Douyin fitness videos often omit key safety information, content analysis finds

A diagnostic analysis of 250 publicly available exercise and health videos on the Chinese short-video platform Douyin found that more than half contained at least one severe problem in how they communicated expected effects, suitable users or safety conditions. Researchers assessed whether videos specified who could follow along, when to stop, what risks might occur, and where their exercise guidance came from. They did not test the clinical effectiveness of the demonstrated movements, follow users’ physical outcomes or document injuries. Instead, the study examined the information supplied alongside each routine, treating a follow-along video as a form of executable health communication in which viewers may act on the presentation rather than merely read or watch it.

The most common severe expressive problem was inadequate risk warning, found in 36.8% of the sampled videos. Excessive efficacy claims met the study’s severe-problem criterion in 34.4%, while severely unclear applicability conditions occurred in 25.6%. Severely deficient professional justification was much less common, affecting 2.0%. Overall, 54.4% of videos had at least one severe problem, 36.0% had two or more, and 7.6% had at least three. The most common mutually exclusive combination was unclear applicability plus insufficient risk warnings, present in 35 videos, or 14.0% of the sample. Sixteen videos, representing 6.4%, simultaneously met the severe criteria for excessive efficacy claims, unclear applicability and inadequate warnings.

Researchers used a diagnostic evaluative content-analysis method with a frozen coding manual. They screened candidates against six requirements: videos had to be publicly verifiable, aimed at ordinary users, directly imitable, oriented toward health or physical improvement, non-medical in diagnostic or prescriptive focus, and not primarily commercial. The final sample was drawn from searches covering posture correction, pain relief and general rehabilitation, fat loss and body shaping, stretching and desk exercise, and Baduanjin or wellness exercises. Videos published from January 2024 through July 2026 were collected on August 17–19, 2026, using the platform’s default ranking order. Because sampling followed those rankings rather than a probability design, the results describe the retrieved quota sample rather than all Douyin content.

After screening, each video was assigned a content category and scored on four dimensions. Efficacy claims concerned whether a routine was linked too strongly or absolutely to a physical outcome. Applicability boundaries concerned identification of suitable and unsuitable users, conditions, prerequisites or restrictions. Risk warnings covered contraindications, precautions, pain signals, excessive load and stopping conditions. Professional justification covered the creator’s identity, sources, teaching materials, research basis or rationale for the movements. A score of 1 indicated no evident problem, 2 a general insufficiency and 3 a dimension-specific severe trigger. Such phrases as “one move” or “beginner” did not automatically produce the highest score; the outcome claim or broader suitability statement had to meet the predefined trigger.

Formal coding followed a pilot on 50 videos, ten from each sampling stratum. Two coders then independently assessed a random set of 50 videos, again with ten from each stratum, using a recorded random seed. Before discussion, they agreed on all 50 content classifications and on 243 of 250 dimension-specific judgments, an overall raw agreement rate of 97.2%. Unweighted Cohen’s kappa for content type was 1.000. Linearly weighted kappa was 0.945 for efficacy claims, 0.807 for applicability boundaries, 0.963 for risk warnings and 0.947 for professional justification. The authors noted seven disagreements in the four scored dimensions, including unresolved evidence or consistency issues for videos 77, 81 and 90 that required verification against the original recordings and reconciliation with the analysis dataset.

The composition of the final sample after recoding was 60 posture-correction videos, 59 pain-relief or general-rehabilitation videos, 50 fat-loss and body-shaping videos, 41 Baduanjin or wellness-exercise videos, and 40 stretching, desk-exercise or brief-movement videos. The patterns differed markedly by subject. Among fat-loss and body-shaping videos, 86.0% met the severe criterion for efficacy claims, compared with 30.0% of posture-correction videos and 30.5% of pain-relief or general-rehabilitation videos. Such content often associated movements with rapid weight loss, localized shaping or visible physical changes. The study could identify the prominence of these expressions, but it could not determine whether they changed viewers’ expectations or influenced behavior.

Pain-relief and general-rehabilitation content showed a different profile. A total of 86.4% met the severe criterion for unclear applicability boundaries, and 81.4% met the severe criterion for insufficient risk warnings. These videos frequently used low-back, neck, shoulder, knee or sciatica-related discomfort as their starting point, yet often did not simultaneously identify unsuitable users, contraindications, stopping conditions or situations in which further assessment might be needed. For fat-loss and body-shaping videos, 58.0% met the severe risk-warning criterion, with gaps particularly associated with high-intensity, continuous, weight-bearing or fat-burning routines. Stretching and desk-exercise videos had no severe risk-warning cases in the final classification, although inadequate efficacy or applicability information was still present.

Content type was statistically associated with the distribution of severe problems in efficacy claims, applicability boundaries and risk warnings; all three omnibus comparisons had Holm-adjusted p-values below 0.001. The corresponding Cramér’s V values were 0.579, 0.778 and 0.648. Professional justification was the exception: its overall association with content type did not reach significance in a two-sided Fisher–Freeman–Halton exact test, which yielded p = 0.369 and Cramér’s V = 0.138. The sparse distribution of five severe cases limited comparisons on that dimension. These tests describe within-sample associations and should not be read as evidence that one content type causes user behavior or injury.

The authors argue that these findings support evaluating whether follow-along content supplies the information needed to make a basic suitability and risk judgment, rather than treating every incomplete video as demonstrably false. Their results suggest that governance should be tailored to content: body-shaping videos need clearer outcome conditions and training intensity; pain and rehabilitation videos need unsuitable populations, contraindications and stopping rules; and lower-intensity routines still need applicable conditions. The analysis did not test platform review processes or governance effects, and its sample cannot represent every platform, all exercise content or actual follow-along behavior. It therefore identifies gaps in public-facing information, rather than estimating injury rates, proving clinical harm or showing how viewers acted on the videos.

Subject of Research: Diagnostic analysis of risk communication and information completeness in 250 Douyin follow-along fitness videos.

Article Title: Public-health risk expression and governance of low-threshold exercise and health content on Douyin short-video platforms: a case study of follow-along fitness short videos

Article References: Zhang, Y., Qu, Z., Guo, X., & Zhao, Y. (2026). Public-health risk expression and governance of low-threshold exercise and health content on Douyin short-video platforms: a case study of follow-along fitness short videos. Frontiers in Public Health, 14, Article 1985289. https://doi.org/10.3389/fpubh.2026.1985289

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1985289

Keywords: Douyin, short-video health communication, fitness videos, risk warnings, applicability boundaries, efficacy claims, content analysis, platform governance, Public-health, risk, expression, governance