Maximum Views: Analyzing Twitch role-play live stream content through language

On April 15th, 69 days after the launch of the latest iteration of NoPixel, a massive video game server, the popular subreddit Livestreamfail banned all Grand Theft Auto V role play (GTAV RP) clip submissions for two weeks. 

This was quite literally a controversial decision, given that the announcement received a near equal amount of upvotes and downvotes. With 1.2 million members, the Livestreamfail subreddit is one of the most popular Reddit forums for users to share live stream content, usually from Twitch. Grand Theft Auto V, a best-selling video game from 2013, is currently the most viewed game on Twitch. It tops all other categories. This is likely due to the popularity of the NoPixel server, which is heavily populated by streamers.

The moderators said that the subreddit was too overwhelmed with GTAV RP content and that it would be best to make more room to diversify the posts.  

Ten days after Livestreamfail’s announcement, Streamer Hasan Piker (whom I featured in my last paper) declared that he was taking a break from GTAV RP. 

For the time being, he would not log into the NoPixel server to role play as Humberto Antonio Donato Pecorino. In that stream, he said he wasn’t feeling like he “wasn’t offering the best kind of content.” In a later stream, Hasan explained that GTAV RP was taking over his life to the point where he was addicted.

Since I finished my last paper on GTAV RP and continued to consume this content, I thought I was just now observing the crumbling of the GTAV RP content industrial complex. Perhaps the market of live streaming content was oversaturated by a single category that served a specific purpose. Hasan became overwhelmed by the task of creating compelling content, and the subreddit moderators had made a deliberate effort to intervene in the monopolization of GTAV RP clips. This process seems to be inevitable in the practice of “content aggregation.” 

In his article, “Giant Pools of Content: Theorizing Aggregation in Online Media Distribution,” author Andrew J. Bottomley defines content aggregation as “the practice of pulling media content from various sources and making it accessible in one dedicated, easy-to-find location.” Bottomley makes a historical analysis of content aggregation, dating this practice as early as the 1960s, and taking on a new form in the Dot Com Boom era of the 1990s. Through his analysis, Bottomley found that, because of it falling under a capitalist model encouraging growth, internet content became so abundant that its access “invariably [brought] with it a problem of overabundance that must be addressed through the methods of sorting and selection,” i.e., content aggregation sites.

Predating Twitch, there was a rise of human-aggregated websites like Broadcast.com, founded in 1995 as AudioNet. Broadcast.com’s purpose was to simulcast content like radio and television stations and provide on-demand content from audiobooks to music albums. Most importantly, this was offered at no cost. However, this was not a new concept, Bottomley explains. Instead, it was “modeled explicitly on broadcasting precedents and imposed those mass media logics on the internet media system.” 

Broadcast.com is long gone, but the practice of content aggregation and distribution is not. Bottomley cited Youtube, Netflix, and Spotify as “content distributors that operate under basically the same intermediation principle of pulling together a vast array of content at a single site from which individuals can then sort and filter the programming according to their particular needs…” Bottomley does not cite Twitch as one of those sites. However, it falls under the definition that he provides. Twitch, much similar to Broadcast.com, sources streaming content from users for wide distribution. It plays a dual role as both a place for people to create live content and a place for on-demand content in the form of “clips.” The dominating live stream platform has inevitably disrupted the live streaming community through its content aggregation service. Bottomley raises that “the consolidation of content through aggregators raises concerns about concentration power and control.”

Through its profit-driven model, Twitch has spawned an overabundance of live stream content that users seek to control by parsing through it to curate capital C content. But the overabundance of content has led to an overtaking of all other kinds of live stream content. In my paper, I ask the following questions: Through a linguistic anthropology lens, when does role-play content become Content? Moreover, what are the consequences of its overabundance?

If you are speaking the language of the content consumer, the top entry in Urban Dictionary provides an apt definition of Content: 

It is worth noting the slang term “the shit” and the “maximum” descriptor for views. From an audience’s perspective, Content can be defined by its wide acceptance as something good, has the potential for virility, and irresistibly warrants engagement. Also, it can be media created for the intention to beat the algorithms of these aggregator sites and get a user’s content in front of as many eyes as possible.

From this acknowledgment that there is a self-aware practice of Content creation, plus the practice of content aggregation being effortless through major platforms, I infer that both are happening in the context of GTAV RP live stream content, and there may be significant consequences that obfuscate the idea of an open and democratic internet.

About NoPixel

NoPixel is a game server hosted by FiveM, a platform that modifies the original GTAV game to accommodate large multiplayer servers. In NoPixel, up to 200 players can be on the server at a time. The server environment is based on Los Santos, a fictional city where the original game takes place. All the content I have observed and collected comes from this server specifically, as it is the most popular among Twitch streamers. 

There is an application process to join the NoPixel server, with questions that ask the player to expand on their potential character and responses to hypothetical scenarios that could happen in-game. Currently, a player must donate monthly to be a part of the server. 

There are a set of rules that all players must abide by to avoid being banned. These rules have to do with role-play philosophy and general decency guidelines. The rules are an essential framework to analyze the data because they act as an authority on the ideology of participation in role-play. This informs what kind of Content can potentially be derived from these Twitch streams.

About the Data

The clipping feature on Twitch is an intuitive way for users to generate content suitable for curation and virility. In less than a few minutes, a user can parse through a small section of the stream, highlight the part that they want to clip, and publish the clip for other users to watch on-demand. 

In order to collect well-rounded data, I watched live streams (content I would consider organic content) and parsed through live stream clips already created by other users through Twitch (I would consider this curated content) with a preference for those with more views. I found that a combination of these methods helped trace the linear timeline of Content curation. By first exploring generated clips, I could infer what could be considered Content in my live observations. To help expand on those observations, I generated my own live stream clips to embed in this paper. These are indicated by my username (marsisms). 

Twitch limits the clipping feature so that the clips can be no longer than 60 seconds. I found myself watching a total of 8 hours of live streaming to capture these clips. 

For this paper, I will present just a small sample of the thousands of GTAV RP clips generated over time. 

Analysis

Rule-abiding role play as Content

As I mentioned in my last paper, NoPixel role-play has strict rules regarding in-character performance, resulting in unique strategies to help the role player maintain the suspension of disbelief for their audience:

“The success of NoPixel streaming on Twitch could be attributed to its seemingly rigid ideology of participation that actually facilitates an expansive role-playing experience, both for streamers and watchers. In the server, the gamer’s role is exclusively that of a NoPixel character. There can be no referencing the fact that you are experiencing this world on a computer. If you are not sure how to move your arm, you will be told to “flex your G muscle,” referring to the button on your keyboard. If characters observe that your mic isn’t working, you may be asked to take a cough drop. If a player cannot adhere to this rule, among others, they will be kicked.”

With this in mind, I wanted to present an instance where extreme in-character, immersive participation generated successful Content. Twitch streamer Sykkuno has joined the NoPixel as the character Yuno Sykk, a kind criminal who conceals his face out of insecurity. He has received the reputation as one of the friendliest people in Los Santos by cops and criminals alike.  

In this clip, generated by user Taycroofficial, Yuno is seen helping his friend Amon Gus (streamer DisguisedToast) get away from the police by directing him while he is driving. This was a somewhat popular clip, garnering 268 upvotes on Reddit and over 12,000 clip views. 

The motivation to choose this clip to exemplify rule-abiding as Content came from the title of the Reddit post: “Sykkuno and his squad get sucked up into a tornado while attempting getaway.” Given the weather mechanics of the server, I was curious about the actual existence of a tornado overpowering a car. After watching the clip, I realized what the title was really referencing and felt it was very relevant to analyze for this paper through linguistic anthropology terms. 

At 0:23, you see the car spontaneously spring in the air and convulse. While it is obviously a glitch outside of the players’ control. Yuno attempts to normalize the situation by frantically asking what Amon is doing to cause the car to move that way. Amon responds by saying he hit a speed bump. 

The following live stream recording shows the glitch happening from the perspective of officer Claire Everly, played by streamer LunaOni. She exits her vehicle to catch the attention of the players trapped in the car, telling them to jump out. 

The whole stream was watched by a total of over 249,000 viewers, which trumps all other LunaOni streams, which have 31,000 to 180,000 views. The players who have jumped out are effectively injured, validated by one player saying, “It hurts so bad,” at 8:59:53. To effectively code-switch in and out of character, LunaOni is laughing at the absurd situation while her in-game mic is off so that she can handle the situation seriously in her role. 

In this extended clip, you can also hear Yuno maintaining the normalization of the situation by indexing the glitch as a natural disaster. He says, “We got caught in a tornado or something.” Officer Claire does not question this. 

Right after, Claire’s cop car also glitches, jumping through the air in a similar fashion. This is when she notices that Amon is provoking these glitches and needs to log out of the server to reset the car placements. She does this in character by yelling out to him, “Amon, you need a nap! Please nap! Sleep! Sleep! Sleep!” 

Telling Amon to sleep is an apt indexical reference because when a character logs out and logs back into the server, they are spawned back to their apartment and seen literally getting out of bed. Claire telling Amon to sleep reinforces this widely accepted use of language, so DisguisedToast knows exactly what she is talking about. He logs out soon after.

Within these clips, you can see and hear the players show their adoption of a particular language of NoPixel role-play that is not official but falls under the ideology of participation. Had the role players gone out of character using language, such as Yuno saying there was a literal glitch, and officer Everly telling Amon to literally log out, the viewers would most likely reject this instance as Content, as it robs them of the suspension of disbelief. When the Reddit user posted the clip, there was further validation of the indexical reference to the glitch. So successful role-play can be effectively categorized as Content. Later in this paper, I show an example of consequential out-of-character behavior that viewers revolt against.    

Memes that turn content into Content

I featured streamer Buddha in my last paper because of his status as a well-known veteran role play streamer. He has accumulated 1181hrs 8mins of on-stream game time since the launch of NoPixel 3.0.

In 2017, during a previous version of the NoPixel server, Buddha interviewed with an Esports-related YouTube channel and gave viewers a rare opportunity to hear from the player behind the character.

He explains how the initial appeal of Lang Buddha came from his voice. While he controls all of the character’s mechanics, he makes his character unique through his accent. Despite Lang being of Chinese descent, his accent is not explicitly coded as Chinese. Instead, Lang speaks like Lang, inspired by the voiceovers from cartoons like Family Guy and the Simpsons. 

He also gives a succinct response to the appeal of the NoPixel server for longtime role play. “It feels real.” Through each iteration, the NoPixel server’s graphics have improved so significantly that it enhances role-play. According to Buddha, the secret to good role-play is going with the flow. His personal participation ideology is to keep a conversation going, similar to the “Yes, and…” rule-of-thumb in improvisational comedy. Even if the conversation starts with something ordinary and evolves into something more absurd, you should let the conversation continue its course, avoiding abrupt endings. 

In practice, Buddha executes his strategies near flawlessly in his streams. He will constantly pick up on current storylines and build new ones as he meets other characters. Buddha surveys and takes advantage of the city landscape by interacting with it effortlessly. He knows his way around the city, and the characters know who Buddha is through his iconic voice and appearance.

In response, Buddha’s audience can reflexively pick up on entertaining moments and categorize them as Content. They do this through language, specifically while chatting during the live stream. Twitch chats can be very fast-paced and impulsive. Live moments warrant quick reactions, so there is much brevity in a chat like Buddha’s, where he can have 16,000 viewers watching him. Because Twitch gives prominence to the chat function, fitting it right next to the streaming video, it can be an integral part of the live stream watching experience. 

I found two studies that explored the significance of Twitch chat, one concerning viewer satisfaction and the other about predicting the popularity of content. In the study of the first article, “Watching Players: An Exploration of Media Enjoyment on Twitch,” the researchers found that the chat function was influential in a viewer’s overall experience of a live stream. Through an online survey of 548 Twitch users, they found that “…using the chat… was associated with higher levels of media enjoyment (b ¼ .20, p < .001).” By defining Twitch as a community, the chat function plays an important role as a conduit for social interaction between not only the streamer and the viewers but between the viewers themselves. It also facilitates more entertaining experiences on the platform.

In the second study examining the relationship between Twitch chat and the popularity of a certain stream (“Learning How Spectator Reactions Affect Popularity on Twitch”), the researchers found that the use of emojis can be a predictor of popularity. This is because they denote “‘relatable’ emotion toward the streamer—rather than simple cheers or happy mood.”   

In keeping with this method, I observed Buddha’s live stream and chat relay simultaneously to see if I could find a connection between how chat reacts to the stream and the Content being derived from these utterances. I specifically looked for moments where a significant number of non-textual, memetic utterances were relayed in chat. I hypothesized that a stream of emotes in reaction to a moment of the stream signified a piece of Content react. I have included clips from Buddha’s stream where I was able to both associate the stream with the realization of Content and glean the type of emotional response that content was receiving. 

This first clip is a moment that elicited positive reactions from the chat because they found it to be humorous. As Lang runs across the street, he gets hit by a car and falls, responding with a vocal utterance that may not express pain but is keeping with a joking register. It sounds like a giggle. He immediately gets up from his fall, jumps on the road down to a street where he is struck by a car for a second time. This time it seems intentional as he giggles again. 

Within that 30 second clip, from the time Lang falls the first time to the end, I counted 67 non-textual utterances indicating laughter. These were represented by the emotes KEKW, LUL, LULW, and OMEGALUL. Each emote is iconic of a person laughing out loud.

In this second clip, chat used emotes to react sadly to content. Buddha is on the phone with Marlo, a character looking to collect the debt from Humberto, him being the character HasanAbi has retired for the time being. Chat is aware that Donnie is “gone,” so they react using the sadge emote, which is the character Pepe the Frog frowning in a slouched position. From time 0:13 to the end of the call at 0:46, the sadge emote is relayed 60 times, with a few messages, including “Donnie” or “Humberto.” 

For chat, this was content that referenced the universe of the NoPixel server and showed how choices outside the server affect what happens inside the server, building onto the already-established lore. Viewers responding memetically meant that they felt a sense of solidarity within Buddha’s community, an important facet that does contribute to viewers’ enjoyment, and established a consensus that this clip was Content.

Finally, I have a clip of a piece of content that evoked a particular type of non-textual utterance that represented the memeification of Buddha as a streamer. While there are “Global” emotes provided to all Twitch users, like the aforementioned LULW, there exists subscriber-only emotes. These can only be used by users who have subscribed to the particular streamer and usually visually relate to the streamer’s persona. So for this instance, the emote reactions could only be shared by Buddha’s subscribers. 

Buddha is driving his vehicle with Yuno, and when he drives up a ramp, flies through the air, and lands relatively unscathed, Buddha subscribers are using the buddhaWICKED emote. It is a pastiche of the WICKED emote (where Pepe dons a pair of sport sunglasses) and Buddha’s icon. 

From the moment Buddha first starts to land, the emote shows up in chat 25 times, significantly less because of its exclusivity. Subscribed viewers demonstrate a more dedicated, though still a parasocial, relationship with the streamer’s content and actually play a unique role in curating Content. In fact, only Buddha’s subscribers can clip his streams, which I found surprising. As a result, the subscribers have the final authority to determine what is worth spreading to the masses. This is a rare but very influential way for the streamer to regulate the type of content that can become Content. 

When language deprives users of role play Content

As evidenced in the clips and background research I have presented so far, the role-players of NoPixel have intentions to make the server welcoming and fun through immersion and maintain a certain sense of humor around their activities. Also, there is a correlation between what is defined as Content and viewers’ reaction to that moment in the live stream. Now, there have been popular instances where hostile and/or non-immersive role-play becomes too egregious to be considered Content, or at the very least fun Content that maintains the positive key of NoPixel role-play language because of viewer reactions. I will expand upon a disruption to the GTAV RP content industrial complex from a streamer named xQcOW, also called xQc.

xQc is an extremely popular streamer, being the most popular in 2020 in terms of hours watched. In this article, the publication argues that this is the most important statistic to determine the popularity of a streamer because it “combines a streamer’s dedication to going live and their ability to attract large audiences consistently.” 

His presence in the NoPixel server has inevitably garnered attention from Twitch users. The top two most view GTAV clips of all time came from his streams. 

Despite his popularity, xQc was a notorious role player. Most notably, viewers began to question his communicative competence. The NoPixel developers have banned xQc four times due to rule-breaking behavior.

However, they have never cited the rule(s) that he broke, so viewers were left to speculate on Livestreamfail. In this thread, the original poster curated a group of clips, presenting them as content that may have led to his ban. You will notice in the post that there was another deliberate effort to keep this content from taking over the subreddit. 

I will focus on one of the clips cited because it best highlights xQc’s language breaches that could have broken the NoPixel server’s rules.  

As other users have in the comments, I speculated that the breach committed from xQc came from his excessive use of expletives, up to the point where it was questioned if he was still in character when he was degrading the cop. The first alarming utterance, “Fuck you, you piece of shit,” is an atypical response for a GTAV RP character. Right after xQc declares, “I hope you fucking die.” Again, very atypical, alarming language that may have broken NoPixels guidelines around “Not Valuing Life (NVL).”

Within these guidelines is a rule that, in character, you should not be “antagonizing gangs/police or armed individuals for no character reason.” Surely xQc was antagonizing the cop, but the question was whether he had no character reason. To him, he was being arrested by a cop, so his language warranted it. 

But I would argue that that is not the case. Throughout my paper, I have shown that live stream viewers have tremendous power to constitute role play Content, especially through how they respond or talk about it. In this thread about this particular clip on Livestream fail, the most upvoted comments have a common element of dismissing this content as in-character role play. 

The phrase, “I hope you fucking die,” was an obvious trigger for users to determine xQc’s role play as unacceptable. And because of the extreme nature, it was enough for them to deem his language out of character and justify the ban. So the fact that there was substantial reaction and discussion from viewers could mean that this was unique content, but it is not Content because it goes against the spirit of GTAV role play that viewers expect from streamers. 

XqC has since been unbanned, so perhaps he has adopted a more friendly language that viewers can enjoy. He is still dominating the role-play live stream world.

Conclusion

To explore and refine the nebulous concept of content was an incomplete and challenging journey. I believe I found some material criteria to determine Content, but I am no oracle at predicting. That is not to say that it is impossible or that the word still has no meaning. Between my last paper and this one, I have found compelling evidence that language is critical to unearthing and understanding potential Content.

In my research and observations, I found parallels between GTAV RP and the findings from Inmaculada M. García-Sánchez’s article “Serious Games.” Through her observations of young Morrocan girls’ role play, she noted how the peer group is a “primordial locust for immigrant children’s language socialization.” When these peer groups are diverse, some girls will socialize with others by exposing them to previously inaccessible parts of life through pretend-play. They also will enforce rules on how to speak in pretend-play. In turn, these girls adopt a socio-cultural identity that elevates certain languages, activities, and ways of being over others in order to belong. Particularly, being a “high society” Spanish-speaking woman was the ideal that could be achieved through role-play for the time being. 

In GTAV RP, a similar kind of language socialization happens among characters. NoPixel’s rules provide some baseline standards on role-playing ethics and procedures. While playing, characters pick up on acceptable language, communicative practices, and modes of participation. As a result, there comes a widely accepted participation of ideology. When your language does not align with the values and normative stances of NoPixel, not only is the character outcast, but so is the streamer. 

The same processes are happening amongst viewers. Audience consensus shapes the GTAV RP content industrial complex. Through a variety of instrumentalities (such as memes and discussions), users can quickly be socialized, reinforce content standards, and shape the identity of their community, whether it be a subreddit or a streamer’s subscriber base. These can be positive ways to create a sense of belonging among disparate viewers. However, too much can disrupt the greater live stream community’s pool of Content. 

This is not entirely the fault of GTAV RP streamers or viewers. Twitch as a platform intentionally promotes the spread of homogeneous content. I witnessed this personally during my research. The more I watched and followed GTAV content on Twitch, the more it was recommended to me. Furthermore, as I watched more of a particular streamer, Twitch would try to launch me down a rabbit hole of those streamer’s clips. More users streaming for longer hours means more watch time on the platform, which means more opportunities to show users ads. The most efficient way to do this is to reinforce popular content. 

There is no doubt that these trends will follow. The next popular game will capture the attention of many streamers and viewers, and then live streams will transform into Content, memes, and points of discussion. Further research on past and future popular game phenomena is needed, but the construction of the GTAV RP content industrial complex could indeed be a part of a pattern that comes with video game content life cycles. This Content may be curated organically through user practice and consensus but can be distributed artificially via an algorithm. 

So there are both positive and negative consequences of content aggregation that breed hive-mind attitudes. They can simultaneously facilitate community-building and an unbalanced power dynamic within it. I see why NoPixel and viewers are quick to implement stop-gap measures and call out inappropriate behavior to keep the role play live stream community balanced and welcoming. The best way they and streamers can learn how to enforce rules of engagement better is to keep using language as a tool to propagate positive content. GTAV RP has unexpectedly brought so much enjoyment into people’s lives in a time of prolonged isolation; it deserves that kind of care.  

---

Works Cited

Bottomley, Andrew J. “Giant Pools of Content: Theorizing Aggregation in Online Media Distribution.” JCMS: Journal of Cinema & Media Studies, vol. 59, no. 1, Fall 2019, pp. 149–156. EBSCOhost, search-ebscohost-com.libproxy.mit.edu/login.aspx?direct=true&db=asu&AN=141941602&site=eds-live&scope=site.

García-Sánchez, Inmaculada M. “Serious Games: Code-Switching and Gendered Identities in Moroccan Immigrant Girls’ Pretend Play.” Pragmatics, vol. 20, no. 4, Dec. 2010, pp. 523–555. EBSCOhost, search-ebscohost-com.libproxy.mit.edu/login.aspx?direct=true&db=edb&AN=59210206&site=eds-live&scope=site.

Kim, Jeongmin, et al. “Learning How Spectator Reactions Affect Popularity on Twitch.” 2020 IEEE International Conference on Big Data and Smart Computing (BigComp), Big Data and Smart Computing (BigComp), 2020 IEEE International Conference On, Feb. 2020, pp. 147–154. EBSCOhost, doi:10.1109/BigComp48618.2020.00-84.

Wulf, Tim, et al. “Watching Players: An Exploration of Media Enjoyment on Twitch.” GAMES AND CULTURE, vol. 15, no. 3, May 2020, pp. 328–346. EBSCOhost, doi:10.1177/1555412018788161.