Expert Counter-Strike teams made 34.63% more between-round statements than nonexpert teams and spoke with a 29.39% higher word count per second (Eldadi and Tenenbaum, Sport, Exercise, and Performance Psychology 2025). Coordination pays off on the scoreboard too: in 31,055 ranked League of Legends matches, teams scored as cooperative won 89.54% of the time, against 20.18% for non-cooperative teams (Caldeira et al., arXiv 2025). The team channel also carries risk, because Call of Duty players were two to over three times more likely to hear toxic language from a teammate than from an opponent (Morrier, Mahmassani and Alvarez, arXiv 2024). This roundup covers the private team channel in competitive play, not the open lobby talk measured in our proximity voice chat usage statistics. We aggregated data from the Journal of Sport and Exercise Psychology, ACM CSCW and CHI, Valve’s Major rulebook, Riot Games, Activision-linked research, and a dozen other peer-reviewed and preprint studies of recorded team comms.
TL;DR
- Expert CS:GO teams made 34.63% more between-round statements than nonexpert teams (Eldadi and Tenenbaum, SEPP 2025).
- League of Legends teams averaged 537.2 voice utterances per game, 45,660 in total over 84 games (Son et al., arXiv 2025).
- Counter-Strike tournament players made 16.2 callouts per minute; Halo 3 players made 25.5 (Tang et al., CSCW 2012).
- Cooperative League of Legends teams won 89.54% of ranked games vs 20.18% for non-cooperative teams (Caldeira et al., arXiv 2025).
- A CHI 2016 study of 84,489 players found a positive but diminishing link between pings and performance (Leavitt, Keegan and Clark, CHI 2016).
- Master-rank League players send 0.489 ‘On My Way’ pings per minute vs 0.267 across all ranks (League of Graphs, via Lee et al., CHI 2025).
- Teammates were two to over three times more likely than opponents to be the source of toxic voice language (Morrier, Mahmassani and Alvarez, arXiv 2024).
- A deliberately toxic teammate lowered team and individual performance in a 716-player experiment (Monge and O’Brien, Media Psychology 2022).
- Counter-Strike Major coaches may talk to players only in warmup, half-time and tactical timeouts (Valve, Major Supplemental Rulebook).
- Riot’s 2026 Coach Voice pilot allows up to 3 sessions of 45 seconds per set (Riot Games, via Sports Kyunghyang 2025).
1. Volume and Pace: How Much Competitive Teams Talk
Skilled teams do not go quiet under pressure; they talk more, faster and more evenly. In a matched sample of eight CS:GO matches, expert teams spoke 15,727 of the 22,490 words recorded during live play, or 69.9% (15,727 / 22,490), and they spread that talk more evenly across the five players, chaining factual-to-factual and action-to-action statements while nonexpert teams drifted between off-task remarks and emotional statements (Eldadi, Fitoussi and Tenenbaum, Journal of Sport and Exercise Psychology 2025). The gap persists between rounds, when teams regroup and plan the next round (Eldadi and Tenenbaum, Sport, Exercise, and Performance Psychology 2025).
Volume alone is a weak signal, though. The most talkative League of Legends team in one dataset produced about 1.9 times as many utterances per game as the quietest team (648.8 / 333.2), yet professional status did not guarantee a high count: one pro team averaged 384.7 utterances per game, below an amateur team at 505.8 (Son et al., arXiv 2025). Who speaks, and what follows, matters more than the raw count. Where that channel runs, inside the game or in a separate party app, is a separate question covered in our console party chat and cross-play voice statistics.
| Metric | Value | Source |
|---|---|---|
| Words spoken during live play, expert teams vs nonexpert teams (4,040 statements, 8 CS:GO matches) | 15,727 vs 6,763 (experts 69.9% of 22,490) | Eldadi, Fitoussi and Tenenbaum, Journal of Sport and Exercise Psychology 2025 |
| Statements between rounds, expert teams compared with nonexpert teams | 34.63% more statements by expert teams (about 1,052 statements, 8 matches) | Eldadi and Tenenbaum, Sport, Exercise, and Performance Psychology 2025 |
| Words per second between rounds, expert teams compared with nonexpert teams | 29.39% higher for expert teams | Eldadi and Tenenbaum, Sport, Exercise, and Performance Psychology 2025 |
| Voice utterances per League of Legends game (3 pro and 2 amateur teams, 84 games): average, and most vs least talkative team | 537.2 average (SD = 188.7); 648.8 vs 333.2 | Son et al., arXiv 2025 |
| Callouts per minute in the final stages of two local tournaments: Halo 3, Counter-Strike, Team Fortress 2 | 25.5, 16.2 and 10.5 phrases per minute | Tang et al., CSCW 2012 |
Context note: the callout rates come from about six hours of tournament audio from 49 players. Most recent available data: Tang et al., CSCW 2012. No newer study we found publishes per-minute callout rates across several shooters.
2. Inside the Callout: What Teams Actually Say
A callout is a short spoken report that tells teammates where something is or what is about to happen. The data show callouts are overwhelmingly about place: 69.5% of callouts about the speaker or the enemy team included a location (Tang et al., CSCW 2012). That is why every competitive map develops a shared vocabulary of named spots. An ethnographic study of two CS:GO teams of 17- and 18-year-old players, taught in Swedish and switching into Finnish, found their callouts were mostly English, known-in-common words, and that teams had about 15 seconds between rounds to agree on a plan (Rusk, Ståhl and Jusslin, Eludamos 2022). Location words exist partly because ears alone are unreliable: direction errors in headset audio are measured in our gaming headset HRTF localization statistics.
In MOBA comms the mix shifts toward information and instructions. Of the 45,458 League of Legends utterances that an automatic classifier assigned to a dialogue act (a label for what an utterance does, such as informing or instructing), 20,291 were Inform, or 44.6% (20,291 / 45,458) (Son et al., arXiv 2025). Coaches complain about the rest of the stream: in one example game from a study co-written by a GiantX assistant coach, 39.1% of the 23 sentences spoken by one player used filler or off-topic phrasing (Vinot and Perez, arXiv 2024).
| Metric | Value | Source |
|---|---|---|
| Counter-Strike callouts in which the speaker reports on their own status or position | 38.7% | Tang et al., CSCW 2012 |
| Counter-Strike callouts that name a specific map location | 43.5% | Tang et al., CSCW 2012 |
| Length of most callouts (Halo 3, Team Fortress 2, Counter-Strike) | 8 words or fewer | Tang et al., CSCW 2012 |
| League of Legends utterances by dialogue act: Inform, Directive (instruction), Question, Commissive (commitment to act) | 20,291 (44.6% of 45,458 classified), 10,060, 9,626 and 5,481 | Son et al., arXiv 2025 |
| Time available between CS:GO rounds to negotiate the next plan | About 15 seconds | Rusk, Ståhl and Jusslin, Eludamos 2022 |
| Share of one player’s sentences flagged as filler or off-topic (‘parasite’ communication) in a coach-annotated example game | 39.1% of 23 sentences | Vinot and Perez, arXiv 2024 |
Outlier note: the 39.1% figure describes a single player in a single example and illustrates the coaching problem; it is not a rate for pro play. The callout-content shares come from the same 2012 recordings as the rates in section 1, and the CS:GO ethnography was recorded between May 2017 and May 2018.
3. Comms Patterns and Winning: Coordination vs Results
Newer research moves past word counts to sequences: what kind of utterance follows what. Professional League of Legends teams showed denser Inform-to-Directive sequences than amateur teams, 0.122 vs 0.109 (p = .001), meaning shared information turned into instructions more often (Son et al., arXiv 2025). The CS:GO sequences described in section 1 point the same way: expert talk links one task statement to the next, while nonexpert talk loops through off-task and emotional remarks.
Coordination also shows up in data with no audio at all. Across 31,055 ranked League of Legends matches and 164,571 players, teams scored as cooperative from assists, objective help and map pressure won far more often, and team-level features predicted wins much better than individual skill features (Caldeira et al., arXiv 2025). An earlier CSCW study reached a related conclusion: in League of Legends, tacit coordination played a larger role than verbal communication (Kim et al., CSCW 2017). In ad hoc teams, a study of 48 teams (135 players) found a positive relationship between communication word frequency and team cohesion (Tan et al., ACM CHI PLAY 2022).
| Metric | Value | Source |
|---|---|---|
| Inform-to-Directive sequence density, professional vs amateur League of Legends teams | 0.122 vs 0.109 (p = .001) | Son et al., arXiv 2025 |
| Question-to-Inform sequence density during key moments (kills, elite monsters, buildings) vs the rest of the game | 0.145 vs 0.131 (p < .001) | Son et al., arXiv 2025 |
| Ranked win rate, cooperative vs non-cooperative League of Legends teams (measured from in-game actions, not voice) | 89.54% vs 20.18% | Caldeira et al., arXiv 2025 |
| Win rate of mostly low-to-medium-skill cooperative teams against mostly high-skill non-cooperative teams | 81.46% (8,503 of 10,438) | Caldeira et al., arXiv 2025 |
| Win-prediction accuracy, team-level features vs individual-level features (logistic regression) | 0.8734 vs 0.7036 | Caldeira et al., arXiv 2025 |
| Ad hoc League of Legends teams in a communication and cohesion study | 48 teams, 135 players; word frequency positively related to cohesion | Tan et al., ACM CHI PLAY 2022 |
Context note: the Caldeira authors warn that matchmaking and the zero-sum nature of wins can inflate the cooperative win rates. Most recent available data for the cohesion study: Tan et al., ACM CHI PLAY 2022.
4. Pings and Silent Coordination
A ping is a one-click map marker with a sound cue that tells teammates “enemy missing” or “on my way” without a word. A foundational ping study covered 84,489 players across 10,293 League of Legends matches and found a positive but diminishing relationship between ping volume and individual performance (Leavitt, Keegan and Clark, CHI 2016). More pings help up to a point, after which they become noise.
Higher-ranked players ping more, which suggests the tool grows with skill rather than replacing voice. Interviews with 22 League of Legends players about 22 analyzed games produced a blunt title, “Less Talk, More Trust”: players tended to see communication, regardless of its content, as a precursor to team breakdowns (Lee et al., CHI 2025). Ping habits differ by player, too. A study of Heroes of the Storm found players use pings in distinct styles that have different effects on team performance (Zheng, Stein and Farzan, ACM Transactions on Social Computing 2022), and a Dota 2 survey identified six distinct motivations for using pings and map annotations (Wuertz, Bateman and Tang, CHI 2017).
| Metric | Value | Source |
|---|---|---|
| League of Legends players and matches analyzed for ping use | 84,489 players, 10,293 matches | Leavitt, Keegan and Clark, CHI 2016 |
| Relationship between number of pings and individual performance | Positive but diminishing (concave) | Leavitt, Keegan and Clark, CHI 2016 |
| ’On My Way’ pings per minute, all ranks vs Master rank and above | 0.267 vs 0.489 | League of Graphs 2024, via Lee et al., CHI 2025 |
| ’Enemy Missing’ pings per minute, all ranks vs Master rank and above | 0.164 vs 0.245 | League of Graphs 2024, via Lee et al., CHI 2025 |
| Games analyzed with interviewed League of Legends players | 22 games (each at least 24 minutes), 22 players | Lee et al., CHI 2025 |
| In-game objective votes in those 22 games that drew more than three votes | None | Lee et al., CHI 2025 |
| Distinct motivations for using pings and annotations among Dota 2 players | 6 | Wuertz, Bateman and Tang, CHI 2017 |
Most recent available data for the ping-performance link: Leavitt, Keegan and Clark, CHI 2016; for ping motivations: Wuertz, Bateman and Tang, CHI 2017. The per-minute rates come from League of Graphs, a third-party statistics site built on match data, as cited by the CHI 2025 authors.
5. Tone Under Pressure: Supportive Talk and Toxic Talk
The team channel is where tone spreads fastest, in both directions. Players were two to over three times more likely to hear toxic language from a teammate than from an opponent, even though exposure of any kind occurred in under one-tenth of one percent of cases, in an analysis of 56,464,489 player-match observations from Call of Duty: Modern Warfare III Team Deathmatch voice chat (Morrier, Mahmassani and Alvarez, arXiv 2024). The same study found that exposure significantly raised the probability that players used similar language, with the effect varying by whether the source was a teammate from the same party (players who queued together) or a different one.
The causal link to results is clearest in an experiment where a confederate teammate (a player working for the researchers) either played neutrally or berated participants: toxic behavior lowered both team and individual performance across 716 League of Legends players (Monge and O’Brien, Media Psychology 2022). The positive mirror image comes from a lab study: novice teams in a supportive climate produced about twice as many emotional and motivational statements and completed more orders on the easy level of a cooperative cooking game (Eldadi, Dekimhi and Tenenbaum, arXiv 2025). A player-perspective taxonomy of League of Legends identified five main types of toxic behavior and five contextual factors that trigger them (Kou, CHI PLAY 2020).
| Metric | Value | Source |
|---|---|---|
| Probability that players were exposed to toxic voice language from opponents or teammates (Call of Duty: Modern Warfare III) | Under one-tenth of one percent | Morrier, Mahmassani and Alvarez, arXiv 2024 |
| Likelihood of hearing toxic language from teammates compared with from opponents | Two to over three times more likely from teammates | Morrier, Mahmassani and Alvarez, arXiv 2024 |
| Estimated effect of exposure to toxic language on the time before a player’s next match | 16.18 to 60.68 hours | Morrier, Mahmassani and Alvarez, arXiv 2024 |
| Players in an experiment with a deliberately toxic teammate | 716; toxicity lowered team and individual performance | Monge and O’Brien, Media Psychology 2022 |
| Emotional or motivational statements per team, supportive climate vs neutral climate (Overcooked! 2) | 15.50 vs 7.63 (d = 1.94) | Eldadi, Dekimhi and Tenenbaum, arXiv 2025 |
| Orders completed on the easy level, supportive climate vs neutral climate | 27.40 vs 22.80 (d = 1.11) | Eldadi, Dekimhi and Tenenbaum, arXiv 2025 |
Context note: the supportive-climate study used 40 novices in 20 teams of three, each including a trained confederate, so its statement counts partly reflect the manipulation itself. On the hard level, the score gap disappeared (222.10 vs 231.70 points, d = -0.07). Most recent available data for the toxic-teammate experiment: Monge and O’Brien, published online in 2021.
6. Rules, Coaches and Moderation of the Team Channel
At the top level, who may speak on the team channel is a rules question. At a Counter-Strike Major, the single registered coach may talk to players only during warmup, half-time and tactical timeouts, and no staff may communicate with players during online matches (Valve, Counter-Strike Major Supplemental Rulebook). Riot’s Coach Voice pilot in the LCK and LCP opened short live windows during the 2026 LCK Cup group stage, without pausing the match and with coaches seeing only their team’s vision (Riot Games, via Sports Kyunghyang, December 3, 2025).
In ranked play, the team channel is now recorded and reviewed. Riot keeps VALORANT voice audio for a minimum of 24 hours regardless of reports and escalates comms restrictions through fixed tiers (Riot Games Support, Comms Restrictions and Audio Retention 2026). The most recent monthly enforcement count we found is much older: over 400,000 text and voice chat restrictions in January 2022 (Riot Games, VALORANT Systems Health Series 2022). Automated review of this kind starts by cleaning and transcribing audio, the same steps benchmarked in our neural noise suppression and voice isolation statistics.
| Metric | Value | Source |
|---|---|---|
| People allowed in the competition area at a Counter-Strike Major | 5 players and 1 registered coach | Valve, Counter-Strike Major Supplemental Rulebook |
| Moments a Counter-Strike Major coach may talk to players | Warmup, half-time and tactical timeouts only | Valve, Counter-Strike Major Supplemental Rulebook |
| Coach Voice sessions allowed per set in the LCK and LCP pilot | Up to 3, each 45 seconds, used by up to 2 registered staff | Riot Games, via Sports Kyunghyang 2025 |
| VALORANT voice audio retention: no report, after a report, after a penalty | 24 hours, 7 days, 2 years | Riot Games Support 2026 |
| VALORANT comms restriction lengths | 1, 3, 7, 14 and 365 days | Riot Games Support 2026 |
| VALORANT text and voice chat restrictions issued in January 2022 | Over 400,000 (plus over 40,000 bans) | Riot Games, VALORANT Systems Health Series 2022 |
Most recent available data for enforcement volume: Riot Games, 2022. Riot’s voice evaluation started in North America in English only, with a beta from February 2023 (Riot Games, VALORANT Voice Evaluation Update).
Summary: Esports Team Voice Comms and Callouts by the Numbers
| Metric | Value | Source |
|---|---|---|
| Between-round statements, expert vs nonexpert CS:GO teams | 34.63% more by expert teams | Eldadi and Tenenbaum, SEPP 2025 |
| Live-play words spoken by expert teams in a matched sample | 15,727 of 22,490 (69.9%) | Eldadi, Fitoussi and Tenenbaum, JSEP 2025 |
| Voice utterances per League of Legends game | 537.2 average (45,660 total, 84 games) | Son et al., arXiv 2025 |
| Counter-Strike callouts per minute in tournament play | 16.2 | Tang et al., CSCW 2012 |
| Counter-Strike callouts naming a specific map location | 43.5% | Tang et al., CSCW 2012 |
| Inform-to-Directive sequence density, pro vs amateur teams | 0.122 vs 0.109 (p = .001) | Son et al., arXiv 2025 |
| Ranked win rate, cooperative vs non-cooperative teams | 89.54% vs 20.18% | Caldeira et al., arXiv 2025 |
| Win-prediction accuracy, team vs individual features | 0.8734 vs 0.7036 | Caldeira et al., arXiv 2025 |
| Players and matches in the CHI 2016 ping study | 84,489 players, 10,293 matches | Leavitt, Keegan and Clark, CHI 2016 |
| ’On My Way’ pings per minute, all ranks vs Master and above | 0.267 vs 0.489 | League of Graphs, via Lee et al., CHI 2025 |
| Toxic language heard from teammates vs from opponents | Two to over three times more likely from teammates | Morrier, Mahmassani and Alvarez, arXiv 2024 |
| Time available between CS:GO rounds to negotiate the next plan | About 15 seconds | Rusk, Ståhl and Jusslin, Eludamos 2022 |
| Players in the toxic-teammate experiment | 716 | Monge and O’Brien, Media Psychology 2022 |
| When Counter-Strike Major coaches may talk to players | Warmup, half-time and tactical timeouts only | Valve, Major Supplemental Rulebook |
| Coach Voice sessions per set, LCK and LCP pilot | Up to 3 of 45 seconds | Riot Games, via Sports Kyunghyang 2025 |
| VALORANT text and voice chat restrictions, January 2022 | Over 400,000 | Riot Games, VALORANT Systems Health Series 2022 |
Methodology and Sources
- Eldadi, Fitoussi and Tenenbaum: Verbal Communication, Coordinated Effort, and Performance in Esports Teams, Journal of Sport and Exercise Psychology 47(5), 2025
- Eldadi and Tenenbaum: Between-Rounds Verbal Communication in Expert and Nonexpert Esports Team Players, Sport, Exercise, and Performance Psychology 2025
- Eldadi, Dekimhi and Tenenbaum: Motivational Climate Effects on Communications, Emotional-Social States, and Performance in Collaborative Gaming Environment, arXiv 2025
- Son, Jang, An, Kang and Park: /UnmuteAll: Modeling Verbal Communication Patterns of Collaborative Contexts in MOBA Games, arXiv, December 23, 2025
- Tang, Massey, Wong, Reilly and Edwards: Verbal Coordination in First Person Shooter Games, CSCW 2012
- Rusk, Ståhl and Jusslin: Understanding Esports Teamplay as an Emergent Choreography, Eludamos 13(1), 2022
- Vinot and Perez: Voice Communication Analysis in Esports, arXiv 2024
- Caldeira, Maharjan, Majumdar and Pournaras: Collective Intelligence Outperforms Individual Talent: A Case Study in League of Legends, arXiv 2025
- Kim, Engel, Woolley, Lin, McArthur and Malone: What Makes a Strong Team? Using Collective Intelligence to Predict Team Performance in League of Legends, CSCW 2017
- Tan, Rogers, Nacke, Drachen and Wade: Communication Sequences Indicate Team Cohesion, Proceedings of the ACM on HCI (CHI PLAY) 2022
- Leavitt, Keegan and Clark: Ping to Win? Non-Verbal Communication and Team Performance in Competitive Online Multiplayer Games, CHI 2016
- Lee, Kim, Park, Kim, Jang and Seering: Less Talk, More Trust: Understanding Players’ In-game Assessment of Communication Processes in League of Legends, CHI 2025
- Zheng, Stein and Farzan: Use Ping Wisely: A Study of Team Communication and Performance under Lean Affordance, ACM Transactions on Social Computing 2022
- Wuertz, Bateman and Tang: Why Players use Pings and Annotations in Dota 2, CHI 2017
- Morrier, Mahmassani and Alvarez: Uncovering the Effect of Toxicity on Player Engagement and its Propagation in Competitive Online Video Games, arXiv 2024
- Monge and O’Brien: Effects of Individual Toxic Behavior on Team Performance in League of Legends, Media Psychology 25(1), 2022
- Kou: Toxic Behaviors in Team-Based Competitive Gaming: The Case of League of Legends, CHI PLAY 2020
- Valve: Counter-Strike Major Supplemental Rulebook
- Riot Games, reported by Sports Kyunghyang: LCK introduces Coach Voice, December 3, 2025
- Riot Games Support: Comms Restrictions and Audio Retention, 2026
- Riot Games: VALORANT Systems Health Series: Voice and Chat Toxicity, February 10, 2022 and VALORANT Voice Evaluation Update
- Data watch: recorded team comms are rarely public, so the core studies use small samples (eight matches in each Eldadi CS:GO study, five teams in Son et al.) and their findings are correlational. The Son et al. classifier counts sum to 45,458, not the 45,660 utterances reported overall; our 44.6% Inform share uses the classified total. Morrier et al. estimate that toxic language from a same-party teammate raises the probability of a player using toxic language by 59.93 to 69.48 percentage points (1.63 to 2.08 points for a different-party teammate), but these instrumental-variables estimates apply to a subset of players and the paper does not reconcile them with its own observed baseline incidence of 0.083%, so we report them here only and not as headline figures. Caldeira et al. measure cooperation from in-game actions, not speech, and note that matchmaking may inflate win rates. Coverage of the Coach Voice pilot conflicts on the unit: Korean outlets report 3 sessions per set, while at least one English-language outlet wrote per game. The between-rounds Eldadi paper was listed as in press in 2025, and Son et al., Caldeira et al. and Eldadi, Dekimhi and Tenenbaum are preprints whose figures may change in peer review. We found no Riot monthly VALORANT enforcement count newer than January 2022. The callout-content shares rely on 2012 audio and remain the most recent available benchmark.
Last updated: November 6, 2026. We update this roundup quarterly, and the next refresh is expected when Riot publishes its review of the 2026 Coach Voice pilot and the Son et al. /UnmuteAll study appears in peer-reviewed form.