Quality and reliability of osteoporosis-related health information on short-video platforms: a multi-source cross-sectional study Authors: Yexi Shi, Wangnan Mao Published: 2026 ABSTRACT AbstractBackgroundOsteoporosis (OP) affects approximately 145.86 million people in China, imposing a substantial disease burden and broader societal costs. Effective management requires evidence-based patient education. Short-video platforms have become increasingly important channels for health communication, yet few studies have provided systematic comparisons of OP content across Chinese platforms using concurrent measures of overall quality, reliability, and transparency.MethodsThis cross-sectional study analyzed 300 OP-related short videos (100 each from TikTok, RedNote, and Bilibili). Two trained raters independently assessed all videos using the Global Quality Scale (GQS), modified DISCERN (mDISCERN), and Journal of the American Medical Association (JAMA) benchmarks; a third reviewer adjudicated disagreements. The primary outcomes evaluated in this study were GQS, mDISCERN, and JAMA scores. Scores were summarized primarily as median (interquartile range [IQR]) and compared using Kruskal–Wallis test, Dunn-Bonferroni pairwise comparisons, and Holm adjustment across the three omnibus tests. Spearman correlations were used to explore associations between quality scores and engagement metrics.ResultsTikTok videos had the highest descriptive engagement metrics, whereas Bilibili videos had the longest duration. Platform differences were observed for GQS, H(2) = 39.878, P = 2.19 × 10⁻⁹, Holm-adjusted P = 6.57 × 10⁻⁹, η2H = 0.128; mDISCERN, H(2) = 20.453, P = 3.62 × 10⁻5, Holm-adjusted P = 7.24× 10⁻5, η 2H = 0.062; and JAMA, H(2) = 9.651, P = 0.008, Holm-adjusted P = 0.008, η2H = 0.026. Bilibili had the highest mDISCERN and JAMA distributions, whereas RedNote had the lowest GQS distribution. Institutions had higher JAMA scores than all other uploader groups; their GQS advantage was limited to comparison with ordinary users, and mDISCERN did not differ significantly by uploader identity. In the pooled analysis, no statistically significant association was detected between engagement indicators and any quality score (all |ρ|< 0.10).ConclusionOverall, the quality, reliability, and transparency of OP information on major short-video platforms were suboptimal, varied markedly across platforms, and showed a striking disconnect between popularity and credibility.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12889-026-29150-x.Keywords: Osteoporosis, Short‑video, Health information quality, Cross‑platform analysis, Reliability Introduction Osteoporosis (OP) is a chronic, systemic bone disease marked by decreased bone mass and progressive degradation of bone microstructure, leading to diminished bone mineral density and a substantially increased susceptibility to low-trauma (fragility) fractures [1–3]. Globally, OP affects approximately 19.7% of the population, with incidence rising steadily amid demographic aging—disproportionately burdening women and older adults [1, 4]. In China, the disease burden is especially severe: contemporary epidemiological estimates applied to the 2020 national population indicate approximately 145.86 million cases [1]. Beyond impairing ambulation, activities of daily living, and health-related quality of life, OP imposes multifaceted strain on healthcare delivery systems, labor productivity, long-term care infrastructure, and public welfare frameworks [5, 6]. Given this broad systemic impact, effective OP prevention and control hinges critically on a dual-pronged, non-pharmacological strategy: integrated, scalable, and evidence-informed patient education delivered via accessible platforms, coupled with sustained, person-centered self-management support—a point now widely recognized [7, 8]. Over the past few years, short-form video platforms—particularly TikTok and Bilibili—have risen to prominence as essential conduits for delivering health-related content to broad audiences [9]. The high accessibility of short-video platforms lowers structural and cognitive barriers to health information access, enabling rapid and widespread dissemination—yet it simultaneously amplifies risks including information overload, declining content credibility, and the proliferation of health misinformation [10, 11]. A growing body of systematic reviews and empirical studies consistently documents that health information on short-video platforms is frequently characterized by weak evidentiary support and a high prevalence of factual inaccuracies [12–14]. For instance, evaluations of spinal cord injury–focused clips hosted on short-video sharing platforms reveal that, despite diversity in format, their overall quality remains suboptimal [15]. Exposure to such content can distort patients’ health-related decision-making and self-management practices, erode trust in clinical professionals, and—through these mechanisms—ultimately compromise clinical outcomes [12, 15, 16]. Given the vast audience for OP-related content on short-video platforms, the dissemination of erroneous or fragmented information poses a tangible risk to digital public health, health literacy, and misinformation exposure, highlighting the urgent need for systematic quality assessment. Previous studies have evaluated OP-related information on TikTok alone [17] and have subsequently compared TikTok with Bilibili [18]. These studies established that overall information quality was limited, but RedNote was not included and the appraisal dimensions were narrower. The present study extends this literature by concurrently comparing TikTok, RedNote, and Bilibili; incorporating three widely used standardized instruments—the GQS, mDISCERN, and JAMA benchmarks; using a four-category classification of uploader identity; and examining associations between engagement indicators and information-quality scores in a 300-video sample. This study aims to: (1) systematically evaluate cross-platform variations in the quality of OP-related information present on China’s major short-video platforms; (2) examine differences according to uploader identity and content category; and (3) explore associations between video characteristics, engagement indicators, and information-quality scores. Methods Ethics statement and data availability All data were extracted exclusively from publicly accessible short videos on TikTok, RedNote, and Bilibili, thereby preserving personal privacy. Account information was used solely to classify uploader identity; usernames and other direct identifiers were not reported, and all analyses were presented in aggregate. Moreover, the research protocol entailed no direct interaction with platform users, rendering ethical approval [19]. Data retrieval and screening To conduct a rigorous, standardized assessment of the accuracy, completeness, and dependability of health information pertaining to OP that is circulated across China’s short-form video platforms, candidate search results were captured on 12 March 2026 within a one-hour window in China, using the same desktop computer (manufacturer: HONOR; model: GLO-GXXX), Windows 11 version 25H2, and Microsoft Edge version 145.0.3800.58. The browser and platform language were set to Simplified Chinese. No browser extensions or virtual private networks were used. Screening, coding, and quality assessment were conducted subsequently. We established dedicated, neutral user accounts on TikTok (https://www.douyin.com/), Bilibili (https://www.bilibili.com/), and RedNote (https://www.xiaohongshu.com/), each with no prior browsing history, follower relationships, or user-generated interactions. These procedures were intended to provide a standardized baseline retrieval condition rather than to eliminate algorithmic bias or reproduce a typical individualized user feed. We executed identical search queries using the Chinese term “骨质疏松” (OP) on TikTok, RedNote, and Bilibili; the videos were recorded in the default order displayed by each platform’s search or recommendation interface. Sample size was determined based on observations in earlier cross-sectional investigations of health-related digital content; we aimed to collect 100 analyzable videos from each of the three platforms, leading to an overall target of 300 videos. Videos were eligible only if OP constituted their central health topic. The following were excluded: (1) promotional material or content explicitly intended for commercial gain; (2) videos with no clinically relevant information on OP; and (3) repeated posts from the same platform, where only the earliest-published version was kept. A single reviewer applied a pre-established, sequential screening procedure—reviewing results in the order in which they were displayed—until 100 qualifying videos per platform were selected, forming the basis of the final analysis. The systematic selection workflow of the study is depicted schematically in Fig. 1. Data Screening Flowchart for Quality & Reliability Analysis Short-video categorization For the complete sample of 300 videos, we systematically extracted six platform-derived metrics: number of likes, comments, shares, and favorites; the uploader’s fans; and video duration (in seconds). Uploaders were classified into four categories—orthopedic surgeons, other medical personnel, institutions, and ordinary users—through a comprehensive evaluation of verifiable profile information, including account ID, posted content, personal profile description, and platform-issued official verification badges (e.g., blue checkmarks). Institutions included hospitals, universities, professional societies, and government or public-health organizations. Accounts with a verifiable healthcare role other than orthopedic surgeons were categorized as other healthcare professionals. Personal accounts without verifiable healthcare or institutional status were categorized as ordinary users. Each video’s primary health communication focus was independently assigned to one of five predefined thematic domains: disease overview, diagnosis, symptoms, prevention, and treatment. Ambiguous uploader classifications and content categories were reviewed independently by two investigators and resolved by consensus. Instruments and evaluation procedure We systematically evaluated OP-related content disseminated via short-video platforms using three widely used content-appraisal instruments: the GQS, mDISCERN, and JAMA. This multimodal evaluation strategy facilitated a robust, evidence-informed appraisal of informational accuracy, scientific credibility, and overall quality of health communication (methodological details are outlined in the Supplementary Materials). Their cross-platform reliability has been empirically established, and they have been rigorously adapted and implemented within the Chinese short-video landscape to assess OP-related health videos [20–23]. The GQS assigns an overall score ranging from 1 (indicating very low educational value and limited applicability) to 5 (reflecting high educational quality and strong real-world relevance), capturing the video’s combined instructional effectiveness and functional usefulness [12, 24, 25]. The mDISCERN is a commonly used five-item modification of the original DISCERN instrument, explicitly engineered to critically appraise the scientific validity and communicative fidelity of treatment-oriented health information in short-video formats. This approach underscores objectivity and trustworthiness through the evaluation of five key dimensions—clarity of purpose, evidence objectivity, source reliability, supplementary resource provision, and acknowledgment of uncertainty—each assigned a binary score (1 if the criterion is satisfied, 0 otherwise), yielding a total score ranging from 0 to 5, with higher scores indicating greater scientific validity and communicative fidelity [24–26]. The JAMA benchmarks assess the transparency and credibility of health information using four criteria: credit for authorship, attribution practices, up‑to‑dateness, and conflict‑of‑interest disclosure, likewise scored on a binary scale (1 for satisfaction, 0 for non-satisfaction), producing a total score between 0 and 4, with higher scores reflecting stronger credibility [25, 27, 28]. All videos were independently assessed in a blinded manner by two orthopedic surgeons (Shi and Mao) with a background in medical statistics, each having over three years of experience, who were blinded to each other's ratings but not to platform identity. Prior to assessment, all raters completed a standardized, competency-based training program covering the theoretical foundations, item-level interpretation, and scoring conventions of the GQS, mDISCERN, and JAMA benchmarks. After independent scoring, discrepancies were adjudicated by a third expert (Chen), who has clinical qualifications and research experience in OP. Final analyses used consensus scores. Inter-rater agreement was calculated from the two independent pre-consensus rating sets using weighted Cohen’s kappa. Agreement was substantial to almost perfect: κ = 0.82 (95% confidence interval [CI], 0.76–0.88) for GQS; κ = 0.78 (95% CI, 0.71–0.85) for mDISCERN; and κ = 0.80 (95% CI, 0.74–0.86) for JAMA. Statistical analysis Version 11.0 of GraphPad Prism and version 31.0 of IBM SPSS Statistics were employed for all statistical computations. Normality was examined by means of the Shapiro–Wilk test. Given that all variables exhibited a significant departure from normality (all P < 0.001), continuous metadata variables and ordinal quality scores were summarized using medians and interquartile ranges (IQR) and quality assessment scores (GQS, mDISCERN, and JAMA). For each co-primary outcome, platform differences were assessed using the Kruskal–Wallis test. Significant omnibus tests were followed by Dunn’s pairwise comparisons with Bonferroni-adjusted P values. Holm correction was additionally applied across the three co-primary omnibus tests. The Kruskal–Wallis effect size was reported as η2H = (H—k + 1)/(n—k), where H is the Kruskal–Wallis statistic, k is the number of groups, and n is the total sample size. Given that the variables exhibited non-normal distributions, bivariate associations were assessed employing Spearman’s rank correlation coefficient (ρ). All statistical analyses were performed using R version 4.4.3 with the Hmisc (version 5.2–3) and corrplot (version 0.95) packages for correlation analyses. A two-tailed α threshold of 0.05 was pre-established for determining statistical significance in all hypothesis tests. Results Summary statistics of short‑video metadata and quality ratings We profiled 300 OP-related short videos (100 per platform) to compare engagement metrics and information quality. Detailed results are presented in Table 1. TikTok generated the highest median likes (818.50), comments (50.00), shares (268.00), and favorites (386.50), whereas Bilibili featured substantially longer videos (median 352.00 s) compared with median 146.50 s on TikTok and median 122.00 s on RedNote. RedNote occupied an intermediate position in most metrics. However, higher engagement did not translate into reliable health information. Using three widely used instruments, we found overall video quality to be suboptimal. Median GQS scores were 3.00 (IQR, 3.00–4.00) for TikTok, 3.00 (IQR, 2.00–3.00) for RedNote, and 3.00 (IQR, 3.00–4.00) for Bilibili. Median mDISCERN scores were 1.00 (IQR, 1.00–2.00), 1.00 (IQR, 1.00–2.00), and 2.00 (IQR, 1.00–3.00), respectively. Median JAMA scores were 1.00 (IQR, 1.00–1.00) for TikTok and RedNote, and 1.00 (IQR, 1.00–2.00) for Bilibili. These values indicate that most videos lack credible references, balanced discussion, and essential disclosure elements. In qualitative terms, the overall quality of OP content on these platforms ranges from poor to moderate, with Bilibili showing marginal superiority but still falling short of acceptable standards. Table 1Descriptive statistics of videosVariablesTikTok (n = 100)RedNote (n = 100)Bilibili (n = 100)Total (n = 300)Basic Information Likes, median (IQR)818.50 (96.25–6732.25)90.00 (27.50–388.25)23.50 (4.00–128.25)93.50 (24.00–899.25) Comments, median (IQR)50.00 (9.00–284.00)4.00 (1.00–31.25)2.00 (0.00–13.50)7.00 (1.00–52.50) Shares, median (IQR)268.00 (12.75–1846.00)54.50 (9.00–197.00)25.50 (4.25–167.50)54.50 (7.25–352.25) Favorites, median (IQR)386.50 (21.00–3481.75)72.00 (19.00–225.25)44.50 (4.25–200.00)84.00 (12.00–515.25) Fans, median (IQR)139,000.00 (28,250.00–647500.00)27,000.00 (2431.75–62,000.00)5918.00 (83.25–27,000.00)27,000.00 (2918.25–159,000.00) Durations, median (IQR)146.50 (91.25–248.00)122.00 (76.50–171.75)352.00 (187.00–889.25)161.50 (95.50–316.00)Quality Scores GQS, median (IQR)3.00 (3.00–4.00)3.00 (2.00–3.00)3.00 (3.00–4.00)3.00 (3.00–4.00) GQS, mean ± SD3.29 ± 0.782.76 ± 0.883.56 ± 0.813.20 ± 0.89 mDISCERN, median (IQR)1.00 (1.00–2.00)1.00 (1.00–2.00)2.00 (1.00–3.00)1.50 (1.00–2.00) mDISCERN, mean ± SD1.63 ± 0.771.53 ± 0.902.07 ± 1.001.74 ± 0.92 JAMA, median (IQR)1.00 (1.00–1.00)1.00 (1.00–1.00)1.00 (1.00–2.00)1.00 (1.00–1.00) JAMA, mean ± SD0.98 ± 0.471.05 ± 0.641.28 ± 0.841.10 ± 0.68 Descriptive statistics of videos Spatial patterns in uploader identities and types of content A systematic audit of 300 OP videos hosted on China's three leading short-video platforms reveals significant platform-specific stratification in both content production and thematic focus. While healthcare professionals dominated the overall sample (75.33%), their representation varied substantially across platforms (Fig .2A). A-D (A. Aggregate distribution of publisher identities across short-video platforms; B. Platform-resolved publisher identity composition: TikTok, Bilibili, and RedNote; C. Overall proportional breakdown of content categories released on short-video platforms; D. Content category composition by platform: TikTok, Bilibili, and RedNote) As shown in Fig. 2B, the creator composition on Bilibili was more heterogeneous, with medical practitioners accounting for only 42.00% of uploaders (orthopedic surgeons 24% + other medical personnels 18%). Conversely, the short videos on RedNote were driven predominantly by orthopedic surgeons (65.00%). On TikTok, the content generation was exclusively confined to medical professionals (orthopedic surgeons: 85.00%; other medical personnels: 15.00%). Content analysis (Fig. 2C) further delineates these divergences. Across the corpus, disease overviews (32.33%) and prevention (32.00%) constituted the primary informational foci, overshadowing treatment (22.67%) and diagnosis-related content (4.67%). Platform-level comparison (Fig. 2D) reveals distinct content priorities: Bilibili predominantly features comprehensive educational content; RedNote emphasizes prophylactic measures (43.00%); whereas TikTok demonstrates a curative bias, with surgical outcomes and interventions comprising 30.00% of its content and symptomatic guidance accounting for only 1.00%. Video quality assessment To quantify between-platform differences in video quality across TikTok, RedNote, and Bilibili, we performed the Kruskal–Wallis non-parametric test followed by Dunn’s post hoc test (α = 0.05). As shown in Fig. 3, GQS and mDISCERN scores differed significantly among the three platforms (all P < 0.001). JAMA scores also showed statistically significant inter-platform variation, albeit with a smaller effect size compared with GQS and mDISCERN. Comparison of GQS, mDISCERN, and JAMA scores across platforms. Overall differences among platforms were assessed using the Kruskal–Wallis test; where significant, pairwise post-hoc comparisons were performed with Dunn’s test corrected for multiple comparisons (*adjusted P < 0.05, ** adjusted P < 0.01, *** adjusted P < 0.001, **** adjusted P < 0.0001) For GQS, although all three platforms shared a median score of 3.00, the underlying distributions were divergent. A Kruskal–Wallis test revealed a statistically significant difference in GQS scores across the different platforms. For GQS, H(2) = 39.878, P = 2.19 × 10⁻⁹, Holm-adjusted P = 6.57 × 10⁻⁹, and η2H = 0.128. Post hoc comparisons revealed that RedNote received significantly lower GQS ratings than both TikTok and Bilibili (all adjusted P < 0.0001). In combination with a narrow IQR of 2.00–3.00, this points to a notable deficiency in the overall production and presentation quality of RedNote videos. With respect to mDISCERN, Bilibili (median = 2.00) scored significantly higher than TikTok (median = 1.00) and RedNote (median = 1.00). A Kruskal–Wallis test revealed a statistically significant difference in mDISCERN scores across the different platforms. For mDISCERN, H(2) = 20.453, P = 3.62 × 10⁻5, Holm-adjusted P = 7.24 × 10⁻5, and η2H = 0.062. Post hoc analysis confirmed that Bilibili differed significantly from the other two platforms (adjusted P < 0.0001 compared with RedNote; adjusted P = 0.005 compared with TikTok), indicating relatively superior information quality and reliability. However, no significant difference was found between RedNote and TikTok (adjusted P = 0.637). Nevertheless, mDISCERN scores remained low across all three platforms, reflecting an overall paucity of reliable video information. For JAMA scores, median values were uniformly low at 1.00, yet the distribution patterns differed markedly. Bilibili exhibited a more dispersed distribution (IQR = 1.00), which encompassed a broader range of higher scores, whereas both TikTok and RedNote were highly concentrated at the lower end of the scale (IQR = 0). A Kruskal–Wallis test revealed that there was significant overall heterogeneity in the distribution of the measured indicators across the three platforms. For JAMA, H(2) = 9.651, P = 0.008, Holm-adjusted P = 0.008, and η2H = 0.026. Post hoc testing further substantiated that Bilibili had significantly higher JAMA scores than TikTok (adjusted P = 0.008), whereas the difference between Bilibili and RedNote was not statistically significant (adjusted P = 0.077), underscoring a marginal but measurable advantage in professional rigor. We also conducted a stratified comparative analysis of the quality scores across the 300 short videos according to uploader provenance. As presented in Fig. 4, Kruskal‑Wallis tests revealed significant differences among uploader groups for both GQS and JAMA scores, with divergent effect sizes, whereas mDISCERN scores did not differ significantly across groups. Comparisons of quality scores across publisher identities. a, GQS scores. b, mDISCERN scores. c, JAMA scores. Differences were assessed using the Kruskal–Wallis test followed by Dunn’s post‑hoc test; all P values were adjusted for multiple comparisons (*adjusted P < 0.05, ** adjusted P < 0.01, *** adjusted P < 0.001, **** adjusted P < 0.0001) For GQS, H(3) = 12.25, η2H = 0.031, P = 0.006. Post‑hoc pairwise comparisons showed that institutions scored significantly higher than ordinary users (adjusted P = 0.004), but no significant differences were detected between institutions and either orthopedic surgeons or other medical personnel. For mDISCERN, H(3) = 2.77, η2H = −0.001, P = 0.429, and no uploader-group difference was detected. For JAMA, H(3) = 33.25, η2H = 0.102, P < 0.0001. Post‑hoc pairwise comparisons substantiated that institutions significantly outperformed all other categories, including orthopedic surgeons, other medical personnel, and ordinary users (adjusted P = 0.0027 compared with other medical personnel; adjusted P < 0.0001 compared with orthopedic surgeons; adjusted P < 0.0001 compared with ordinary users). Taken together, these findings reveal a selective institutional advantage—robust and comprehensive only in structured presentation metrics (JAMA), partial in global quality scoring (GQS, limited to comparisons with ordinary users), and absent in fundamental information reliability (mDISCERN). Correlation analysis For the purpose of analyzing how OP-related short video quality correlates with dissemination characteristics, we performed Spearman’s rank correlation analyses using R (correlation matrix shown in Fig. 5). Three main findings emerged. Correlation matrix of video popularity and engagement metrics with professional quality assessment instruments (GQS, mDISCERN, and JAMA). Each cell displays the correlation coefficient (range: 0–1) for the corresponding variable pair; darker shading indicates a stronger correlation First, dissemination characteristics—likes, favorites, comments, and shares—were very strongly and positively intercorrelated (all ρ > 0.95), reflecting highly consistent user interaction patterns. Second, the three quality instruments—GQS, mDISCERN, and JAMA—showed significant positive intercorrelations of varying strength (ρ ranging from 0.22 to 0.58). The strongest association was between mDISCERN and GQS (ρ = 0.58, P < 0.001), suggesting that videos with higher content objectivity and evidence reliability tended to receive higher evaluations of educational value and practical utility. In contrast, the GQS–JAMA correlation was weaker (ρ = 0.22, P < 0.001), indicating that formal adherence to disclosure standards does not necessarily translate into high educational quality. Most critically, we found no significant associations between any of the three quality scores and dissemination characteristics (likes, favorites, comments, shares; all |ρ|< 0.10, all P > 0.05), demonstrating a clear decoupling of video information quality from user engagement that may reflect a misalignment between professional evaluation systems and the culture of short-video platforms. Discussion Key findings This study found that OP-related short videos originating from different platforms and content creators exhibit marked disparities in information quality. Also, uploader identity was associated selectively with quality: institutions showed a consistent advantage in JAMA transparency, a limited advantage in GQS relative to ordinary users, and no advantage in mDISCERN. No statistically significant association was detected between engagement metrics and quality scores in the pooled analysis. These findings preliminarily illuminate aspects of the social‑media health‑information ecosystem while also exposing a practical dilemma in public health practice, and may thus offer a modest reference for future work exploring potential remedies. Characteristics of cross-platform short-form video health communication Drawing on a sample of 300 OP-related videos sourced from TikTok, RedNote, and Bilibili, this study delineates the divergent landscapes of digital health communication across three major Chinese short-video platforms. We identified a pronounced trade-off between user engagement and perceived content quality across platforms. Dissemination characteristics—including likes, comments, shares, favorites, and fans—were significantly elevated on TikTok, reflecting superior reach and user interaction. While RedNote lagged behind TikTok in both dissemination characteristics, mDISCERN and GQS, it demonstrated marginally higher information reliability as assessed by JAMA. Conversely, although Bilibili exhibited the lowest engagement levels, it maintained relatively robust GQS, mDISCERN and JAMA performance. A striking disparity in uploader demographics was evident: orthopedic surgeons constituted over 60% of content creators on TikTok and RedNote, whereas Bilibili featured a markedly more heterogeneous creator base, with healthcare professionals accounting for only 24% of uploaders. These patterns align with prior characterizations reported in the literature: TikTok videos achieve high virality and user engagement but often lack substantive depth; Bilibili videos reflect a broader diversity of occupational backgrounds and exhibit greater content rigor, albeit with comparatively limited reach; and RedNote is characterized by a high density of promotional and advertising content [29–31]. Collectively, these findings underscore a critical fragmentation in social media–based health education: platforms that maximize audience reach frequently diverge from those that prioritize clinical rigor. Video quality assessment Prior studies have examined the quality of online health information about OP, yet they have largely focused on single platforms and rarely employed multiple standardized instruments for cross-platform comparison [17, 32]. This narrow evaluative scope risks overlooking critical dimensions, such as the balance of treatment recommendations and the transparency of information source [30]. To address this gap, we systematically assessed OP-related short videos on three major platforms—TikTok, RedNote, and Bilibili—using GQS, mDISCERN, and JAMA. Overall video quality was moderate to low across all platforms, but the pattern of strengths differed by platform. Bilibili and TikTok led in GQS scores, indicating superior overall quality relative to RedNote. TikTok scored relatively low on the JAMA, pointing to clear deficiencies in transparency indicators such as authorship and references. For mDISCERN, Bilibili outperformed the other two platforms with the highest scores, TikTok ranked second, and RedNote placed at the bottom, reflecting a marked failure to communicate uncertainties and risks when disseminating health information. Analysis by uploader identity yielded a mixed picture. For JAMA scores, institutions obtained higher scores than all other groups, with these differences reaching statistical significance. For GQS, institutions scored significantly higher than ordinary users only; no significant differences were observed among institutions, orthopedic surgeons, and other medical personnel. The mDISCERN scores did not differ significantly across any uploader categories. These findings suggest that institutional content holds an advantage in transparency and accountability (JAMA) over individual creators, while its overall quality advantage (GQS) appears limited to comparisons with ordinary users. The lack of mDISCERN differences raises the possibility that OP-related information reliability was conveyed at a similar level across uploader types. The JAMA advantage may stem from team-based production involving content strategists, medical editors, and clinical consultants, which facilitates systematic disclosure of authorship, credentials, and references. Non-profit or academic institutions may also enforce internal compliance checks that discourage unsubstantiated claims; hospitals face stronger reputational constraints, likely encouraging more proactive disclosure. Regarding GQS, the significant difference between institutions and ordinary users may reflect the benefit of multi-round professional review for comprehensive coverage of etiology, diagnosis, treatment, and daily management, whereas individual medical professionals can achieve comparable depth. The absence of mDISCERN differences could indicate that core OP-related messages remain relatively consistent across sources, though alternative explanations are possible. In summary, the three platforms exhibit distinct shortcomings in presenting OP health information: Bilibili fares relatively well on GQS and mDISCERN but remains fragmented; TikTok delivers acceptable GQS yet falls noticeably short on JAMA transparency; and RedNote performs worst on mDISCERN. Institutions showed a consistent advantage only in JAMA transparency scores, while their GQS advantage was limited and no uploader-group difference was detected for mDISCERN. These findings indicate that the institutional advantage was most apparent in information transparency rather than consistently across all quality dimensions. Correlation analysis between basic characteristics and quality scores The relationship between fundamental attributes of short videos and their content quality scores was analyzed, revealing no positive correlation between information quality and popularity-related metrics. This result is consistent with recent health communication findings on knee osteoarthritis [33], osteosarcoma [34], and oral cancer [35]. These data demonstrate that dissemination characteristics—such as likes, shares, and comments—are unreliable proxies for the accuracy and professionalism of OP-related educational material. The proposed exacerbation of the communication paradox in bone health self-management by short‑video algorithmic curation—where sensational pseudoscience tends to overshadow evidence‑based guidance—remains a hypothesis awaiting empirical verification. One possible explanation for the weak association between engagement and quality scores is that these measures capture different dimensions of value. GQS, mDISCERN, and JAMA emphasize organization, evidentiary support, source attribution, and disclosure, whereas users may also respond to personal experience, narrative accessibility, emotional resonance, and interaction within comment communities. Strengths and limitations This study extends previous single- and two-platform analyses by simultaneously evaluating 300 OP-related videos from TikTok, RedNote, and Bilibili using GQS, mDISCERN, and JAMA benchmarks. Previous studies of OP short videos on Chinese platforms are more limited to single platforms, narrow instrument sets and older data [17, 18]. We fill these gaps by concurrently analyzing TikTok, RedNote and Bilibili with a larger, more recent sample, using GQS, mDISCERN and JAMA to assess multiple quality dimensions. Importantly, we uncover novel disparities through a refined uploader classification and show that engagement metrics do not correlate with content quality. Methodologically, the study employed validated, standardized evaluation instruments, revealing significant inter-platform disparities in information quality—findings that might help inform, to a modest extent, the design of more precisely targeted health communication interventions. Several limitations, however, should be considered. First, owing to its cross-sectional nature, the study design merely records a time-bound snapshot of informational attributes and cannot reveal temporal dynamics, dissemination trends, or potential causal relationships. Second, we must point out that these neutral accounts cannot avoid algorithmic differences caused by request-level features (such as time, device, and inferred locale) and platform-level non-personalized signals (such as popularity or freshness); moreover, constrained by platform-specific recommendation algorithms and retrieval strategies, the sample may not encompass the full spectrum of relevant online content, introducing possible selection bias. Third, although ambiguous cases about uploaders’ categorization were independently reviewed by two coders and resolved by consensus, we have acknowledged that reliance on platform-displayed information may still result in misclassification. Fourth, the interpretation and grading of content inevitably retain a degree of subjectivity, which may affect assessment precision. Nevertheless, inter-rater reliability was formally tested: for GQS, weighted kappa was 0.82 (95% CI, 0.76–0.88); for mDISCERN, weighted kappa was 0.78 (95% CI, 0.71–0.85); and for JAMA, weighted kappa was 0.80 (95% CI, 0.74–0.86). A separate limitation is the absence of intra‑rater reliability assessment, which precludes quantification of within‑rater consistency over time. Fifth, the study remained at the level of content quality evaluation and did not assess the real-world effects of this content on patients’ health literacy, perceptions, behavioral intentions, or clinical outcomes. Conclusion OP-related short videos on TikTok, RedNote, and Bilibili showed limited and platform-specific quality, reliability, and transparency. Bilibili had relatively higher mDISCERN and JAMA distributions, whereas RedNote had the lowest GQS distribution. Institutional uploaders showed their clearest advantage in JAMA transparency; they were not uniformly superior across all quality measures. In the pooled analysis, no statistically significant association was observed between engagement indicators and information-quality scores. Future longitudinal and user-centered studies are needed to determine how these content differences affect health literacy, self-management, and clinical decision-making. Supplementary Information Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Acknowledgements The authors thank Dr. Liangliang Chen for adjudicating the independent scoring discrepancies. Authors’ contributions SY and WM independently assessed all videos. SY collected and coded the video data and performed the primary statistical analyses. WM conceived and supervised the study, conducted the correlation analysis, and critically revised the manuscript. All authors read and approved the final manuscript. Funding Not applicable. Data availability The original contributions presented in the study are included in the article/Supplementary Material. These data can be provided by the corresponding author upon justified and scholarly request, who can also be contacted for further inquiries. Declarations Ethics approval and consent to participate Not applicable. This study was based solely on publicly accessible short-video content retrieved from open-access platforms. Publicly available profile information was accessed solely for uploader classification. Direct identifiers were not reported in the manuscript or shared dataset, and all analyses were presented in aggregate. As such, formal ethical review and approval were not required. Consent for publication Consent for publication was not required in this study. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Data Availability Statement The original contributions presented in the study are included in the article/Supplementary Material. These data can be provided by the corresponding author upon justified and scholarly request, who can also be contacted for further inquiries.