Abstract
Videos in endless loop (VEL) are a popular form of entertainment on social media platforms such as TikTok and Instagram, with an estimated 1.5 billion users on TikTok alone in 2025. Despite their popularity, concerns have emerged regarding their impact on users’ perception of time. This study (N = 151) investigated the relationship between watching VEL and time perception, valence, and arousal. A novel digital time-guessing task was developed and embedded within a 14-day experience sampling design to ensure ecological validity and assess the real-life effects of VEL consumption. In addition, a short-term retest approach, nested within the daily assessments, allowed us to examine the temporal dynamics of possible effects. While no immediate differences in time perception were found between the reference measurement and directly after VEL consumption, significant differences in time estimation were observed 15 and 30 minutes later. In contrast, VEL had no lasting effects on participants’ valence and arousal, apart from brief changes directly after viewing. These findings indicate that watching VEL is associated with subjective time perception, albeit with a delayed effect. Given the small-to-medium effect size (β ≈ 0.20) within this specific sample (young, educated, female, and German-speaking) and the high frequency with which individuals engage with VEL in daily life, this effect may have meaningful implications for users’ temporal orientation and media consumption behavior.
Introduction
With an estimated 1.5 billion users, TikTok is among the most widely used social media platforms in 2025. 1 It is characterized by short, vertical videos presented in an endless feed curated through a highly precise algorithm. These three features form the basis of a new entertainment format referred to here as videos in endless loop (VEL). Other platforms have adopted similar formats, 2 such as Instagram Reels or Facebook Reels. Therefore, VEL unites both typical short-form video applications 3 (e.g., TikTok) and sub-formats of other social media platforms characterized by the same three features (e.g., Reels), even when these are not the primary feature of the respective platform (e.g., Instagram: videos, but also pictures and text). Given that these platforms collectively reach billions of users, 1 a large proportion of the global population may be affected by consequences that remain largely unexplored.
A review of the current literature reveals that most studies on digital media consumption have focused on general smartphone use, social media use, or specifically TikTok use, with particular attention to effects on mental health and cognition. In terms of mental health, higher usage has been associated with increased risks of depression, perceived stress, anxiety, and reduced feelings of social connectedness.4,5 From a cognitive standpoint, smartphone use has been shown to negatively impact reaction time, working memory, 6 and self-perceived attention performance. 7 Excessive TikTok use has also been linked to memory impairments 4 and reduced analytical thinking. 8 However, large-scale studies suggest that the overall cognitive effects of social media use are relatively small 9 (N = 12,043).
Flow effect
One frequently reported yet understudied phenomenon on VEL platforms is that users become fully immersed in watching videos that they face difficulties in stopping or resisting their urge to continue. 10 This may be explained by the flow effect—a state often triggered during social media use 11 —which mediates the relationship between user motivation and problematic usage, 12 and has been specifically observed on TikTok. 13 Within this context, time distortion and concentration difficulties are particularly associated with excessive use. 14 Compared with reading, TikTok use results in significantly greater time distortion 15 and induces stronger and more frequent flow experiences than Instagram. 16
Time perception
Time perception is defined as the subjective experience of the passage of time 17 and can be further categorized into succession perception and duration perception. 18 For the purposes of this study, duration perception is of greater importance and is defined as the perceived interval between the onset and offset of an event. The cognitive mechanisms underlying this form of perception are memory and attention. Time distortion is defined as any deviation between real time and perceived time.
Current study
The present study diverges from prior research by focusing on the immediate effects of VEL. Its primary aim is to examine how VEL consumption correlates with users’ time perception and whether these effects persist over time. In addition, affective responses were assessed using the Affect Grid 19 to explore changes in valence and arousal following VEL consumption.
Importantly, this study used a field-based experience sampling method (ESM) via participants’ own smartphones, enhancing ecological validity. Unlike prior cross-sectional or laboratory-based studies, our design enables measurement directly after VEL exposure without interrupting natural viewing patterns or disrupting the underlying algorithm. This is particularly relevant given the algorithm’s central role in shaping this entertainment format and its potential contribution to excessive engagement. 10
Methods
Participants
Participants were recruited via various social media platforms in two phases, corresponding to adjustments made to the power analysis. According to this analysis, the recommended sample size was N = 120 participants (for further details, see Supplementary Data S1). Phase 1 began in June 2024, followed by phase 2 in November 2024. In total, 182 individuals were recruited. After data cleaning, 159 participants remained in the sample. We excluded duplicate cases (e.g., completing the initial questionnaire twice) and participants who enrolled in the study but did not complete a single questionnaire. Further eight participants had to be excluded from analysis after we excluded invalid cases (i.e., retest intervals were too broad; for detailed exclusion criteria, see https://osf.io/ts42g). Therefore, the final analyzed sample consisted of 151 participants. Twenty-six percent of participants dropped out before the end of the study. Nevertheless, because missing data are not problematic for multilevel models and the final questionnaire assessed online cognition (Online Cognition Scale) in addition to a retest of sociodemographic variables, we decided not to exclude these participants. Based on the overall sample, the mean compliance rate was 12.8 completed time-based questionnaires (out of 28 possible; standard deviation SD = 9.2).
The mean age of participants was 27.0 years (SD = 8.53). The majority identified as female (59.6%), followed by male (34.4%) and other (1.3%), while 4.6% did not report their gender. In terms of nationality, most participants were from Austria (51.7%) or Germany (35.8%), whereas 8.6% reported another nationality (4.0% missing data). Regarding relationship status, 44.4% reported being in a relationship, 36.4% were single, 11.3% were married, and 2.6% were divorced (5.3% missing data). Educational attainment was relatively high: 45.0% had completed A-levels, 37.1% held a university degree, 8.6% had a vocational qualification, 5.3% had completed only mandatory schooling, and 4.0% had not completed mandatory schooling (4.0% missing data).
Measures and procedure
Data were collected using the scientific smartphone application ESMira. 20 Over the 14-day study period, participants completed a series of questionnaires, including both cross-sectional and longitudinal measures (for an overview, see Figure 1). To take part in the study, individuals were required to download the ESMira app onto their smartphones and enroll via a QR code or passphrase. Upon enrollment, participants received detailed information about the study’s purpose and procedures and provided informed consent before participation.
Procedural flowchart of the ESM study design. ESM, experience sampling method.
The initial questionnaire was delivered to participants via app notification 60 seconds after entering the study in ESMira. This questionnaire assessed sociodemographic information cross-sectionally, including participants’ gender (male, female, and other), age (in years), current relationship status (single, in a relationship, married, widowed, and divorced), highest level of education (no mandatory school, mandatory school, vocational qualification, A-levels, and university degree), and nationality (Austrian, German, and other).
The study included three main types of longitudinal questionnaires: (a) time-based prompts at random intervals (reference measurement for the time-guessing task), (b) event-based prompts following self-initiated reports of VEL consumption, and (c) retest prompts administered 15 and 30 minutes after the event-based questionnaire. All longitudinal questionnaires incorporated the German version of the Russell Affect Grid, 19 a Visual Analog Scale (VAS) measuring valence (pleasure) and arousal. The grid ranges from 1 (unpleasant [sehr unangenehm]; sleepy [Schläfrigkeit]) to 100 (very pleasant [sehr angenehm]; highly aroused [hohe Erregung]).
The central element of all longitudinal questionnaires was a custom-designed task aimed at assessing participants’ time perception (i.e., the time-guessing task). The task followed a standardized structure: First, participants were informed about the procedure. A start button was then presented, which triggered an invisible timer set to a random duration between 5 and 20 seconds. The exact duration varied randomly with each instance, except for the retests, which used the same duration as the initial event-based measurement. When the timer expired, a green checkmark appeared on the screen, and a brief sound was played to alert participants (see Supplementary Figure S1). On the subsequent screen, participants were asked to estimate the elapsed time between pressing the start button and the appearance of the green checkmark. This estimate served as the operational measure of subjective time perception. Participants were explicitly instructed not to use any technological aids and to rely solely on their intuitive sense of time.
To account for potential confounding effects due to the naturalistic setting of the study, an additional item was included to assess perceived environmental distraction while completing the questionnaires. This was measured using a VAS ranging from 1 (not distracting [reizarm: leise, ruhig]) to 100 (overstimulating [Reizüberflutung: laut, geschäftig]). This variable is referred to as the level of distraction.
The term VEL was uniformly defined for participants as “videos in endless loop.” This definition encompassed video formats such as TikTok (e.g., Instagram Reels); several platforms were provided as response options to support participants’ understanding of the format. The duration of consumption was not specified because there was no a priori knowledge of the time required to develop a flow-like state while watching. Nevertheless, to establish a methodological foundation for the study, participants were instructed to complete the event-based questionnaire once they felt immersed in watching, as this experience often occurs during a flow-like state in which time distortion is expected.
During the time-based reference measurements, participants were asked when they had last viewed VEL content. Then, they had to do a time-guessing task as in the event-based questionnaire in order to have a reference about how accurate participants can guess time durations without watching VEL. In the retest questionnaires following the event-based assessments, they were asked whether they had watched any VEL content since completing the previous questionnaire. These items were later used during data cleaning to ensure that measurements were not confounded by intervening VEL exposure (e.g., notification for the time-based reference measurement comes during a VEL session).
During the 14-day study period, participants received notifications for the time-based questionnaire (i.e., the reference measurement of time perception) at two random time points per day between 8:00 a.m. and 8:00 p.m. (with optional adjustment by the participants). In contrast, the event-based questionnaire was self-initiated and had to be completed whenever participants had watched VEL content. To enhance adherence, participants were reminded once per day to complete the event-based questionnaire whenever they watched VEL for a longer period, for example, through an informational message displayed at the end of the second time-based questionnaire. Retest notifications were automatically sent 15 and 30 minutes after completion of each event-based questionnaire.
At the end of the 14-day testing period, participants completed a final questionnaire that repeated the initial sociodemographic items and included the German version of the Online Cognition Scale, 21 which measures excessive Internet use. The inclusion of this scale was informed by previous findings indicating a link between time distortion and problematic smartphone 22 or social media use.12,14
Statistical analyses
We computed a bias score as the difference between the actual timer duration (randomly set between 5 and 20 seconds) and participants’ estimated duration. Because the distribution contained extreme values, we excluded bias scores exceeding ±20 seconds (4.1%) from further analyses. This corresponds to a very conservative exclusion criterion of ∼±4 SD (SD ≈ 5 seconds) and removed only a small number of clearly aberrant values (e.g., implausible four-digit responses). The resulting bias variable had a mean of 1.11 seconds (SD = 5.54; range = −18 to 19; n = 3,770). Skewness (S = 0.64) and kurtosis (K = 0.87) indicated only minor deviations from normality.
We employed multilevel regression analyses using R for all statistical analyses. A two-level model was applied, with measurement occasions (level 1) nested within participants (level 2). The reference category for the short-term retests was the time-based reference measurement (i.e., without recent VEL exposure). For further information on the statistical analyses, please see Supplementary Material.
Results
We collected demographic data in both the initial and final surveys. Only two participants showed inconsistencies in either gender or age. Since the remainder of the data appeared consistent and free from anomalies, these two cases were retained in the dataset.
To ensure that time-based reference measurements were not confounded by recent VEL usage, participants were asked when they last used VEL. If the interval was <30 minutes, data from that specific questionnaire were excluded from the analyses. On average, participants had last used VEL ∼5 hours before completing the reference measurements (median = 300 minutes; range = 30–12,780 minutes). We also examined whether participants had used VEL between the event-based and retest measurements. If so, the corresponding data were excluded as well.
Over the 14-day assessment period, participants reported a total of 1,017 VEL sessions. The most frequently used platforms were TikTok (42.4%) and Instagram (36.5%), followed by YouTube (12.2%), Facebook (5.9%), and Snapchat (3.0%).
As shown in Table 1, we observed a time-estimation bias of ∼1.3 seconds during the reference measurements. This bias persisted even after controlling for situational distraction, participant gender, age, and general Internet use patterns (see Supplementary Table S1). Participants tended to overestimate the actual duration by ∼1.3 seconds. This aligns with previous findings showing similar overestimation biases in time-guessing tasks (e.g., Polti et al. 23 ; ∼7-second overestimation for a 30-second duration).
Results of the Multilevel Analyses
Note: Reference category was the time-based reference assessment randomly chosen at two time points within a day (i.e., time points when participants had not watched VEL). Gender = category ‘other’ was excluded because only two participants: 1 = men, 2 = women. First retest after 15 minutes and second retest after 30 minutes. CI, 95% confidence interval; β, standardized B; ICC, intraclass correlation coefficient; OCS, Online Cognition Scale; SD, standard deviation.
+p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001 (two-sided).
Directly after watching VEL, this naturally occurring bias decreased slightly (by 0.16 seconds), although the change was not statistically significant. However, the reduction became significant after 15 minutes (difference = 1.01 seconds) and 30 minutes (difference = 1.09 seconds), both corresponding to small-to-medium effect sizes (see Figure 2). In summary, watching VEL was significantly associated with systematic differences in time perception, manifesting as a measurable underestimation of time that persisted for at least 30 minutes.
Time-perception changes during a complete assessment cycle (reference, immediately after VEL, 15 minutes retest, 30 minutes retest). VEL, videos in endless loop.
Interestingly, participants’ arousal levels were slightly reduced immediately after watching VEL (small effect size; see Table 1), but this effect diminished within 15 minutes. Moreover, participants did not report increased pleasure after watching VEL, either immediately or 15 or 30 minutes later, compared with the reference measurement. Notably, when situational distraction was not controlled for, pleasure ratings were significantly higher immediately and 15 minutes after watching VEL (small effect size; see Supplementary Table S1). This suggests that perceived pleasure is more likely attributed to the viewing context (e.g., relaxed or undisturbed settings) rather than to the content of VEL itself.
Discussion
Although the immediate effects of VEL on time perception appeared negligible, the findings indicate a delayed impact, thereby supporting the initial hypothesis. Specifically, participants tended to underestimate time after consuming VEL. The other variables, arousal and pleasure, showed small to negligible effects; however, both were correlated with the level of situational distraction. This finding is noteworthy, as watching VEL did not result in participants feeling better afterward.
The acceptance of the hypothesis is consistent with findings from recent research.14,15 Indeed, VEL consumption was significantly associated with individuals’ perception of time. However, the newly observed phenomenon that this time distortion occurs with a delay following consumption has not been previously documented. Possible explanations include cognitive post-processing, as time perception relies on multiple cognitive mechanisms, such as memory and attention, 18 or recovery from flow states, 11 which often occur during VEL consumption and are associated with time distortion. However, these possibilities require further investigation.
The time-guessing task
For the purposes of this study, a newly developed, self-designed task was introduced to assess individuals’ time perception. Given that this was the task’s first implementation, it is important to highlight several key observations in relation to the study’s findings. The results show that participants generally overestimated time during both the reference measurement and the measurement taken immediately after watching VEL. This pattern of overestimation is consistent with findings by Polti et al., 23 thereby supporting the validity and effectiveness of the newly introduced task.
Limitations and further directions
One limitation of this study is the homogeneity of the sample. The predominance of highly educated female participants from German-speaking regions may introduce bias, as it does not fully reflect the broader social media audience. 1 Although our analyses indicated no substantial effects of age or gender on time perception, the implications of this imbalance should be considered.
A further limitation is that, despite adequate power, we did not run platform-specific models for each social media service offering VEL. 2 Platform effects may differ in strength (e.g., TikTok is often described as more addictive than other platforms), 16 potentially because of differences in recommendation algorithms. As prior work rarely distinguishes between platform-wide effects and the VEL feature itself, it would be valuable and necessary for future research to compare VEL implementations across platforms, as results may differ.
Repeated assessments may also be prone to habituation or sensitivity effects (i.e., participants may “improve” with repetition). Disentangling such effects from VEL-related changes in time perception would require reference measurements not only for the time-guessing task (included here) but also for the short-term retests.
Relatedly, retests were scheduled 15 and 30 minutes after the event-based questionnaire. More frequent retests and a longer window (e.g., up to 60 minutes) could map the onset and duration of effects more precisely and help rule out accumulation or overlap with subsequent VEL sessions.
Furthermore, lengthening the task interval (e.g., up to multiple minutes) could strengthen the interpretation of the observed estimation shift. Longer intervals may be associated with greater time distortion, suggesting that the relatively small distortion observed in the present study (1.0–1.3 seconds) could scale to more substantial distortions under extended temporal conditions.
Finally, the field-like nature of the ESM design limited the assessment of contextual moderators, such as multitasking, viewing context, or the emotional valence of the viewed content. Given participant burden, these variables were not assessed in detail; future studies could therefore adopt a more laboratory-like setting to measure or control these factors.
It would also be worthwhile to assess the duration of VEL engagement prior to questionnaire completion. Prior research links greater screen time to poorer time perception, 22 and this may generalize to VEL, given the underestimation observed in our data. However, because subjective reports often diverge from objective screen-time measures, 24 and because objective tracking of VEL usage was not technically feasible, this variable could not be integrated. Future research should incorporate reliable objective measures of VEL exposure to strengthen the robustness of conclusions.
Conclusion
This study provides evidence for a significant association between watching VEL and subjective time perception, specifically leading to a delayed underestimation of duration following consumption. This effect persisted for at least 30 minutes and occurred independently of all covariates, situational distraction, arousal, and pleasure.
The study’s design—characterized by high ecological validity, multiple measurement time points, and a focus on real-life usage patterns—enabled novel insights, particularly regarding the delayed onset of time distortion. While certain limitations, such as the inability to differentiate between platform-specific effects or to measure VEL session duration objectively, constrained the scope of interpretation, they also point to promising directions for future research. Addressing these limitations through refined methodologies, such as extended retest intervals, platform comparisons, and objective screen-time tracking, could deepen our understanding of VEL’s cognitive and emotional impact.
Authors’ Contributions
A.S.: Conceptualization, investigation, methodology, data curation, and writing—original draft. S.S.: Conceptualization, formal analysis, data curation, supervision, methodology, writing—review and editing, resources, and visualization. S.V.: Methodology and writing—review and editing.
Availability of Data and Materials
The raw data and materials can be found at https://osf.io/ts42g/.
Code Availability
The code used for the analyses is available in the same OSF repository, https://osf.io/ts42g/. The software used, ESMira, is an open-source project, which can be found at https://github.com/KL-Psychological-Methodology/ESMira.
Footnotes
Acknowledgment
The author would like to thank Elena Prochaska for her help in collecting part of the data. The article was revised by ChatGPT-5.2 to improve style and grammar while retaining the authors’ tone. We acknowledge support by Open Access Publishing Fund of Karl Landsteiner University, Krems, Austria.
Ethics Approval
The study was conducted in accordance with the Declaration of Helsinki, the standards of German and Austrian Psychological Societies, the American Psychological Association, and the guidelines of the Karl Landsteiner University of Health Sciences. The study was noninvasive and did not include institutionalized participants (e.g., patients). Participants were 18 years of age or older, and no health-related issues were assessed. Informed consent was obtained from all individual participants. Participation was voluntary and anonymous, and participants had the right to withdraw from the study at any time without penalty. Therefore, the study was exempt from formal approval by the ethics committee (Waiver number 1039/2024).
Author Disclosure Statement
The authors declared no potential conflicts of interest.
Funding Information
No funding was received for this article.