Source-linked AI summary
Moderation Challenges in Voice-based Online Communities on Discord
Jialun Aaron Jiang, Charles Kiene, Skyler Middler, Jed R. Brubaker, Casey Fiesler
TL;DR
Moderating real-time voice is difficult because content is ephemeral and lacks the persistent record available in text-based communities. Through interviews with Discord moderators, the paper finds that proving violations is a central challenge and discusses tools that support moderators while accounting for voice-specific constraints.
Problem
Real-time voice lacks a persistent written record, complicating how moderators locate, remove, and verify rule-breaking and identify its speaker.
Source: §1 · p. 2Method
The study analyzes 25 interviews with Discord moderators about voice-specific rules, moderation practices, and evidence-gathering challenges.
Source: §1 · p. 2Results
Moderators often struggled to prove violations and could act post hoc; voice-only recording did not let them link members’ unique voices to their identities.
Source: §6 · p. 17, §7 · p. 20Takeaways & Limitations
Voice moderation tools can automate work while keeping humans in charge and should account for infrastructure limits before importing text-based practices.
Source: §1 · p. 2, §6.2 · p. 19Blanket bans of content automated systems deem inappropriate may harm communities, especially where voice-channel rules and norms are evolving or unclear.
Source: §6.2 · p. 19
Abstract
from arXiv · showhide
Online community moderators are on the front lines of combating problems like hate speech and harassment, but new modes of interaction can introduce unexpected challenges. In this paper, we consider moderation practices and challenges in the context of real-time, voice-based communication through 25 in-depth interviews with moderators on Discord. Our findings suggest that the affordances of voice-based online communities change what it means to moderate content and interactions. Not only are there new ways to break rules that moderators of text-based communities find unfamiliar, such as disruptive noise and voice raiding, but acquiring evidence of rule-breaking behaviors is also more difficult due to the ephemerality of real-time voice. While moderators have developed new moderation strategies, these strategies are limited and often based on hearsay and first impressions, resulting in problems ranging from unsuccessful moderation to false accusations. Based on these findings, we discuss how voice communication complicates current understandings and assumptions about moderation, and outline ways that platform designers and administrators can design technology to facilitate moderation.
1 Introduction
Voice communication changes moderation by making rule-breaking harder to identify and prove without a persistent record. Interviews with Discord moderators reveal new challenges and constrained strategies, motivating tools that support human moderators.
1 Introduction
- Real-time voice lacks a persistent written record, complicating how moderators locate content, remove it, identify speakers, and determine whether a violation occurred.The authors connect these issues to other ephemeral settings, including immersive games and virtual reality.Source: §1 · p. 2
- Voice and audio-based communities introduce unfamiliar rule-breaking, including disruptive audio, while moderators develop tactics and workarounds for these problems.Discord users also use the platform to play audio such as music or noise.Source: §1 · p. 2
- Interviews reveal that moderators struggled to gather evidence of rule-breaking, sometimes relying on hearsay and first impressions when deciding how to act.The authors also report that evidence-gathering strategies could fail in different scenarios.Source: §1 · p. 2
- The authors argue that voice precludes commonplace text-moderation tools and recommend moderator tools that automate work while leaving humans in charge.They also caution against importing existing rules and strategies without considering technological infrastructure.Source: §1 · p. 2
2 Related Work
Prior work frames moderation as a mix of techniques and ongoing challenges, while research on new technologies, voice, and ephemerality helps explain why voice chat poses distinct moderation problems.
2 Related Work
- The review links moderation research, voice as a newly adopted community technology, and ephemerality, which the authors identify as the biggest challenge for voice moderation.It uses these research threads to situate how voice can introduce or exacerbate moderation problems.Source: §2 · p. 3
2.1 Moderation and Its Challenges
- Moderation includes excluding, pricing, organizing, and norm-setting, but volunteer labor, imperfect tools, and evolving rules complicate enforcement.Prior work also identifies difficulties adapting automated tools and scaling nuanced human decisions.Source: §2.1 · p. 3
2.2 New Technologies and New Interactions
- New technologies can create unexpected behavioral-regulation challenges as users find ways to break rules and push against community norms.The review compares voice moderation with earlier problems arising from novel uses of technology.Source: §2.2 · p. 4
2.2.1 Voice Communication.
- Voice-only communication can intensify speaker-identification and coordination problems, while prior work also describes voice abuse and ambient noise.Recorded voice snippets in rural Indian communities permit moderation actions such as deleting, categorizing, and ranking, unlike real-time ephemeral voice.Source: §2.2.1 · p. 4
2.2.2 Ephemerality.
- Prior work associates ephemerality with spontaneous interaction and experimentation, but persistent records also matter to moderators as evidence of rule-breaking.The paper examines evidence gathering as a central challenge in ephemeral voice.Source: §2.2.2 · pp. 4–5
3 Research Site: Discord
Discord organizes communities into servers with separate text and voice channels, while its real-time voice features and moderation permissions shape what moderators can do.
3 Research Site: Discord
- Discord servers range from small friend groups to communities with hundreds of thousands of members, and the platform had over 200 million unique users as of December 2018.Source: §3 · p. 5
- Servers contain separate text or voice channels; the interface places server and channel lists beside text chat and a role-categorized user list.Source: §3 · p. 5, Figure 1 · p. 6
- Voice chat is real-time and not recordable or storable through Discord, and users can mute or deafen themselves; some channels use music bots and playlists.Source: §3 · p. 5
- Role permissions let moderators mute, deafen, remove, or ban users, while third-party bots can add temporary actions and automate text moderation; the paper reports no voice-moderation bots.Source: §3 · pp. 5–6
4 Method
The study analyzes 25 semi-structured interviews with Discord moderators, recruited from open servers and examined through iterative thematic analysis.
4 Method
- The researchers analyzed 25 interviews with volunteer moderators from 16 Discord servers, recruited through outreach and snowball sampling.Most studied servers had more than one thousand members, which the authors note may not represent smaller groups.Source: §4 · p. 6
- Interviews lasted 42–97 minutes and covered moderation stories, discovery of rule-breaking, actions taken, team deliberations, hypothetical scenarios, and participants’ reflections.Participants were 18–39 years old (M = 24, SD = 5.43).Source: §4 · pp. 6–7
- The team used inductive open coding followed by iterative coding and discussion to develop higher-order categories and clarify the final themes.The first author produced descriptive theme memos grounded in interview data, which all authors discussed.Source: §4 · pp. 7–8
5 Findings
The findings characterize voice-specific rule violations, moderators’ responses, and the difficulty of gathering evidence for enforcement.
5 Findings
- The findings cover new voice-specific rule violations, moderators’ responses, and evidence-gathering strategies that often fail.The section proceeds from rule types and comparisons with text communication to enforcement actions and evidence challenges.Source: §5 · p. 8
5.1 Rules in Voice and How People Break Them
The study examines actively enforced explicit and implicit rules, focusing on ways voice changes rule-breaking and moderation. Slurs and hate speech were explicit rules across the studied servers, while disruptive noise, music queue disruptions, and raids involved voice-specific implicit rules.
5.1.1 Explicit Rules.
- All 16 discussed servers prohibited slurs and hate speech, but moderators differed on whether context and intonation changed enforcement.Some treated slur variants uniformly, while others made case-by-case judgments, increasing the role of personal judgment.Source: §5.1.1 · pp. 8–9
5.1.2 Implicit Rules.
- Moderators also enforced unwritten voice-specific rules against disruptive noise, music queue disruptions, and raids.These behaviors were complex and nuanced enough that moderators did not always articulate them as explicit rules.Source: §5.1.2 · p. 9
5.1.2.1 Disruptive Noise.
- Disruptive noise can monopolize voice conversation, and moderators distinguish intentional noise from loudness caused by sensitive microphones.Because voice has limited channel capacity, loud speech can silence others; moderators may warn users about sensitive microphones before acting and consider how they join a channel.Source: §5.1.2.1 · pp. 9–10
5.1.2.2 Music Queue Disruption.
- Music queue disruptions include loud or lengthy tracks and skipping others’ selections, violations that deny members their turns.Queues are sequential, so playing an excessively long track or skipping another person’s music disrupts turn-taking.Source: §5.1.2.2 · p. 11
5.1.2.3 Raids.
- Raids involve coordinated voice-channel disruption, and moderators may have to handle accounts individually while risking false positives.Interviews included raids of up to thousands of bot accounts; similar names or profile pictures could prompt punishment without confirmation, including for community in-jokes.Source: §5.1.2.3 · pp. 11–12
5.2 Moderation Practices
Voice moderation relies on context-sensitive judgment because moderators cannot inspect a persistent record, so they often warn first, sometimes act on hearsay or first impressions, and use broad authority rules.
5.2.1 Warn Before Punishment.
- 23 of 25 moderators said they usually warned before imposing real punishment, using warnings to enforce rules without risking a mistaken punishment when voice violations could not be unambiguously verified.Warnings could also let moderators learn the context of alleged violations while avoiding upsetting community members.Source: §5.2.1 · p. 12, §5.2 · p. 12
- Public warnings also served as examples of unacceptable behavior, especially because voice rules may be ambiguous or unfamiliar to newcomers accustomed to text communities.Moderators described other community members learning from public warnings.Source: §5.2.1 · pp. 12–13
5.2.2 Punishment Based on Hearsay and First Impressions.
- Despite the usual warning-first approach, moderators sometimes punished users based on a member report or first impressions, leaving room for false accusations and mistaken bans.One team banned a user based on a report without hearing the alleged conduct; another moderator described immediately banning newcomers whose behavior suggested they had no interest in the rules.Source: §5.2.2 · p. 13
5.2.3 Catch-all Rules.
- Only two examined servers had voice-specific behavior guidelines; because rules could not cover every behavior and violations involved nuance, servers relied on catch-all rules granting moderators broad authority.The umbrella rules reduced the work of creating rules and arguing with members, while placing greater responsibility on moderators to decide correctly.Source: §5.2.3 · pp. 13–14
5.3 Acquiring Evidence
Because real-time voice leaves no persistent record, moderators used presence, witness reports, and recordings to gather evidence, but each approach had practical or ethical limits.
5.3 Acquiring Evidence
- Ephemeral voice makes evidence of rule-breaking a major challenge: unlike text, voice channels have no persistent record, and the available workarounds remain imperfect.The workarounds could contribute to mistaken punishment or potentially breaking the law.Source: §5.3 · p. 14, §5.3.3 · p. 17
5.3.1 Entering Voice Channels to Confirm Rule Breaking.
- Joining a channel helps only if a moderator is available while the violation is ongoing; this approach is nearly impossible for brief violations and crowded conversations can obscure who spoke.When a moderator arrives too late, the speaker may leave, erasing the connection between the heard content and the person to punish.Source: §5.3.1 · pp. 14–15
5.3.2 Relying on Witness Reports.
- Witness reports can substitute for missed events, and multiple consistent accounts may seem more credible, but moderators cannot verify their truth and investigating inconsistent accounts can consume disproportionate effort.False reports or coordinated brigading may also leave accused members without concrete evidence to appeal against.Source: §5.3.2 · pp. 15–16
- Some moderators sent trusted members or undercover moderators into voice channels as “spies,” but concealed identities can undermine members’ expectations of privacy and discourage voice participation.Members may not know who they are speaking to or who can hear them.Source: §5.3.2 · p. 16
5.3.3 Recording.
- Recording can make ephemeral voice persistent, but screen capture is needed to connect speakers to identities; it still cannot resolve overlapping speech and creates larger files.Some moderators used concealed recordings with non-moderator accounts when users stopped speaking after a moderator joined.Source: §5.3.3 · pp. 16–17
- Recording also raises jurisdiction-dependent legal concerns: laws differ, and Discord and the examined servers had no consent process for voice-channel recording.The paper notes that two-party consent is required in some jurisdictions.Source: §5.3.3 · p. 17
6 Discussion
Voice’s real-time ephemerality removes familiar moderation tools and makes violations harder to verify. The discussion considers design responses and argues that moderation must account for each technology’s infrastructure.
6.1 The Unique Challenges of Voice
- Ephemeral voice eliminates deleting, editing, or annotating content, leaving open communities to exclude rule breakers or set norms; exclusion targets a person’s presence.In voice, muting or banning is the only way to prevent problematic content from appearing because it depends on the speaker’s presence.Source: §6.1 · pp. 17–18
- Direct rules struggle to cover nuanced voice violations, while public example-setting takes time; moderators also lack infrastructure-level tools to prevent violations before they occur.Voice has no equivalent to keyword filtering, so moderators can only react to violations.Source: §6.1 · p. 18
- Ephemeral voice makes evidence difficult to obtain; reporting-based channel entry and recording are unreliable and may risk breaking the law.Moderators may be unable to prove a violation and can only act after it occurs.Source: §6 · p. 17, §6.1 · p. 18
6.2 Designing for Voice Moderation
- Automated detection could address some violations, but nuanced context and speech differences may challenge systems, while blanket bans could unpredictably shift community norms.The paper notes potential unfair punishment of speakers whose English differs from the speech-recognition systems’ training.Source: §6.2 · pp. 18–19
- Designers could detect disruptive volume and lengthy audio, and platforms could provide recordings with metadata identifying who spoke and when.The paper suggests prompting loud users to check hardware and letting moderators judge whether lengthy audio violates rules.Source: §6.2 · p. 19
- Always-on listening and recording raise privacy and legal concerns; the paper proposes platform-level consent or advice for moderators to obtain consent in their communities.The discussion notes that one-party-consent recording is illegal in eleven U.S. states and some other countries.Source: §6.2 · pp. 19–20
6.3 Beyond Voice
- The identified problems may also arise in other real-time, unrecorded spaces such as social VR, but the study’s specific issues may not generalize beyond voice.VR adds virtual physical presence as another moderation complication.Source: §6.3 · p. 20
- Because moderation strategies can break down across technologies, moderators should revisit implicit rules as communities change and can share solutions across communities.The interviews did not reveal cross-community collaboration, though the authors suggest it as a way to address new-technology problems.Source: §6.3 · p. 20
7 Conclusion
The study finds that voice-based communities challenge familiar moderation practices: enforcing rules and proving violations are both difficult. It argues that platform technology and policies should support moderators while protecting them from technology-related harm.
7 Conclusion
- Moderators struggled both to enforce rules in voice communities and to prove that violations had occurred.Voice-only recordings were insufficient because moderators could not link distinctive voices to member identities.Source: §7 · p. 20
- Vague platform policies on recording may expose volunteer moderators to legal ramifications.Source: §7 · p. 20
- The authors argue that platform designers and administrators should develop technology and policies that support moderators and protect them from unexpected technology-related harm.Source: §7 · p. 20