowdTangle, we interpret the findings as reflecting visibility rather than normalized engagement. Finally, given multiple statistical tests, findings should be interpreted cautiously as exploratory.

Despite these limitations, this study has several strengths and implications. First, this research remains the largest of its kind, to our knowledge, examining mpox on Instagram, one of the most used social media platforms in the United States. Only one other research study has examined mpox on Instagram, and that research focused on a smaller number of posts and was limited to assessing misinformation [30]. To date, a few other studies have examined mpox content via other social media platforms, including Reddit [31] and X [11,32]. Given that mpox-related posts by public figures and celebrities were among the most liked in this sample, medical experts could partner with celebrities and influencers to lead awareness campaigns on mpox.

Instagram has functioned as a consequential channel in the mpox information ecosystem. In a large, randomized sample of posts spanning the initial 2022 to 2023 outbreak period, we found that organizational accounts—especially news/media and businesses—and public figures disproportionately produced the most-liked content; visual stills, lesion depictions, prevalence statistics, and explicit citations were each associated with higher engagement; and prevention and treatment content—when present—generally aligned with authoritative guidance.

Taken together, these findings imply that public health communicators can improve Instagram reach and relevance by packaging guidance as clear static visuals (eg, infographics and single-image posts) that include succinct prevalence/context cues, citing sources (eg, WHO) prominently in the caption or image, and partnering with trusted creators and news accounts to copublish updates and frequently asked questions. Given cross-platform evidence of misinformation, stigma, and toxicity—particularly on TikTok and X—strategies that foreground verified provenance, rapid myth-busting, and community-informed moderation remain essential [23].

Finally, our engagement-contrast design suggests a practical playbook for future outbreaks: monitor which content attributes travel on each platform and then rapidly iterate message formats and partnerships accordingly. Future work should extend this analysis to reels and video formats, test message experiments (A/B) with creator collaborations, and link post-level exposures to downstream behaviors (eg, vaccine information-seeking and clinic clicks). As emergent mpox developments continue to surface—and new emergencies arise—Instagram can be leveraged as a scalable, equity-minded conduit for timely, accurate risk communication when campaigns adopt these evidence-based practices.