Source-linked AI summary
How the Design of YouTube Influences User Sense of Agency
Kai Lukoff, Ulrik Lyngs, Himanshu Zade, J. Vera Liao, James Choi, Kaiyue Fan, Sean A. Munson, Alexis Hiniker
TL;DR
The paper addresses how app design can undermine users’ agency and whether internal mechanisms can better support it than external controls. Using a survey of YouTube users and co-design sessions, it finds that autoplay and recommendations generally reduce agency, while search and playlists support it, with preferences depending on users’ intentions.
Problem
App mechanisms such as autoplay can undermine users’ sense of agency, which is associated with negative life effects, while prior digital-wellbeing work has focused mainly on external controls.
Method
The paper studies YouTube’s internal mechanisms through a survey of 120 users followed by co-design sessions evaluating potential redesigns.
Results
Recommendations, ads, and autoplay mostly reduced perceived control, whereas playlists, search, subscriptions, play controls, and watch history and stats mostly increased it; users preferred more control for specific intentions.
Takeaways & Limitations
Designers should support user agency through internal mechanisms such as aspiration-aligned recommendations and lightweight microplanning, while recognizing that users do not always want more control.
Takeaways & Limitations
Both studies measured participants’ stated preferences rather than their behavior in situ, so actual responses to agency-oriented redesigns may differ.
Abstract
from arXiv · showhide
In the attention economy, video apps employ design mechanisms like autoplay that exploit psychological vulnerabilities to maximize watch time. Consequently, many people feel a lack of agency over their app use, which is linked to negative life effects such as loss of sleep. Prior design research has innovated external mechanisms that police multiple apps, such as lockout timers. In this work, we shift the focus to how the internal mechanisms of an app can support user agency, taking the popular YouTube mobile app as a test case. From a survey of 120 U.S. users, we find that autoplay and recommendations primarily undermine sense of agency, while search and playlists support it. From 13 co-design sessions, we find that when users have a specific intention for how they want to use YouTube they prefer interfaces that support greater agency. We discuss implications for how designers can help users reclaim a sense of agency over their media use.
1 INTRODUCTION
The paper shifts digital-wellbeing design from external controls toward redesigning YouTube’s internal mechanisms to support user agency. Two studies examine which mechanisms influence agency and when users prefer greater control.
- Attention-economy mechanisms such as autoplay and clickbait seek to maximize users’ time while potentially exploiting psychological vulnerabilities.
- Low agency over technology use is associated with conflicts between media consumption and personal goals, problematic use, and negative effects including lost sleep.
- External interventions such as lockout timers monitor or police multiple apps without changing the internal mechanisms that make an app problematic.
- The studies investigate existing YouTube mechanisms and test redesigns of recommendations, playlists, search, and autoplay with users.
- The work identifies internal mechanisms that influence agency and proposes microplans that can guide behavior within a single session.
- Users preferred greater control for specific intentions, but preferred letting YouTube take control when their intention was nonspecific.
2 BACKGROUND AND MOTIVATION
The background contrasts attention-capturing and external-control approaches with designing YouTube’s internal mechanisms around user agency. It motivates agency as a more targeted alternative to reducing screen time alone.
- Attention-capture dark patterns manipulate users into spending time and attention in apps against their best interests.
- YouTube is a prominent setting for studying attention capture because it is widely used, often reported as distracting, and recommendations drive more than 70% of watchtime.
- Reducing screentime can be a poor proxy for users’ goals, which may concern specific features or contexts rather than total app use.
- YouTube’s Notifications digest and Autoplay toggle provide more granular controls than general screentime tools.
- Sense of agency concerns experiencing oneself as the initiator of actions, including judgments that attribute an action to the self or another.
- Internal mechanisms redesign an app experience, whereas external mechanisms monitor or police multiple apps without substantially changing individual app experiences.
- Internal mechanisms may support more targeted interventions and align users’ interests rather than merely fighting distractions.
3 STUDY 1: SURVEY OF 120 YOUTUBE USERS
Study 1 examines how existing mechanisms in the YouTube mobile app support or undermine users’ sense of agency before Study 2 evaluates potential redesigns.
- Study 1 examines existing YouTube mechanisms’ effects on sense of agency as the first step before designing and evaluating changes in Study 2.
- The study addresses RQ1 by investigating users’ experiences in the current YouTube mobile app.
- The sequential design connects investigation of the current app to evaluation of potential mechanism changes.
3.1 Participants
The survey recruited 120 U.S. YouTube mobile-app users with substantial recent YouTube use, using Mechanical Turk and documented device and demographic characteristics. The sample was not representative of all user populations.
- Recruitment: Participants were recruited from U.S. Mechanical Turk workers who owned smartphones and had used YouTube for at least three hours during the prior week.
- Devices: The sample included 80 Android users and 40 iPhone users because the researchers found nearly identical app features across devices.
- Sample: 120 participants completed the survey after meeting inclusion criteria, although seven additional respondents were excluded for noncompletion.
- Scope: The sample oversampled men, Asians, and young people relative to 2019 U.S. Census estimates, limiting generalization to other populations.
- YouTube use: Participants reported a median of 101 daily YouTube minutes across devices and estimated that a median 50% occurred in the mobile app.
3.2 Procedure
The survey asked YouTube users which app features made them feel most or least in control of their time, using randomized question tables and six required responses per participant.
- Participants answered online survey questions about demographics, YouTube use, and mechanisms affecting their sense of control.
- The survey randomized the order of the “more in control” and “less in control” question tables.
- The survey used “in control” wording because prior HCI studies measured sense of agency with that phrasing and self-report scales.
- Each participant submitted 6 responses: 3 for feeling most in control and 3 for feeling least in control.
3.3 Coding reliability thematic analysis
Researchers used coding reliability thematic analysis to identify interface mechanisms in 720 responses, establish shared codes, and organize them into higher-level themes.
- Researchers iteratively coded 720 responses, treating each response as one combined answer to the Thing and Explain questions.
- They removed 112 responses without substantive content before analyzing the remaining material.
- Coding included mechanisms designers could directly change, such as recommendations, while excluding largely situational factors outside designers’ control.
- The codebook ultimately contained 21 design mechanisms, including autoplay and playlists, with three subcodes for frequently mentioned recommendations.
- Researchers mapped the mechanism codes to three higher-level themes and recorded when individual mechanisms had split effects.
3.4 Results and Analysis
The survey identified nine frequently mentioned YouTube mechanisms with contrasting effects on agency: recommendations, ads, and autoplay usually reduced control, while several user-directed features increased it.
- 3.4.1 Design Mechanisms.: 467 responses named a specific design mechanism, and nine mechanisms mentioned at least 15 times formed the main analysis.The nine mechanisms covered 392 of 467 mechanism-referencing responses.
- 3.4.1 Design Mechanisms.: Recommendations were the most frequently mentioned mechanism, accounting for 27% of all responses.
- 3.4.1 Design Mechanisms.: Recommendations, ads, and autoplay primarily made respondents feel less in control, whereas playlists, search, subscriptions, play controls, and watch history and stats primarily increased control.
- 3.4.2 The spectrum of consent.: Participants experienced agency when app actions matched active or ongoing consent, such as using play controls or subscribing to creators; autoplay acted without consent.
- 3.4.2 The spectrum of consent.: Unpredictable behavior or violated expectations intensified users’ reactions to ads and autoplay because it felt deceptive or misleading.
- 3.4.3 Planning ahead.: Playlists supported advance planning by limiting viewing to a defined list and providing a stopping point, unlike endless recommendation browsing.
- 3.4.3 Planning ahead.: Autoplay and recommendations pressured users toward unplanned extended viewing, while Watch Later reduced pressure to watch immediately.
- 3.4.3 Planning ahead.: Notifications had mixed effects: they could prompt unwanted viewing or help users plan by signaling new content and reducing random checking.
4 STUDY 2: CO-DESIGN WITH YOUTUBE USERS
Study 2 used co-design sessions to examine how redesigning YouTube’s internal mechanisms might affect agency, presenting users with participant-generated sketches and researcher-generated mockups.
- Study 2 approach: The study held 13 sessions with individual YouTube users featuring participant sketching and evaluation of researcher-generated redesigns.
- 4.1 Preparatory Design Work: Five authors brainstormed 67 ideas, produced 33 sketches, and scored them for expected agency impact, novelty, and implementation feasibility.Expected agency impact received twice the weight of the other criteria.
- 4.1 Preparatory Design Work: Researchers created mockups for the seven highest-scoring sketches, each with low-, medium-, and high-control versions.
- 4.1 Preparatory Design Work: The recommendations redesign varied homepage exposure from unlimited recommendations to three recommendations with “show more” or none.
- 4.1 Preparatory Design Work: The research scope was limited because specific design details, such as task-oriented versus leisure-oriented search examples, could produce different effects.
- 4.2 Participants: The 13 co-design participants were Seattle YouTube users aged 18–36 who reported a median of 52 minutes of daily use.
4.3 Procedures
The co-design sessions combined think-aloud demonstrations of current YouTube use with participant sketching and evaluation of redesigned app mechanisms.
- The median co-design session lasted 73 minutes, with sessions ranging from 57 to 105 minutes.
- Sessions began with participants explaining a typical YouTube engagement cycle and identifying mechanisms that made them feel most or least in control.
- Participants sketched changes over minimal home, search, and video-player mockups designed to elicit new ideas rather than evaluate the existing interface.
- Participants evaluated three randomly ordered versions of each redesigned mechanism without being told which represented low, medium, or high control.
4.4 Codebook Thematic Analysis
Researchers analyzed transcripts and sketches using codebook thematic analysis, supported by session debriefs, team review, and vivid exhibits used in analytical memos.
- The team used codebook thematic analysis, generating themes that were more interpretive than summary but less centered on researchers’ subject positions than reflexive thematic analysis.
- Figure 3 contrasts a more-control sketch using user-chosen recommendation topics with a less-control sketch that hides the search bar beneath recommendations.
- Codes covered features affecting control, design mechanisms for control, clear user expectations, and triggers to stop, with further subcodes.
- Researchers applied codes to all transcripts and sketches, reviewed the results, and extracted vivid exhibits to write analytical memos.
4.5 Results and Analysis
Participants’ preferred level of control depended on their intention: specific goals favored greater control, while nonspecific aims favored delegating control to YouTube. They also associated active, informed choices with greater agency.
- Participants generated two themes: greater control was preferred for intentional use, and active, informed choices were expected to increase control.
- Specific intentions call for more control.: Specific intentions, such as cooking a recipe or watching a targeted tutorial, led participants to prefer mechanisms that provided greater control.
- Specific intentions call for more control.: Participants wanted stronger controls so recommendations would not hijack a specific intention, especially when entertaining results could invite time-wasting clicks.
- Specific intentions call for more control.: For moderately specific topical or temporal intentions, participants wanted to retain some control while delegating some decisions to YouTube.
- Active and informed choices.: For nonspecific intentions such as unwinding or exploring, participants often preferred YouTube to lead, because greater control could require more time and effort.
- Active and informed choices.: Participants envisioned active choices ranging from selecting algorithmically personalized topics to generating the choice set through a direct search.
- Active and informed choices.: Participants described search as user-led choice and often foregrounded it, while recommendations could cover screens with infinite scrolling that encouraged passive rabbit holes.
5 OVERALL DISCUSSION
The studies show that YouTube mechanisms can either undermine or support agency, and that users’ preferred level of control depends on their viewing intention. The discussion therefore emphasizes context-sensitive recommendations, customizable control, and caution against assuming that more agency is always better.
- Recommendations, ads, and autoplay mostly reduced users’ control, while playlists, search, subscriptions, play controls, and watch history mostly increased it.
- Rethinking What ‘Relevance’ Means for Recommendations: Recommendations undermined agency through both irrelevance and excessive relevance, with highly appealing videos sometimes conflicting with users’ longer-term goals.
- Rethinking What ‘Relevance’ Means for Recommendations: Recommender systems should decide whether to show recommendations before optimizing which items to display, using factors such as time, screentime goals, or explicit preferences.
- Rethinking What ‘Relevance’ Means for Recommendations: Recommendation design should consider users’ explicit preferences and aspirations rather than relying only on behavioral traces and immediate click likelihood.
- Microplanning: Microplans could support agency by tying lightweight plans or playlists to a specific situation, such as using a guitar tutorial playlist in a studio.
- Different Levels of Control for Ritualized and Instrumental Use: The authors caution that increasing agency is not always desirable because participants sometimes preferred low-control features and mixed experiences with autoplay and recommendations.
- Different Levels of Control for Ritualized and Instrumental Use: Users preferred higher-control mechanisms for specific intentions but lower-control mechanisms for general intentions, suggesting that one fixed control level will not fit all use.
- Towards a Language of Attention Capture Dark Patterns: The paper proposes a shared language of attention capture dark patterns to help designers recognize interfaces that impose attentional harms.
6 CONCLUSION
The paper reframes digital wellbeing around increasing users’ sense of agency rather than only reducing screentime. It identifies YouTube mechanisms that shape agency and proposes microplanning and a vocabulary for attentional harms.
- The paper identifies YouTube mechanisms that participants reported as influencing their sense of agency and examines how they might be changed.
- The authors propose microplanning mechanisms that make lightweight plans to guide a single session of use.
- The paper proposes language that designers can use to recognize patterns imposing attentional harms on users.