Short video platforms as sources of health information about HPV vaccine: A content and quality analysis

Abstract

Background

Numerous HPV vaccine-related videos are available on Kwai, Bilibili, and TikTok; however, their quality and professionalism vary considerably. This study evaluates the quality and reliability of HPV vaccine-related videos on these platforms to offer Chinese-speaking users a reference for informed vaccination decisions.

Method

A keyword search for the top 100 relevant videos on “HPV vaccine” was performed on Kwai, Bilibili, and TikTok, yielding a total of 238 eligible videos. A comparative analysis was conducted on video characteristics, uploader profiles, content categories, uploader attitudes, and public responses. The Global Quality Score (GQS) and the modified DISCERN (mDISCERN) instrument were applied to assess video quality and reliability.

Result

TikTok had the highest median number of likes (1744.5) and shares (1338), while Bilibili led in median comments (179.25) and favorites (527). Public support for the HPV vaccine was highest on Kwai (63.2%), whereas TikTok showed notable opposition (15.8%). Interestingly, only Bilibili lacked a neutral stance. The proportion of physician uploaders was highest on TikTok (61.8%), whereas Bilibili had the largest share of self-media contributors (66.3%). Among professional uploaders, 92.3% supported the HPV vaccine, and their videos received 55.4% public approval—significantly higher than the 34.6% for individual users (p = 0.021). Significant differences in mDISCERN scores were observed across all three platforms (all pairwise comparisons, p < 0.001).

Conclusion

Videos uploaded by professionals tend to have higher engagement and greater informational reliability, making them more effective in promoting public support for vaccination. TikTok videos scored highest on both GQS and mDISCERN metrics and had the largest proportion of professional uploaders, indicating superior overall quality and reliability. These findings should be interpreted within the context of the mainland-Chinese short-video ecosystem and may not be generalizable to non-Chinese-speaking populations.

Introduction

The HPV vaccine has been demonstrated to be effective in preventing high-risk HPV types responsible for cervical cancer.^1,2 The World Health Organization (WHO) recommends HPV vaccination and cervical cancer screening as key primary prevention strategies. The vaccine plays a pivotal role in global efforts to eliminate cervical cancer. Despite its well-established benefits, global HPV vaccination coverage remains suboptimal—ranging from 11.3% in low- and middle-income countries to 51.5% in high-income countries. As of 2021, global coverage among 15-year-old females was merely 12.2%. ³ Contributing factors include economic barriers—such as limited vaccine access in lower-income regions—and insufficient public understanding, which fosters safety concerns. ⁴ Therefore, raising public awareness is essential to improving access to accurate and timely HPV vaccine information.

With the rise of short-video platforms, these media have emerged as key intermediaries between health educators and information seekers. Such platforms hold significant potential to enhance HPV vaccine coverage. Although YouTube is not accessible in China, alternative platforms such as TikTok, Kwai, and Bilibili are widely used and highly influential. These platforms host large volumes of HPV vaccine-related content. However, the reliability and professionalism of such content vary considerably and warrant systematic evaluation. Some videos may disseminate inaccurate, subjective, or unverified information. ⁵ TikTok, in particular, not only has hundreds of millions of users in China ⁶ but also boasts a substantial international following. Consequently, these platforms exert considerable influence on public perception and health behaviors. Content on these platforms is widely disseminated both domestically and internationally. Exposure to inaccurate information may result in misguided decisions, delayed treatment, or adverse health outcomes.^7,8

Empirical evaluations of HPV vaccine-related video content remain limited in the current literature. Validated tools such as the Global Quality Score (GQS) and modified DISCERN (mDISCERN) have been developed to assess short-video quality. Therefore, this study aims to assess the quality and reliability of HPV vaccine-related videos on TikTok, Kwai, and Bilibili using GQS and mDISCERN metrics. The evaluation incorporates video themes, uploader characteristics and attitudes, public sentiment, and content features. The findings aim to inform public access to accurate and trustworthy health information.

The primary objective is to assess the quality, reliability, and dissemination features of HPV vaccine-related short videos on TikTok, Kwai, and Bilibili, and to examine how uploader professionalism affects content credibility and public sentiment.

The study specifically aims to: (a) quantify and compare video quality using GQS and mDISCERN across platforms; (b) characterize uploader profiles (professionals, non-professionals, individual users) and relate them to video themes and quality; (c) assess uploader and public attitudes toward HPV vaccination; and (d) offer evidence-based recommendations for platform governance and public health communication in China and comparable digital environments.

Significance: This is the first study to concurrently evaluate short-form HPV vaccine videos across China's three major platforms—TikTok, Kwai, and Bilibili—using validated instruments (GQS and mDISCERN). By correlating uploader credentials with content quality and public sentiment, the study yields actionable insights for: (a) platform governance—supporting algorithm optimization and content verification mechanisms; (b) public health practice—designing targeted strategies to combat misinformation and increase vaccine uptake; and (c) Mandarin-language health communication—providing a scalable framework for evaluating short-video health content in Chinese-speaking regions where such platforms dominate information flow.

Materials and methods

Ethical considerations

The study was approved by the Hebei General Hospital (2025-LW-0177) and performed in accordance with the ethical guidelines of the Declaration of Helsinki. The data used in this study were sourced from publicly available video content on platforms such as Kwai, Bilibili, and TikTok. These data are publicly accessible and do not involve any personal or private information. The use of all data strictly adheres to the respective platforms’ terms of service and complies with ethical standards for academic research. All keywords and content were in Simplified Chinese, reflecting mainland-Chinese linguistic norms.

Video collection

In this study, short-video data were collected from the platforms Kwai, Bilibili, and TikTok. The data collection process involved searching using a set of keywords that combined both scientific terminology and colloquial expressions to ensure a comprehensive coverage of content. Videos were collected according to the default sorting method provided by each platform, ensuring consistency in the data selection process. For each platform we collected the first 100 videos returned by the default “comprehensive” ranking displayed after entering the keyword “HPV vaccine” in the search bar. This default ranking is a proprietary composite score that weighs multiple engagement metrics—primarily views, likes, shares, and recency—but the exact weighting is not disclosed by the platforms. To ensure reproducibility, searches were performed in incognito mode (logged-out), with cleared cache/cookies and location services disabled; To ensure the stability and relevance of the data, videos that were classified as advertisements or were published recently (videos uploaded ≤ 7 days before the search date) were excluded from the collection. Screenshots of the search-results page and time-stamped browser logs were archived. Because the ranking can shift slightly over time, we captured the complete result lists within a 48-hour window on March 15 to 16, 2025. This method was designed to minimize bias and maximize the representativeness of the sample, ensuring the reliability and robustness of the results (Figure 1).

Figure 1.

Search strategy for videos on HPV vaccine.

Handling similar and irrelevant content

After retrieving the top 100 videos from each platform (total 300), we applied identical exclusion criteria in the following order: (a) similar or near-similar content (same footage/audio or identical script within ±15% similarity); (b) irrelevant themes (e.g. unrelated cancers, non-vaccine HPV treatments, or videos whose primary focus was not HPV vaccination).

Handling of missing data

In addition to the exclusion of similar or irrelevant content, missing or unavailable video metrics (such as likes, comments, shares, and follower counts) were handled by applying listwise deletion. This means that any video with missing data for any of the key metrics was excluded from the analysis of that specific metric. For videos where any key metric was missing, the corresponding data points were removed to ensure consistency across the dataset. This method allowed for a reliable statistical analysis without introducing biases from incomplete data.

Search strategy and data extraction

Various characteristics of the videos were carefully recorded, including the number of likes, comments, shares, and favorites, along with the content creator's follower count.

Uploader characteristic

The video uploaders were grouped into the following categories: Doctors, Other healthcare professionals/students, Official media, Self-media, and Doctors of traditional Chinese medicine (TCM). All doctors in various departments are classified as doctors. Other healthcare professionals include certified nurses, midwives, pharmacists, public health practitioners, and medical or nursing students who have not yet obtained full medical licensure. These individuals may not hold doctoral-level qualifications but are trained in health sciences and play a key role in health education and communication, particularly in community settings. Their contributions to HPV-related content often focus on practical guidance, such as clinic procedures, vaccine scheduling, and patient counseling, which complements the clinical expertise provided by physicians. Professionals: practicing physicians who have obtained the practicing physician qualification certificate and display verifiable information such as hospital work badges and personal physician information on their personal homepages (only including obstetricians and gynecologists); Non-professionals: Doctors from other departments, other healthcare professionals/students, official media (accounts officially operated by the Centers for Disease Control and Prevention, government health departments, or tertiary hospitals), and TCM; Individual users: individual users and self-media accounts without medical or official institution certification.

Video review and categorization

Two authors, Luyang Su, and Jingrun Yao, independently reviewed the videos and removed those that were either redundant or irrelevant. The videos were categorized based on their themes, which included mentions vaccines as safe, vaccine side effects, the age range for the vaccine, doses required, recommendations talking to the doctor, vaccination-related fertility concerns, length of vaccination interval, whether to get tested for HPV before vaccination, and whether regular screening is needed after vaccination. Since some videos covered more than one theme, the total number of themes addressed in each video was noted. Videos that did not align with any of the relevant themes were deemed irrelevant and excluded from the analysis.

Attitudes toward the HPV vaccine

The attitudes of uploaders and the public toward the HPV vaccine were categorized into four types: Pro-vaccination, Anti-vaccination, Neutral, and Unspecified.

Videos assessments

The evaluation of the videos was conducted using two widely recognized tools: the mDISCERN^9–11 and the GQS.^10–12 These tools were selected for their established effectiveness in assessing the quality and reliability of health-related content. The mDISCERN tool was employed to assess the reliability and quality of the health information presented in the videos. This tool evaluates the clarity and comprehensiveness of the information, as well as its relevance to the intended audience. The GQS was used to provide an overall rating of the video's quality, focusing on factors such as content accuracy, production quality, and engagement. Both tools are commonly used in health communication research to ensure objective and reliable assessments of online health information. The mDISCERN scoring system consists of 16 items, each addressing different aspects of the video content, such as clarity of the health message, balanced presentation of benefits and risks, and the inclusion of relevant references. Each item was rated on a 5-point scale, ranging from 1 (poor quality) to 5 (excellent quality) (Table S2). The GQS also employed a 5-point scale, with scores ranging from 1 (poor quality) to 5 (excellent quality), based on the overall assessment of video content and presentation (Table S1).

Two independent reviewers/rators (Luyang Su and Jingrun Yao), both with expertise in public health and medical research, were trained in the use of the mDISCERN and GQS tools and had experience in assessing health-related content online. Their expertise ensured the reliable and accurate application of the scoring criteria. A third arbitrator (Xiaohang Ai) assigned the final score if the two raters’ scores were inconsistent. For instance, if two people give the same score, there is no need for a third person to intervene. However, if their scores are inconsistent, a third person is required to give the score, and the score given by the third person shall prevail. To ensure the reliability of the assessments, inter-rater agreement was measured using Cohen’s Kappa statistic. This method was used to evaluate the level of agreement between the two raters, with values above 0.75 indicating excellent agreement. A Kappa value of 0.40 to 0.75 was considered acceptable, while values below 0.40 indicated poor agreement. Any discrepancies in ratings were discussed and resolved through consensus. The divergence is approximately twenty percent. The results of the video evaluations were analyzed to determine the overall quality of the videos and the extent to which they provided accurate, balanced, and comprehensive health information. Videos receiving higher scores on both the mDISCERN and GQS were considered of better quality, while lower scores indicated a need for improvement in content clarity, accuracy, and engagement. This evaluation process provided a robust assessment of the videos’ effectiveness as health information sources. Detailed descriptions of these tools are provided in Additional Files 1 and 2.

Statistical analysis

The data were analyzed using non-parametric distribution methods. The median and interquartile range (IQR) were used to summarize the data. For comparisons between two groups, the Mann–Whitney U test was employed, while the Kruskal–Wallis H test was used for comparisons involving three or more groups. Conducted non-parametric tests on the classifications of different platforms, then carried out post hoc tests, and added effect sizes. Cohen's Kappa coefficient was calculated to assess inter-rater reliability. To evaluate the relationships between video variables and their correlations with other variables, Spearman's rank correlation analysis was conducted. A p-value of <0.05 was considered statistically significant. Data analysis was performed using IBM SPSS Statistics 25, and data visualization was conducted using GraphPad Prism 9.

Results

Video characteristics

Our study included 76 Kwai videos, 86 from Bilibili, and 76 from TikTok, after excluding similar and irrelevant content (Figure 1). The analysis focused on several key metrics, including video duration, likes, collections, comments, shares, and follower counts. The video characteristics across three platforms—Bilibili, Kwai, and TikTok—exhibited notable differences (Table 1). Bilibili videos had the longest duration (median 261 seconds), significantly longer than those on Kwai (median 78.5 seconds) and TikTok (median 70 seconds). TikTok videos garnered the highest median likes (1744.5) and shares (1338), followed by Kwai (likes: 506.5, shares: 158) and Bilibili (likes: 103, shares: 77). Bilibili videos had the highest median comments (179.25) and collections (527), while TikTok had the lowest (comments: 48.5, collections: 43.25). Regarding uploader attitudes, TikTok had the highest proportion of pro-vaccination content (86.8%), whereas Kwai was exclusively pro-vaccination (100%). Bilibili had the highest proportion of anti-vaccination and unspecified content. These findings highlight platform-specific differences in user engagement and content orientation regarding HPV vaccine videos.

Table 1.

Characteristics of videos about HPV vaccine on Bilibili/Kwai/TikTok.

Variables	Total (n = 238)	Bilibili (n = 86)	Kwai (n = 76)	TikTok (n = 76)	p	Cohen's f
Likes, M (Q1, Q3)	515.5 (82–5620)	103 (13–1184)	506.5 (112.5–4978)	1744.5 (219.75–16,327.25)	0.025	0.16
Collections, M (Q1, Q3)	146 (17.25–1096.75)	66.5 (8.25–527)	111 (17–930.25)	379.5 (43.25–3939.5)	0.001	0.25
Comments, M (Q1, Q3)	73.5 (10–903)	17 (1.75–179.25)	54 (12–1099.5)	223 (48.5–1929.75)	0.010	0.21
Shares, M (Q1, Q3)	209 (25–3013)	77 (7–530)	158 (40–2285)	1338 (174.5– 12,553.5)	<0.001	0.34
Video duration(s), M (Q1, Q3)	100 (60–212)	261 (169–524.25)	78.5 (46.5–104.25)	70 (55–106.5)	<0.001	0.88
Followers, M (Q1, Q3)	13.7 (0.958–138.825)	1.45 (0.026–18.675)	18.4 (8.75–135.3)	65.65 (4.75–334.4)	0.025	0.16
Uploaders attitude (%)					<0.001
Anti-vaccination (%)	7 (2.9%)	3 (3.5%)	0 (0.0%)	4 (5.3%)
Neutral (%)	11 (4.6%)	8 (9.3%)	0 (0.0%)	3 (4.0%)
Pro-vaccination (%)	205 (86.1%)	63 (73.3%)	76 (100.0%)	66 (86.8%)
Unspecified (%)	15 (6.3%)	12 (14.0%)	0 (0.0%)	3 (4.0%)

Uploader characteristics

This study analyzed the distribution of video uploaders across three platforms: Kwai, Bilibili, and TikTok. Doctors were more prevalent on Kwai (51.3%) and TikTok (61.8%) compared to Bilibili (7.0%). The majority of videos on Bilibili (66.3%) and Kwai (18.4%) were uploaded by self-media, significantly higher than on TikTok (10.5%). The proportion of videos from official media was highest on TikTok (17.1%) and lowest on Bilibili (11.6%). Other medical workers or students were relatively evenly distributed across the platforms. Notably, the presence of TCM doctors was minimal across all platforms (Table 2 and Figure 2).

Table 2.

Characteristics of video uploaders about HPV vaccine on Bilibili/Kwai/TikTok.

Type of uploaders, n(%)	Bilibili (n = 86)	Kwai (n = 76)	TikTok (n = 76)	p	Hos	Cramer's V
Doctor	6 (7.0%)	39 (51.3%)	47 (61.8%)	<0.001	T > K > B	0.4
Other medical worker/student	13 (15.11%)	20 (26.3%)	8 (10.5%)	0.04	B > T;K > T	0.55
Official media	10 (11.6%)	3 (4.0%)	13 (17.1%)	0.03	B > K;T > K	0.32
Self-media	57 (66.3%)	14 (18.4%)	8 (10.5%)	<0.001	B > K > T	0.3

Note: T stands for Tiktok, B for Bilibili, and K for Kwai. Hos stands for Post Hoc Inspection. Hos stands for post-hoc test.

Figure 2.

Numbers of video uploaders about HPV vaccine on Bilibili/Kwai/TikTok (all the authors).

This study assessed the quality and reliability of HPV vaccine-related videos on Bilibili, Kwai, and TikTok using the GQS and mDISCERN scores. TikTok videos had the highest median GQS (4, IQR 2–4) and mDISCERN scores (4, IQR 2–4), followed by Kwai (GQS: 3, IQR 1–3; mDISCERN: 3, IQR 2–4) and Bilibili (GQS: 3, IQR 2–4; mDISCERN: 2, IQR 1–4). The consistency coefficients (Cohen's Kappa) of the GQS and mDISCERN scores of the two raters on different platforms were 0.78 and 0.72, respectively. The consistency coefficients for the scores of different groups such as professionals, non-professionals and individuals are 0.75 and 0.80 respectively. In summary, videos on TikTok had the highest quality in terms of both GQS and mDISCERN scores. Significant differences were observed among the three platforms (p < 0.001 for all pairwise comparisons) (Table 3). Figures 3 and 4 further confirm this result.

Table 3.

Quality assessment of videos about HPV vaccine on Bilibili/Kwai/TikTok.

	Bilibili			TikTok			Kwai			p-Value
	M	Min–max	P25-p75	M	Min–max	P25-p75	M	Min–max	P25-p75	p(B&T)	p(B&K)	p(T&K)	Cohen's f
GQS	3	1–5	2–4	4	1–5	2–4	3	1–2	1–3	<0.001	<0.001	<0.001	0.39
m DISCERN	2	1–5	2–4	4	1–5	2–4	3	1–5	2–4	<0.001	<0.001	<0.001	0.43

Figure 3.

Violin plots comparing GQS scores across TikTok, Kwai, and Bilibili. X: Video. Y: GQS score (1–5 scale).

Figure 4.

Violin plots comparing mDISCERN scores across TikTok, Kwai, and Bilibili. X: Video. Y: mDISCERN score (1–5 scale).

The analysis of Table 4, which includes the GQS and mDISCERN scores for videos uploaded by professionals, non-professionals, and individual users, indicates significant variations in video quality and reliability. Professionals scored highest with a median GQS of 5 (IQR 4–5) and mDISCERN of 5 (IQR 4–5), indicating higher quality and reliability. Non-professionals and individual users had lower scores, with median GQS of 3 (IQR 2–4) and 2 (IQR 1–3), and mDISCERN scores of 3 (IQR 2–4) and 2 (IQR 1–3), respectively. Statistical analysis showed significant differences in both GQS (p < 0.01) and mDISCERN scores (p < 0.01) among the three groups, as depicted in Figures 3 and 4. These results underscore the importance of professional involvement in creating high-quality and reliable educational content about HPV vaccines. Figures 5 and 6 further confirm this result.

Table 4.

Quality comparison between the videos uploaded by professionals/non-professionals/individual user.

	Professionals (N = 92)			Non-professionals (N = 65)			Individual users (N = 81)
Scores	M	Min–Max	Q1–Q3	M	Min–Max	Q1–Q3	M	Min–Max	Q1–Q3	p (P-N)	p (P-I)	p (I-N)	Cohen's f
GQS	5	2–5	4–5	3	1–4	2–4	2	1–3	1–3	<0.01	<0.01	0.018	0.67
mDISCERN	5	2–5	4–5	3	1–5	2–4	2	1–3	1–3	<0.01	<0.01	<0.01	0.71

Figure 5.

Video quality GQS score by type of uploaders (professionals, non-professionals, individual uesr). X: Type of uploaders. Y: GQS score (1–5 scale).

Figure 6.

Video quality mDISCERN score by type of uploaders (professionals, non-professionals, individual uesr). X: Type of uploaders. Y: mDISCERN score (1–5 scale).

Video categorization

The distribution of video categories and public attitudes toward HPV vaccines varied significantly across Bilibili, Kwai, and TikTok. Videos discussing vaccine safety were more prevalent on TikTok (43.4%) and Bilibili (38.4%) compared to Kwai (30.3%). Content addressing the number of doses required was most common on TikTok (80.3%) and Bilibili (70.9%), significantly higher than on Kwai (51.3%). Videos recommending consultation with a doctor were more frequent on TikTok (17.1%) and Bilibili (15.1%) than on Kwai (3.9%). Regarding public attitudes, pro-vaccination sentiment was most evident on Kwai (63.2%) and TikTok (46.1%), contrasting with Bilibili (29.1%). Anti-vaccination content was notably present on TikTok (15.8%) and Bilibili (8.1%), with a lower percentage on Kwai (5.3%). Neutral attitudes were more common on Kwai (15.8%) and TikTok (19.7%) compared to Bilibili (0.0%) (Table 5). Figure 7 shows the proportion of uploaders’ attitudes across different platforms.

Table 5.

Categorization of videos about HPV vaccine on Bilibili/Kwai/TikTok.

Type of topics, n(%)	Bilibili (n = 86)	Kwai (n = 76)	TikTok (n = 76)	p	Hos	Cramer's V
Mentions vaccines as safe	33 (38.4%)	23 (30.3%)	33 (43.4%)	<0.001	B > K;T > K	0.38
Vaccine side effects	16 (18.6%)	14 (18.4%)	16 (21.1%)	0.42	–	0.24
Age range for vaccine	50 (58.1%)	32 (42.1%)	50 (65.8%)	<0.001	B > K;T > K	0.11
Doses required	61 (70.9%)	39 (51.3%)	61 (80.3%)	<0.001	B > K;T > K	0.19
Recommends talking to doctor	13 (15.1%)	3 (3.9%)	13 (17.1%)	<0.001	B > K;T > K	0.16
Vaccination-related fertility concerns	21 (24.4%)	14 (18.4%)	21 (27.6%)	0.28	B > K;T > K	0.15
Length of vaccination interval	16 (18.6%)	14 (18.4%)	16 (21.1%)	0.51		0.15
Whether to get tested for HPV before vaccination	14 (16.3%)	6 (7.9%)	14 (18.4%)	0.042	B > K;T > K	0.089
Whether regular screening is needed after vaccination	25 (29.1%)	13 (17.1%)	25 (32.9%)	0.007	B > K;T > K	0.13
Public attitude, n(%)
Pro-vaccination	25 (29.1%)	48 (63.2%)	35 (46.1%)	0.02	K > T > B	0.12
Anti-vaccination	7 (8.1%)	4 (5.3%)	12 (15.8%)	0.018	T > B > K	0.15
Neutral	0 (0.0%)	12 (15.8%)	15 (19.7%)	–	–	0.1
Unspecified	54 (62.8%)	12 (15.8%)	14 (18.4%)	<0.001	B > K;B > T	0.32

Note: T stands for Tiktok,B for Bilibili, and K for Kwai. Hos stands for post-hoc test.

Figure 7.

Numbers of video uploaders attitude about HPV vaccine on Bilibili/Kwai/TikTok. X: Percentage (%). Y: Video.

Attitudes toward the HPV vaccine

The analysis of uploader and public attitudes toward HPV vaccination across different uploader identities—professionals, non-professionals, and individual users—reveals distinct patterns. Professionals predominantly exhibited a pro-vaccination attitude (92.3%), significantly higher than non-professionals (83.5%) and individual users (72.8%). Anti-vaccination sentiment was minimal among professionals (0%), compared to non-professionals (1.5%) and individual users (9.9%). Neutral and unspecified attitudes showed no significant differences across the groups.

Conversely, public attitudes reflected a more varied perspective. While professionals still lean toward a pro-vaccination stance (55.4%), non-professionals and individual users showed a significant shift toward unspecified attitudes (38.5% and 45.7%). Anti-vaccination sentiment was more pronounced among the public via individual users (14.8%) compared to professionals and non-professionals (8.7% and 4.6%). These findings underscore the influence of uploader identity on both the expressed attitudes and the perceived public sentiment toward HPV vaccination. Professionals appear to effectively promote pro-vaccination messages, whereas the public's, attitude, especially among non-professional and individual user groups, shows a tendency toward neutrality or lack of specification, indicating potential areas for targeted public health interventions (Table 6). Table 7 presents an analysis of public attitudes toward HPV vaccines as depicted in videos posted by uploaders of varying professional statuses. Findings indicate that videos posted by professionals elicit more positive public responses, with 55.4% expressing pro-vaccination views, significantly higher than the 34.6% observed in videos posted by individual users (p = 0.021). Anti-vaccination attitudes were more prevalent in videos by individual users (14.8%) compared to those by professionals (8.7%), although this difference was not statistically significant (p = 0.284). Neutral attitudes were least common in videos by individual users (4.9%), significantly lower than those by professionals (16.3%). Videos by individual users also had the highest proportion of unspecified attitudes (45.7%), significantly exceeding those by professionals and non-professionals. These data suggest that the identity of video uploaders significantly influences public perceptions of HPV vaccines, with content from professionals more likely to foster positive public sentiment. Figure 8 shows the correlation between the Type of uploaders and uploaders’ attitude and the Type of uploaders and Public attitude.

Table 6.

The attitude between the videos uploaded by professionals/non-professionals/individual user.

Attitude of the uploader (n%)	Professionals (N = 92)	Non-professionals (N = 65)	Individual users (N = 81)	p	Hos	Cramer's V
Pro-vaccination	85 (92.3%)	54 (83.5%)	59 (72.8%)	0.033	P > I > N	0.105
Anti-vaccination	0 (0.00)	1 (1.5%)	8 (9.9%)	0.017	I > N;I > P	0.21
Neutral	3 (3.2%)	4 (6.1%)	6 (7.4%)	0.06	–	0.13
Unspecified	4 (4.3%)	6 (9.2%)	8 (9.8%)	0.046	I > N > P	0.25

Note: P stands for Professionals, N for Non-professionals, and I for Individual users. Hos stands for Post Hoc Inspection. Hos stands for post-hoc test.

Table 7.

The attitude between the public by professionals/non-professionals/individual user.

Attitude of the public (n%)	Professionals (N = 92)	Non-professionals (N = 65)	Individual users (N = 81)	p	Hos	Cramer's V
Pro-vaccination	51 (55.4%)	29 (44.6%)	28 (34.6%)	0.038	P > N;P > I	0.28
Anti-vaccination	8 (8.7%)	3 (4.6%)	12 (14.8%)	0.015	I > P > N	0.14
Neutral	15 (16.3%)	8 (12.3%)	4 (4.9%)	0.003	P > N > I	0.21
Unspecified	18 (19.6%)	25 (38.5%)	37 (45.7%)	0.012	I > N > P	0.22

Note: P stands for Professionals, N for Non-professionals, and I for Individual users. Hos stands for Post Hoc Inspection. Hos stands for post-hoc test.

Figure 8.

Spearman correlations between video quality scores and engagement metrics. The corelation between type of uploaders and uploaders attitude was 0.21 (p < 0.05). The correlation between Type of uploaders and Public attitude was 0.21 (p < 0.01). *p < 0.05, **p < 0.0l, ***p < 0.001.

Correlation analysis

Employing Spearman's rank correlation analysis, we examined associations between the GQS and mDISCERN video quality indicators and audience engagement metrics, including likes, collections, comments, and shares, as detailed in Table 8. Our analysis reveals a positive correlation between video quality, as assessed by GQS and mDISCERN, and viewer engagement metrics. This suggests that superior video quality is associated with increased audience interaction, as reflected in likes, collections, comments, and shares. In contrast, video duration was found to have no significant impact on video quality or viewer engagement. These insights emphasize the critical role of high-quality content in boosting viewer engagement and effectively disseminating information across social media platforms.

Table 8.

Spearman correlation between video quality and audience interaction.

	GQS	mDISCERN	Likes	Collections	Comments	Shares	Video duration	Followers
GQS	1
mDISCERN	0.908 **	1
Likes	0.422 **	0.394 **	1
Collections	0.341 **	0.289 **	0.94 **	1
Comments	0.393 **	0.367 **	0.943 **	0.887 **	1
Shares	0.383 **	0.362 **	0.924 **	0.952 **	0.89 **	1
Video duration	−0.29 **	−0.39 **	−0.102	0.041	−0.091	−0.05	1
Followers	0.426 **	0.443 **	0.729 **	0.65 **	0.629 **	0.658 **	−0.221	1

^**p < 0.01.

Discussion

This study investigates HPV vaccine content on China's three leading short-video platforms—TikTok, Kwai, and Bilibili—through a comprehensive assessment of information quality and dissemination patterns. The findings highlight distinct features and key challenges in disseminating health information across these platforms.^13–16 TikTok, leveraging its algorithmic preference for interactive content, achieved the highest median Global Quality Score (GQS = 4) and mDISCERN score (4), along with a median of 1744.5 likes per video—significantly outperforming the other platforms. Moreover, TikTok demonstrates strong content innovation by integrating real-life physician appearances with animated visualizations. Official media contributes 17.1% of the content, enhancing the platform's perceived credibility.

In contrast, Bilibili hosts the longest videos (median duration = 261 seconds), which are conducive to detailed scientific explanations, yet it records the lowest quality scores (median mDISCERN = 2). Furthermore, 66.3% of the content is produced by self-media accounts, contributing to fragmented and often inconsistent information. For instance, 27% of Bilibili videos omit discussion of vaccine side effects—substantially higher than the 19% reported in Federica et al.'s (2023) analysis of YouTube long-form videos. ¹⁷ Kwai, which has the highest proportion of physician-generated videos (51.3%), demonstrates moderate viewer engagement (median likes = 506.5) and content quality (GQS = 3). This may be attributed to Kwai's user base, which is predominantly composed of audiences from lower-tier regions and tends to be less receptive to technical or medical terminology.

Compared to global platforms, Chinese short-video platforms exhibit a pronounced “decentralization” in content sourcing and curation. For instance, 61.8% of HPV vaccine videos on TikTok are uploaded by medical professionals, in stark contrast to just 7.0% on Bilibili. This discrepancy reflects distinct platform user ecosystems and underscores the need for platform-specific, audience-tailored health communication strategies. To improve information quality consistency, platforms should implement a “health information certification label” that authorizes only verified professional accounts (e.g. certified physicians, licensed hospitals) to publish vaccine-related content. In addition, optimizing recommendation algorithms to prioritize videos with GQS ≥ 4 and filtering out unverified medical claims (e.g. “no need for screening post-vaccination”) could significantly enhance content quality. Artificial intelligence (AI), particularly natural language processing, can be leveraged to detect low-quality videos—for example, by identifying exaggerated or absolutist language—thereby supporting manual content moderation.

The study demonstrates that videos produced by professionals (e.g. licensed physicians) consistently meet high-quality standards (GQS = 5, mDISCERN = 5). However, approximately 30% of self-media videos contain omissions or potentially misleading information. Notably, 19% of videos fail to clearly specify the recommended vaccination age range, simplifying “9–45 years” to “for adolescents,” potentially discouraging eligible older individuals from seeking vaccination. Moreover, the spread of anti-vaccine narratives presents a growing concern. Although comprising just 2.9% of all videos (5.3% on TikTok), anti-vaccine content is shared 2.3 times more frequently than regular content (p < 0.01). For example, a self-media video falsely claiming that “HPV vaccines cause ovarian failure”—despite lacking scientific evidence—accumulated over 100,000 shares due to its sensationalist headline. This phenomenon aligns with Vosoughi et al.'s (2024) model of misinformation virality.^18,19 The harms associated with low-quality videos should not be underestimated, as exposure to misleading HPV vaccine content may lead women to delay or reject vaccination.^20–24 These findings underscore the urgent need for platforms to implement more stringent content moderation mechanisms.^25–27

The study also found that some self-media videos not only omitted essential information—such as vaccine side effects—but actively disseminated conspiratorial and misleading claims. ²³ The rapid diffusion of such content is an escalating concern, as prior research indicates that misinformation spreads significantly faster than factual information due to its emotional salience and shareability.^28–30 Addressing misinformation requires a multifaceted strategy encompassing algorithmic interventions, public education initiatives, and real-time debunking by trusted institutions.

Correlation between video quality, user engagement, and dissemination challenges

Spearman's rank correlation analysis showed that a higher GQS score was moderately correlated with a higher level of participation (r = 0.422, p < 0.001). These findings suggest that platform algorithms tend to prioritize high-quality content from authoritative sources in recommendation feeds. However, algorithmic preference for brevity presents a barrier to comprehensive scientific communication. On TikTok, the median video duration is 70 seconds, whereas professional videos addressing complex topics (e.g. differences between bivalent and 9-valent vaccines) typically require over 100 seconds, limiting their visibility in recommendation algorithms. Moreover, search trends indicate that queries for “HPV vaccine cost” occur at three times the frequency of “HPV vaccine side effects,” suggesting user priorities are skewed toward practical concerns rather than scientific accuracy. Additionally, videos presented in local dialects (e.g. Cantonese on Kwai) receive approximately 40% higher engagement than Mandarin-language videos, underscoring the role of regional cultural adaptation in content creation. Collectively, these findings indicate that both algorithmic structures and user cognition jointly shape the dissemination ecology of health information on short-video platforms.^29,31 Relying solely on “high-quality content” is insufficient to ensure broad reach in algorithm-driven environments.

Based on the study's conclusions, we propose the following improvement pathways: First, Platform Responsibility: Platforms should introduce a “health information certification label,” ensuring that only verified professional accounts (e.g. medical institutions, certified doctors) are allowed to post vaccine-related content. Algorithmic prioritization of high-quality content (GQS ≥ 4) and restrictions on unverified health claims should also be implemented. ³² Second, Medical Institution Engagement: Top-tier hospitals should collaborate with platforms to launch “HPV Vaccine Q&A Live Sessions” to address public concerns in real-time. Case story videos (e.g. “Cervical cancer screening turning negative after vaccination”) should be created to enhance credibility and emotional resonance. Third, Public Education and Media Literacy: Developing short-video health literacy training programs in collaboration with schools and communities would empower users to critically assess online health information. Creating a series of engaging videos debunking common HPV vaccine misconceptions (e.g. “Top 10 Myths about HPV Vaccination”) could effectively counter misinformation. ³³ Fourth, Algorithmic Optimization and AI Moderation: Platforms should refine AI-based detection of misinformation by identifying emotionally charged, misleading narratives and flagging videos containing unverified medical claims. This will help reduce the virality of misinformation while amplifying evidence-based content.

In addition to the proposed solutions, it is crucial to recognize the practical challenges associated with their implementation, as well as the resources required to ensure their success.

Certification labels for health information

Implementing certification labels to verify professional content would require a robust infrastructure to assess and verify the credentials of content creators. Platforms would need to establish clear criteria for verifying professional qualifications (e.g. licensed physicians or accredited medical institutions). A dedicated team or third-party organizations would likely need to manage this verification process, which could incur substantial costs in terms of both human resources and technological support.

Challenges

Verification Process: Identifying credible professionals among a large pool of content creators may be time-consuming and prone to error, especially with self-media content. Resource Requirements: Platforms would need to invest in verification infrastructure, such as credentialing systems and auditing procedures, potentially requiring partnerships with professional medical organizations. Proposed Solutions: Platforms could start by collaborating with medical accreditation bodies to establish clear guidelines for content certification. This could be piloted on a small scale, such as by focusing on specific medical specialties (e.g. gynecology or pediatrics), before scaling to a broader range of health content.

AI moderation for content verification

The use of AI to detect misleading or unverified medical claims in videos presents significant potential. AI systems could be trained to flag videos that contain language typical of misinformation or exaggeration, thus helping to automate the moderation process. However, AI-based moderation still faces several challenges: Accuracy of AI Models: AI models require large amounts of training data to identify misinformation effectively, and even then, they can produce false positives (flagging accurate content) or false negatives (failing to flag misleading content). Cultural Sensitivity: AI models may struggle with understanding culturally specific nuances or the context of health discussions, leading to errors in flagging content. Resource Requirements: Developing and maintaining these AI systems requires significant investment in machine learning resources and technical expertise. Proposed Solutions: Platforms could partner with AI research institutions to develop more advanced and context-aware content moderation systems. They could also implement a tiered approach where AI flags content for human review, allowing for faster moderation without sacrificing accuracy. Collaboration with medical experts could help refine AI models and improve their accuracy over time.

To the best of our knowledge, this is the first study to simultaneously apply both the GQS and the mDISCERN instrument to short-form HPV-vaccine videos on TikTok, Kwai and Bilibili, and demonstrate that platform-specific algorithmic and cultural factors moderate the reach of high-quality health information. Prior literature has either focused on Western platforms (YouTube) or single Chinese platforms, has lumped all non-physicians into one heterogeneous “non-professional” category, and has rarely linked uploader identity to measurable audience attitudes. By closing these gaps, our findings provide actionable evidence for platform-level governance, accreditation policies for health content creators, and targeted public-health messaging in low- and middle-income settings where short-video platforms are the dominant source of health information.

This study systematically reveals the dissemination features and challenges of HPV vaccine content on China's short-video platforms in 2025, providing data-driven insights for optimizing the health information ecosystem. Moving forward, fostering a collaborative health communication model involving platforms, institutions, and the public will be crucial to establishing a scientifically accurate, inclusive, and effective health communication system, ultimately achieving the goal of “empowering health through knowledge.”

This study has several limitations: firstly, the timeliness of data. Although we utilized the most up-to-date 2025 data, short-video platform algorithms are frequently updated, which may affect the long-term applicability of the conclusions. Secondly, sample representativeness: This study focused on 238 videos from the three major platforms. Future research should expand to more platforms (e.g. WeChat Video Accounts, Xiaohongshu) and multi-language content (e.g. videos in ethnic minority languages). Thirdly, cultural and linguistic factors: Future research should explore cross-cultural comparisons, analyzing differences in HPV vaccine narratives across Chinese and Western platforms (e.g. TikTok vs. TikTok) and the impact of cultural factors on information reception. Fourthly, while our analyses show that uploader identity (professional vs. non-professional vs. self-media) is associated with both content quality and the sentiment of the top-liked comment, we recognize that uploader status is not the sole determinant of audience attitude. Topic emphasis (e.g. safety vs. fertility concerns), tone (fear-appeal vs. reassurance), narrative style (story-telling vs. didactic), and cultural sensitivities (e.g. taboos around female sexuality, regional dialect use) can independently shape viewer reactions. These variables were not explicitly coded in the present study and therefore represent unmeasured confounders. Future mixed-methods work that integrates qualitative coding of tone and cultural cues with multivariable regression modeling is needed to disentangle the relative impact of each factor. Fifthly, algorithmic bias is in default rankings. Although we conducted searches in incognito mode and without user log-in, the platforms’ “comprehensive” or “default” ranking is a proprietary composite score that incorporates real-time engagement velocity, device fingerprinting, geographic signals, and content freshness. These factors create algorithmic feedback loops that favor videos with rapid early interactions, potentially over-representing sensational or emotionally charged content and under-representing high-quality but slower-diffusing material. Because the exact weightings are undisclosed and may differ across TikTok, Kwai, and Bilibili, our sample should be viewed as a snapshot of what an average logged-out user in mainland China would see at a single moment, rather than a fully unbiased cross-section of all HPV-vaccine videos. This algorithmic bias constitutes a major limitation that affects the generalizability of our findings to the totality of content available on each platform and to users in other regions or demographic groups. We explicitly acknowledge this potential bias and its impact on the generalizability of our findings. The videos included in our analysis may not fully reflect the overall spectrum of HPV vaccine-related content on these platforms, particularly under-representing educational content that may not achieve rapid or high engagement. For example, videos featuring professional health experts may receive fewer initial interactions, thereby limiting their visibility in the algorithmic ranking. To address this, we recommend that future studies incorporate additional sampling techniques, such as random sampling, to better capture a wider variety of content. Implementing such methods would help mitigate the bias associated with algorithmic ranking and ensure a more comprehensive assessment of the quality and diversity of health-related videos. Moreover, platform algorithms could be optimized to prioritize the reliability of content by integrating more stringent content validation mechanisms or promoting videos with established medical accuracy, rather than relying solely on engagement metrics. Additionally, we propose that future research explore the use of API-based sampling or systematic datasets provided by platforms, which would allow for a more representative sample of content and reduce biases related to default rankings. These adjustments could help ensure that health-related information shared via social media platforms accurately reflects the range of available knowledge, ultimately improving public access to reliable health information. By highlighting these algorithmic biases and suggesting improvements, we hope to inspire further research and platform-level interventions to enhance the credibility and reliability of health information disseminated through short-video platforms. Sixth, confined to Simplified-Chinese short videos, our findings are embedded in a Chinese cultural context: the pronounced trust in licensed physicians, the euphemistic framing of HPV discussions, platform-specific features such as real-name physician badges and e-commerce vouchers, and collectivist family decision-making norms may all reduce applicability outside Chinese-speaking or culturally distinct populations. Replication across other languages and social settings is therefore essential. Seventh, while this study provides valuable insights into the dissemination of HPV vaccine information on short-video platforms, it is important to recognize the role of cultural context in shaping public perceptions and behaviors related to vaccination. In China, healthcare authority structures and family decision-making patterns significantly influence how information is received and acted upon. First, Chinese healthcare authority structures play a central role in shaping public attitudes toward medical information. Medical professionals, particularly physicians, hold substantial influence in Chinese society, where their recommendations are often viewed with high respect. This cultural trust in medical professionals is evident in our findings, where videos uploaded by doctors, especially those from well-regarded institutions, garnered higher levels of public approval and engagement. In contrast, content from non-professional or self-media sources often received lower engagement, underlining the public's reliance on trusted health authorities for health-related guidance. Second, family decision-making patterns, which are deeply rooted in Chinese collectivist values, also impact how health decisions, such as vaccination, are made. In Chinese culture, decisions regarding children's health, including vaccination, are frequently made within the family unit, with significant input from parents and extended family members. This familial influence can shape an individual's attitude toward vaccination, especially for younger individuals. Therefore, health information conveyed through social media platforms may be further influenced or scrutinized through family discussions, where parents' concerns or support can either reinforce or challenge the information presented. These cultural nuances are particularly important when considering HPV vaccination, as it is often perceived as a sensitive issue tied to sexual health, which may prompt additional family-based deliberation. By incorporating these cultural factors, we gain a more comprehensive understanding of how short-video platforms influence public health behavior in China. The interaction between authority structures, family decision-making, and digital health content underscores the importance of tailoring public health campaigns to align with these cultural dynamics. Finally, AI integration for content moderation: Future studies should investigate the effectiveness of AI-driven misinformation detection tools and how they can be integrated into content moderation frameworks.

Conclusion

Our research results indicate significant differences in the quality and reliability of videos related to the HPV vaccine on Kwai, Bilibili, and TikTok platforms. TikTok videos had the highest quality in terms of GQS and mDISCERN scores, and the quality and reliability are closely related to whether the uploader is a professional. The identity of the video uploader significantly influences the public's perception of the HPV vaccine. Therefore, we advocate for more professionals to create specialized videos about the HPV vaccine to provide the public with reference information, effectively increase HPV vaccination rates, and reduce the incidence of cervical cancer related to HPV, benefiting the population. Additionally, it is recommended that social media platforms strengthen the authentication of uploaders and review the content of uploaded videos to control public discourse and prevent unprofessional information from spreading among the population, ensuring the public receives accurate, professional, and high-quality videos.

Abbreviations

GQS

Global Quality Score

mDISCERN

modified DISCERN

AI

artificial intelligence