“How it actually works”: Algorithmic lore videos as market devices

Abstract

On YouTube, self-styled algorithmic experts claim to know how algorithms “actually work.” However, their knowledge is largely speculative. Developing recent work that pays attention to algorithmic expertise, I argue that algorithmic lore videos are “market devices” which are economically productive for the platform in four ways: they (1) legitimize platform narratives of algorithmic objectivity, (2) teach creators how to calculate the value of content and format it according to platform metrics, (3) encourage creators to build and govern audiences, and (4) justify continued content production even when it does not pay off for creators. Thus, despite claiming to describe “how the algorithm actually works,” algorithmic lore videos are performative; by teaching creators how to understand and act in the platform economy they do important work to bring the platform’s “visibility markets”—its labor market of content creators engaging in content production and its goods market of content to be watched—into being.

Introduction

As the world’s most popular video hosting website, YouTube holds a prominent position in the platform economy. Through its infrastructures, YouTubers or Creators can monetize their content on the platform in the hopes of earning an income. However, these infrastructures are conditioned on visibility, which is difficult to attain and maintain on a platform where over 300 hours of content are uploaded every minute. Recent events have also underlined the uncertainty of a career on YouTube. For example, during the so-called YouTube “adpocalypse” in 2016–2017, advertisers boycotted the platform when they discovered that YouTube’s AdSense algorithms were showing their advertisements alongside extreme content. Creators relying on the YouTube Partner Program (YPP)—which monetizes content by algorithmically matching ads to content and paying creators a fixed amount per 1000 impressions or cost per mille—saw their revenue drop as fewer ads were available for monetization. To bring back advertisers, YouTube increased algorithmic content moderation and revised content guidelines. However, this also increased the uncertainty of the system as creators faced search results demotion and automated demonetization on old and new content. As a result of events like these, creators work hard to learn how the platform and its algorithms work, to optimize visibility and reduce algorithmic uncertainty.

Information asymmetries between platform owners and workers are a key factor in maintaining inequalities in digital labor markets (Gray and Suri, 2019; Ravenelle, 2019; Roberts, 2019; Rosenblat, 2018; Rosenblat and Stark, 2016). However, research has shown how workers develop strategies for overcoming these information asymmetries. These strategies include crowdsourcing information about employers (Irani and Silberman, 2013) and algorithmic systems (Bishop, 2019); or making oneself “algorithmically recognizable” to platform algorithms (Gillespie, 2017) through practices of self-optimization (Bishop, 2018; Cotter, 2018). Despite these efforts to empower algorithmic subjects, self-optimization can end up reinforcing race, gender, and class biases in algorithmic systems. In the context of YouTube, efforts to reverse engineer algorithms to produce valuable algorithmic signals can reinforce the “visibility hierarchies” of the platform, which ultimately encourage the production of content which aligns with the economic goals of the platform and its advertisers (Bishop, 2018, 2020). While intermediaries like multi-channel networks (MCNs) may also assist with these tasks, this assistance typically comes at a cost, for example, by taking a percentage of creator revenue or gaining control over content production (Cunningham et al., 2016; Lobato, 2016; Siciliano, 2021).

This article focuses on videos produced by self-styled experts who teach creators how to navigate YouTube’s algorithmic systems and the information asymmetries they create. These videos contain what Bishop (2020: 1) refers to as “algorithmic lore,” which is “a mix of data-informed assumptions that are weaved into a subjective narrative.” This article develops Bishop’s suggestion that despite its speculative nature and regardless of its accuracy, algorithmic lore ultimately ends up benefiting the platform. Drawing on an analysis of 15 algorithmic lore videos, I argue that algorithmic lore videos are “market devices” (Muniesa et al., 2007). As they become entangled with creators, viewers, and YouTube’s algorithms, these market devices develop certain forms of economic thinking and calculative agency among creators which ultimately benefit the platform. This is because algorithmic lore videos discipline creators in four ways conducive to the platform’s economic goals: they (1) legitimize platform narratives of algorithmic objectivity, (2) teach creators to calculate value and format content according to economic metrics, (3) encourage creators to build and govern audiences for the platform, and (4) justify continued engagement by moralizing the difficulties of content creatorship.

Conceptualizing algorithmic lore videos as market devices clarifies a key mechanic of neoliberal platform governance. Neoliberalism requires careful interventions to establish the conditions under which the “natural” and “free” market flourishes—importantly, by encouraging the adoption of neoliberal subjectivities informed by market principles of entrepreneurialism, individualism, and competition (Foucault, 1979; Gane, 2012; Mirowski and Plehwe, 2009). Algorithmic lore videos encourage creators to become entrepreneurial, self-disciplining subjects and invest moments of breakdown, frustration, and failure with meaning. Together, this keeps creators engaged in the competition for visibility even when it does not pay off. In the process, these market devices reify the objectivity and neutrality of the platform, which further displaces accountability for social, political, and economic inequalities from the platform.

Thus, despite their claims to merely describe “how the algorithm actually works,” algorithmic lore videos are performative; by teaching creators how to understand and act in the platform economy they do important work to bring the platform’s “visibility markets”—its labor market of content creators engaging in content production and its goods market of content to be watched—into being. In the sections that follow, I explore how we can theorize algorithmic lore videos as market devices. After detailing my data and methods, I discuss the lessons offered by algorithmic lore videos, and how the platform stands to benefit the most from their circulation. In concluding, I note potential avenues for further research into algorithmic experts and algorithmic lore.

Eyeballs to money: theorizing algorithmic lore videos as market devices

In many ways, YouTube’s algorithmic experts and academic researchers share a similar goal: to “know” algorithms. To do so, scholarship in the STS tradition has captured algorithmic systems as “intricate, dynamic arrangements of people and code” (Seaver, 2019: 419), the “multidimensional ‘entanglement’ between algorithms put into practice and the social tactics of users who take them up” (Gillespie, 2014: 183), or has paid attention to the “algorithmic imaginaries” we develop through our everyday interactions with algorithmic systems and how they inform our use of platforms (Bucher, 2018). STS approaches to algorithms help capture how their supposedly objective and rational agency is contingent on the networks of actors, cultural assumptions, and negotiations which bring them into being and sustain them, and to examine them and their effects in practice (Crawford, 2016; Neyland and Möllers, 2017; Seaver, 2019).

STS scholarship on the performativity of markets draws our attention to the sociotechnical infrastructures of the platform economy. As Callon (1998) argues, economic theory—as articulated by economists—disentangles actors, objects, and interactions from their social context and recasts them within calculable relationships, develops calculative agency among actors, and mobilizes actors and devices to stabilize these market forms (Callon, 1998; MacKenzie, 2006). This process is often assisted through “market devices” or “material and discursive assemblages that intervene in the construction of markets” (Muniesa et al., 2007: 2). Scholars have examined how sociotechnical elements of financial systems—such as credit scoring algorithms (Poon, 2007) and high frequency trading algorithms (MacKenzie, 2018)—shape and construct markets by translating complex social interactions into abstract models. The process of translation is important, as it embeds moral values within economic infrastructures and obliterates the granularity of lived experience in favor of the model. As Karpik’s (2010) work on “judgment devices” demonstrates, market devices also play a role in the moral valuation of objects. As such, it is not simply about the right value, but the right value for the right object.

Combining STS approaches to algorithms and markets, we can conceptualize YouTube as an economic infrastructure that translates eyeballs into money, algorithmic experts as lay economists of the platform economy, and algorithmic lore videos as market devices. Through the sociotechnical entanglement of actors and market devices such as algorithms and interfaces, the social practices of watching, liking, commenting on, and creating content are translated into economic value through advertising infrastructures like the YouTube Partner Program (YPP). The platform’s algorithmic systems are judgment devices: search, recommendation, trending, and discovery algorithms will match the right content with the right viewer, and the YPP’s algorithmic matching systems can match the right content with the right ad and viewer. As lay economists, algorithmic experts transform the complex sociotechnical interactions that make up the platform—who participates in the production, selection, and circulation of content, how this content circulates and is served, and how actors and goods should behave—and transform them into abstract models for their viewers. These models then circulate through YouTube’s economic infrastructures as algorithmic lore videos, which are selected and served to potential viewers by the very system that experts work to theorize.

As MacKenzie (2006) notes, a performative approach to economic theory presents three benefits. First, it does not focus on the accuracy of economic models so much as it focuses on the potential effects of their adoption. Following MacKenzie’s (2006) formulation, it allows us to approach the theories of experts not as a “camera”—that is, a snapshot of how algorithms make decisions. Instead, we see them as an “engine”; they shape how creators approach and understand “the algorithm.” Second, it draws attention to how the technicalities of models become consequential in market infrastructures. In the pursuit of knowledge about the algorithm, experts draw on the platform’s data infrastructures to develop their theories. A performative approach to algorithmic lore helps us understand not only how technical elements like such as economic metrics shape the output of cultural producers, but also how they are given meaning. Third, it prompts the question of how this could be otherwise. Within this entanglement of viewers, creators, algorithms, interfaces, and infrastructures, algorithmic lore videos make a difference by promoting certain ways of acting and behaving on the platform, and ultimately, making the translation of eyeballs to money possible.

Data and methods

I employ a grounded theory approach to understand the role algorithmic lore videos play in constructing YouTube’s visibility markets. If an individual seeking out information about algorithms is new to the platform or looking to increase their visibility, these videos are one of the most accessible resources to them. I used a purposive sampling strategy to collect videos that a prospective YouTube creator might be served on the platform. Results for a keyword search for “YouTube Algorithm Works” were ranked and refined to focus on videos posted in 2018–2019. ¹ To ensure that the theories of the “algorithm” were relatively recent, videos from algorithmic experts which were posted before 2018 were removed. The sample was later expanded to include a second video from Brian Johnson, a video from Sunny Lenarduzzi, as well as a video from Derek Muller—who operates the channel “Veritasium.” In total, 15 videos were collected and analyzed. I analyzed the latent content of these videos—including how algorithmic experts present themselves, how actors and algorithms are framed, and strategies offered—to understand themes as they emerged. Like recent work in STS and Critical Algorithm Studies, this methodology seeks to understand “algorithms as culture” (Seaver, 2017) or the “ontological politics of algorithms” (Bucher, 2018); that is, how algorithms and their effects are enacted through partial understandings and everyday practices, rather than the code of the black-boxed algorithm.

The videos in the sample can be divided into two broad categories: unofficial and official. First, I analyzed 13 algorithmic lore videos on YouTube made by creators who are not officially affiliated with the platform. Most videos are framed as one-on-one consultations between the expert and the viewer, and range anywhere from 5 minutes in length to over an hour. Of the unofficial videos included in this sample, four were from the “Education” category, four from the “How To and Style” category, three from the “Science and Technology” category, and two from the “Entertainment” category. The majority of unofficial experts included in the sample operate channels which focus primarily on creator development and algorithmic optimization in the creator economy (e.g. Eves, Darmawangsa, Johnson, Blake, Lenarduzzi). Others operate channels which produce educational videos about science, technology, engineering, and mathematics (STEM) subjects, but featured the algorithm as a special topic (e.g. Muller, Crawford). Finally, the vidIQ channel is a business channel which promotes the company’s popular third-party plug-in and its creator development courses through algorithmic lore videos. It is important to note that in the initial search results, the returned videos were overwhelmingly from White male creators who conform to the hegemonic geek subcultures of the platform (Bishop, 2020).

I also analyzed two official videos from YouTube to provide a benchmark for comparison. YouTube publishes content to its “YouTube Creators” channel—often drawn from its “Creator Academy” series of online courses—as well as its “Creators Insider” channel for new and established creators to learn how the platform works. ² These videos are often quite short; they explain how the platform’s technologies work in general terms and how creators can try to make their content more appealing. However, unlike unofficial videos, these often do not go into great detail—as I discuss in further detail later, its overview of “the Algorithm” provides general tips like “follow the audience” rather than telling creators what metrics to focus on. The Creator Insider video, ³ which features a member of the platform’s Search and Discovery team, provides more detail. However, the overall takeaway from this video is simply to make content that people will watch and keep watching.

While official videos do not discuss any of the “hacks” or strategies of unofficial videos, the claims of official videos are frequently referenced and unpacked in more detail by unofficial experts. It is important to include both official and unofficial videos to understand how the platform’s framing of algorithms corresponds with that of algorithmic experts, and how algorithmic experts draw on these videos to develop their own theories. Despite differences in how they describe the algorithm, unofficial videos share a similar premise: to reveal “why you’re not getting views or [subscribers],” ⁴ often by “breaking down,” “manipulating,” or “hacking” the algorithm. The actual suggestions made by experts to achieve this goal varied greatly. Some experts teach creators how to understand and use the platform’s interfaces in ways anticipated by the platform, some teach tricks that put these interfaces and systems to unintended uses as “hacks” or to conduct “tests,” and some dissect data analytics from their own content to give more general advice about content creatorship on the platform.

While this approach attempts to capture how a user might encounter algorithmic lore videos on the platform, there are important limitations to consider. First, with any platform that employs personalization algorithms, it is difficult to claim that observed results of keyword searches are replicable or definitive. To limit this effect, I used a separate browser and remained signed out for the duration of the keyword search. Any subsequent interactions with the dataset on YouTube (e.g. downloading transcripts) were conducted with a different browser. Future sampling strategies may repeat the keyword search at multiple points and on multiple devices to see how these results may change. Second, a sample of 15 videos may seem small for a project looking to understand platform infrastructures and governance. However, a smaller sample allows for in depth analysis of how different algorithmic experts present their theories about “the algorithm.” Furthermore, even a small sample of algorithmic lore videos can have a large impact. It is an important to remember that these videos were recommended by the platform through its search algorithms. At the time of data collection in early 2019, the unofficial videos in this sample had been viewed 1.9 million times. As of the time of writing in late 2020, these videos had accumulated 3.6 million views. However, this sample only includes one or two videos from each expert out of their entire channel. If we consider the channel view count of each unofficial expert, the importance of considering these market devices becomes clear; as of late 2020, their channels had been viewed approximately 2.4 billion times.

“The algorithm looks for value”: lay economic theories of YouTube’s market infrastructure

Algorithmic experts promise to reveal the YouTube algorithm to prospective viewers. As Bishop (2020) argues, experts claim that anyone with the right attitude toward experimentation and data can circumvent the platform’s control. However, this expertise is deeply entangled with platform meritocracy and White male geek culture; appeals to data visualization frame this knowledge as objective and results as achievable by anyone, despite the exclusion of feminized and marginalized genres of content in their data sets. Thus, Bishop argues that despite their self-branding as adversaries of the platform, these experts instead encourage ways of self-optimizing for the platform rather than truly undermining its systems. Conceptualizing algorithmic lore videos as market devices encourages further critical engagement with the models algorithmic experts develop, as well as the potential consequences of their adoption by viewers. In this section, I explore the lay economic theories these algorithmic experts develop, their lessons for creators, as well as their moralizing force.

Experts’ message that creators have to simply follow these models to be economically successful relies on the idea that YouTube’s visibility markets operate rationally and can be knowable. Ironically, experts draw on publicly accessible corporate sources to provide this “insider” knowledge about algorithms’ rationality to their viewers. For example, YouTube’s Creator Academy video frames “the algorithm” as follows:

[The algorithm is] basically a real-time feedback loop that tailors videos to each viewer’s different interests. It does its best to show the right videos to the right viewer at the right time across the whole planet. How do we do this? With data! ⁵

YouTube emphasizes the importance and objectivity of data and reifies “the algorithm” as an independent entity that can make the right call with the right data. Experts then reproduce these claims to objectivity in their videos. Consider Derral Eves, ⁶ a well-known algorithm expert on the platform with approximately 600,000 subscribers and over 58 million channel views. ⁷ Self-described as one of the first experts to be certified in “audience growth,” ⁸ Eves is also a frequent contributor to YouTube industry events such as VidCon. His channel consists almost entirely of algorithmic lore videos ranging from walkthroughs of platform features, analyses of popular channels, interviews with YouTube executives and data engineers, and discussions about industry events. Eves explains that “YouTube makes data-driven decisions off of the viewer and off of the creator and off of the advertiser and as they were looking at the viewer they noticed that the viewer was actually doing something.” ⁹ However, by virtue of their positioning as trusted intermediaries (Bishop, 2020), experts’ claims to platform rationality and objectivity carry more weight than those of the platform. By reproducing the company line, algorithmic experts also help legitimize the platform economy.

Algorithmic lore videos also frame algorithms, creators, and audience members as rational value-maximizing agents. Experts’ basic theory is that the algorithm seeks out the highest value videos, and that viewers will click on a video if its value—communicated through its thumbnail, title, and first two lines of the description—aligns with what they are looking for. Devon Crawford ¹⁰ —a self-taught software and electrical engineer with 412,000 subscribers and 19.8 million channel views ¹¹ —explains that if a video is valuable enough, people will watch it to completion, thus demonstrating the value to the algorithm through its watch-time metric. Crawford, hence, underlines the knowability of the “unknowable” algorithm: the secret is that there is no secret, as he says. To hack the algorithm, creators “just” need to make high-value videos that the rational algorithm and viewer will pick up on. What is important is not whether he accurately knows the algorithm, but instead, how the algorithm is framed as a rational, knowable actor picking up the activities of rational viewers.

How can creators “know” if the algorithm sees their content as high value? Experts claim the rational algorithm draws on objective data, so to know the algorithm creators should do the same. According to experts, these data can be found in platform tools like the YouTube Creator Studio, later rebranded in 2019 as YouTube Studio. ¹² This interface allows creators to manage uploaded content, access data analytics, monitor monetization and copyright, and access free resources for video production. The analytics page presents metrics under the following four tabs: overview, reach, engagement, and audience. Reach metrics evaluate how often content is served to potential viewers. For example, “Click Through Rate” (CTR) measures how often the link to a video is clicked on after it has been shown to potential viewers. This metric quantifies how well content can transform impressions—when a video is shown to a potential viewer—into views. As a result, it often serves as a proxy for how effective a thumbnail, title, and description are. “Watch Time” is an engagement metric which refers to the total amount of time that the overall audience has watched a video. This is a key metric for qualifying for monetization on the platform, as it quantifies the potential draw of creators’ content for advertisers. “Average View Duration,” often referred to as “Audience Retention,” is another engagement metric that measures the average amount of time a video keeps the viewer engaged before they exit the video. Finally, audience analytics include metrics like “unique viewers”—how many independent audience members watched a video—as well as “average views per viewer”—which measures the average number of times viewers rewatch your video.

As I discuss in the following sections, each of these metrics entails different strategies for pursuing visibility, some of which use the YouTube Studio and its metrics for unintended purposes. Experts also differ on which metric is most important, and how they should be used. For example, Miles Beckler—a self-described digital marketing expert with 164,000 subscribers and 8.6 million channel views ¹³ —emphasizes CTR over watch-time to increase the likelihood of clicks, while Eves emphasizes watch-time over CTR to demonstrate the overall quality of content to the algorithm. This points to the constructedness and subjectivity of data analytics. As Christin (2020) argues, metrics have interpretive flexibility. As symbolic resources, metrics can have different uses and meanings depending on their context. Despite disagreements on which metric to favor and how to put it to use, experts agree that creators can rely on data metrics as good and objective indicators for the value of content and the activities of the platform’s visibility markets. Thus, like the experts studied by Bishop (2020), experts in this sample reify the objectivity of metrics and the value of algorithmic experimentation. What a market devices approach brings to this is a focus on how the technicalities of these models—that is, these interpretations of metrics and their role in knowing the algorithm—become consequential.

Experts’ theory that YouTube’s visibility markets markets for content are based on knowable and rational operations forms the basis for their “expert recommendations” to content creators. These recommendations, in turn, become economically productive in the following three ways: they teach creators (1) how to calculate the value of content, and to format their content in a way which maximizes value; (2) how to build and govern audiences for the platform which maximizes audience retention; and (3) that creators must keep working despite YouTube’s shifting data market infrastructures. In combination, these lessons from algorithmic lore videos render YouTube’s visibility markets easier to govern and more economically productive.

“Drive more views to your videos”: developing calculative agency

Based on their theories of YouTube’s market infrastructure, algorithmic experts teach content creators not only how to calculate the value of their content; they teach creators how to increase it. For experts like Crawford, value has both social and economic dimensions. Social value is tied to whether the potential viewer feels satisfied and that the time spent watching was worthwhile. To maximize this value, experts urge creators to ensure their content aligns with its audience-facing elements (e.g. title, thumbnail, and description). Economic value is tied to whether the content can capture an audience and keep them watching for longer. To maximize this value, experts encourage creators to use metrics to inform their content production strategies. Experts argue that if creators carry out these calculations correctly their visibility will increase: “the algorithm” will classify their video as high value and serve it more often in search and discovery, the audience will seek out more of their content, and it will be more competitive for advertising opportunities as it draws a larger crowd. As market devices, algorithmic lore videos develop calculative agency among creators by encouraging them to make their content contingent on the data flows of the platform (Nieborg and Poell, 2018).

Algorithmic lore videos teach creators how to translate metrics into content production strategies. For example, vidIQ—a YouTube channel for the popular third-party analytics extension with approximately 724,000 subscribers and over 49 million channel views ¹⁴ —discusses the theory that uploading higher resolution videos can result in algorithmic favorability, as the platform features a “4K” badge on the search results page. ¹⁵ Rob Wilson, the presenter, argues that while the algorithm might not see higher resolution videos as more valuable, viewers might. VidIQ argues that creators should analyze performance metrics like CTR to decide whether 4K content is actually valuable to the audience. If platform data streams indicate a correlation between 4K content and better engagement and reach metrics, then creators should alter their technical production practices. The value of other technical decisions such as the run-time of content can also be calculated in this way. Experts explain that if the YouTube Studio shows good audience retention, then creators should experiment with longer content to see if they can keep that retention rate up with longer videos.

Algorithmic experts also teach creators how to minimize uncertainty when “betting” on popular types of content. For example, Brian G. Johnson—a digital marketing strategist with 142,000 subscribers and 9.4 million channel views ¹⁶ —teaches creators how to “hack” the notification system to anticipate trending topics. Alluding to neoliberal ideals of free-market competition, Johnson argues that it is important to understand what topics successful channels are discussing to avoid losing out on potential views. To better drive views to their channel, creators should monitor and emulate what successful channels are doing:

They say that amateurs borrow and professionals steal. And if you want to grow fast here on YouTube and if you want to make jaw bending, life changing income, then steal. On an Apple or Android device, access YouTube. Next, click on notifications and turn notifications on, but say no to sounds. You don’t want to be interrupted day-to-day. You see, we’re not going to use notifications like most. We’re going to use notifications to drive more views to our videos. ¹⁷ (emphasis added)

Johnson explicitly encourages copying from successful channels by teaching creators to use the notification tool to follow successful channels in their niche: the platform’s notification tools will automatically notify creators when these channels upload new videos so that creators can piggyback on their content’s popularity. He also encourages creators to emulate traditional media industries by anticipating what he calls “Big, Broad and Proven Hollywood Topics.” For example, Johnson refers to the upcoming film “The Meg” as a signal that viewers may be more interested in content related to sharks. He proposes that producing content that complements trends in Hollywood can similarly help creators to grow their channel on the platform. While this “hack” does not use the platform’s notification system as intended—that is, to deliver helpful suggestions for videos—it is important to note that it does not undermine the platform’s trending algorithms. In a fascinating way, this hack encourages content creators to behave almost like market speculators—to anticipate when trends are about to emerge on the platform and “bet” on content to benefit from its raised popularity before moving on to the next trend.

A market devices approach shows us the role that algorithmic lore videos play in developing calculative agency among creators on the platform. Algorithmic experts teach creators that the ability to calculate and maximize the value of content is essential for success. Not only do they teach creators how to know the algorithm through metrics and the YouTube Studio, but they also teach them how to calculate and maximize the value of their content through these data infrastructures. Through metrics, creators can also calculate what the audience finds valuable, and plot a course for future content production. As seen with Johnson’s notification hack, calculative agency extends beyond the YouTube Studio, because creators are also taught methods to anticipate trends on the platform and to increase the potential value of their content accordingly.

“Follow the audience”: capturing and governing audiences through data

Even highly valuable content can find it difficult to achieve visibility on YouTube. As Johnson points out in his notification hack above, the eyes which are available to view creators’ content are scarce. This supply-demand theory of attention points to the next lesson from algorithmic lore videos: to succeed, creators need to capture the audience and keep them watching. In economic terms, what Johnson is arguing is that creators also need to learn how to control the demand side of the market for content. The YouTube Creator Academy video makes this clear as follows:

“OK, then how can I get the algorithm to like my videos?” It’s pretty simple: get the audience to like your videos. That’s because the algorithm follows the audience. If people love your videos, the algorithm will surface them to others. ¹⁸ (emphasis added)

How do algorithmic lore videos teach creators to do this? In this section, I look at how creators are taught to “follow” the audience through platform metrics. Drawing on the same infrastructures which quantify users and infer algorithmic identities onto them for marketing purposes (Cheney-Lippold, 2011), algorithmic lore videos teach creators how to build and govern audiences for the platform.

Algorithmic lore videos argue that creators need to understand the audience as “fuel.” Consider, for example, Johnson’s video in which he explains the significance of audience retention for maximizing economic value:

And what you need now is fuel, fuel to power your video. You’ve told YouTube where you want it to show up in search. It’s going to understand the kind of audience to push it out to if you’ve got enough fuel to power the video, and the fuel comes from your audience retention, your accumulated minutes watched, all those video metrics, because at the end of the day, it’s the watch time metrics that fuel the videos, and it’s the thing that really separates small struggling channels from the big channels that do well. They have the ability to keep the audience watching, the very thing YouTube talks about again and again. ¹⁹ (emphasis added)

Here, Johnson is theorizing how the platform takes the social activities of the audience and translates them into algorithm-legible metrics like CTR and watch-time. No matter how valuable the video is, without positive audience interactions, it cannot achieve visibility. Thus, a key task for creators according to Johnson is to “fuel” their content by successfully capturing the audience and producing high audience retention statistics—ideally near 100%. Creators need to translate momentary impressions—or how many times your content is seen as an option—into algorithm-legible metrics like CTR and watch-time, and to keep the audience watching.

The primary tool for capturing and governing audiences is the YouTube Studio. In the audience tab, creators can access detailed statistics about their viewers: traffic source, geography, viewer age, viewer gender, date, playlist, device type, and YouTube product. Further granularity can be added—for example, creators can see how viewers on different devices engage with elements within their videos such as cards and links. These metrics can be cast within any temporal range they desire; however, experts and the platform maintain that the first 7 days of an upload or a change to a video—such as a new title, description, thumbnail, or tags—is the most important time period for audience engagement metrics. ²⁰ Armed with this audience data and the metrics discussed earlier, creators are taught to continuously monitor whether their content can reliably attract audiences on the platform, and whether it keeps them engaged. Importantly, experts teach creators how to envision their audience as “measurable types” (Cheney-Lippold, 2017) which can be controlled and governed as a population.

Experts teach creators to use metrics to “chase down audiences” and better capture people in their target audience. They show creators how to evaluate the impact of thumbnails, titles, and keywords through metrics presented in the YouTube Studio as well as other third-party analytics tools like CreatorIQ, TubeBuddy, and Morningfame. Experts also encourage creators to periodically refresh these elements on older content and to A/B-test different thumbnail aesthetics, different titles, and keywords on new content, too. As Crawford argues, the goal is to produce “good” data: he theorizes that if your content has a high CTR, good retention, and steadily growing watch-time, then this must mean that you provide value to the audience and effectively capture them. ²¹ Interestingly, experts emphasize that this careful calculation relies on judgment skills that come with time and experience: creators must balance attention grabbing with the “value” of the content. Otherwise, creators might create “clickbait” which produces “bad data” and provides poor value to the audience, which would be indicated by a high CTR and low audience retention.

If done correctly, experts argue, creators can escape the algorithm’s control by building audiences for the platform. Sunny Lenarduzzi ²² —a digital marketing and entrepreneurship expert with 500,000 subscribers and over 26 million channel views ²³ —suggests that creators can achieve algorithmic favorability by funneling their audience from external sites to their YouTube channel, thus increasing offsite traffic metrics. Lenarduzzi theorizes that this funneling signals to the platform that your content is so valuable it draws off-platform traffic, which makes it more likely to be served by search, discovery, and trending algorithms on the platform. Put differently, algorithmic lore videos propose the idea that creators must produce their own calculated public (Gillespie, 2014) rather than leave audience production to the algorithm. Instead of their content being served to an audience of potential viewers by the algorithm, creators are urged to define the scope of their audience for the algorithm and extend this beyond the confines of the platform to increase traffic to the platform.

Once an audience has been captured, creators are encouraged to govern their viewing practices. As Eves argues, this can be accomplished by designing content which produces valuable audience data or “natural data points” for the platform:

You want to be able to create natural data points for your video and these natural data points are very, very powerful now. If you’re thinking about [it], YouTube’s trying to predict what the viewer will do next. If you help guide the viewer what they’re going to do next, then that helps you as a content creator like really well because you’re creating a natural path where people are able to go . . . And so, you want to create those natural data points because that’s what YouTube’s looking at, [which] is what the viewer does. And when you actually get a higher percentage of viewers doing certain things, that’s great. ²⁴ (emphasis added)

Eves encourages creators to actively guide audience members toward certain actions which produce valuable algorithm-legible data. In other words, Eves is telling creators to predict the platform’s prediction algorithms. Creators can make these “natural paths” in a number of different ways: making playlists to funnel viewers toward their own content; referencing other videos they have made in their content; embedding cards to guide viewers to referred content; breaking up long-form content to make organically connected series; optimizing keywords, title, and thumbnails to increase the odds of appearing in the suggested or the auto-play tab of popular content; and using “calls-to-action” to spur audience engagement such as subscribing, liking, and commenting on the video. As one expert suggested, testing out “hyper-specific calls-to-action”—such as “tell me in the comments down below which of the two optimization strategies you’ll be trying first” rather than “like, subscribe, share, and leave a comment below”—and seeing how they correlate to viewer metrics is one way to encourage specific, algorithm-legible engagement from viewers. Curiously, algorithmic lore videos describe these highly constructed attempts to predict prediction algorithms as “natural” or “organic.” There are striking parallels to how neoliberal free-market theorists draw on natural science metaphors to make the constructed free-market seem natural (Mirowski, 2014). By framing these data in this way, algorithmic lore videos normalize the active governance of viewers by creators and the constructedness of the platform’s visibility markets.

By conceptualizing algorithmic lore videos as market devices, we see how they encourage creators to take an active role in building and governing audiences for the platform. They are taught to see audiences as a resource to capture both on- and off-platform. Creators are also taught to anticipate the audience on the platform and establish “natural pathways” which nudge them toward activities which produce the “right” data—that which shows the platform that their content is potentially valuable to other viewers. A market devices approach adds further dimensions to what Cheney-Lippold (2011, 2017) has referred to as the “soft biopolitics” of algorithmic systems. As market devices, algorithmic lore videos teach creators to quantify audiences and modulate what their audience sees to encourage behavior which produces valuable algorithmic signals. Through algorithmic lore videos, creators are taught to develop and enact dynamic categories as they “follow” the audience, and to self-correct to ensure that they consistently capture and govern their audience members to consume more content on the YouTube platform.

“Blame your content, not YouTube”: moralizing toil and data

To this point, I have discussed how algorithmic experts’ lay economic theories metricize content creation and audience management. Creators are taught how to calculate the value of their videos, adjust content production to data, and how to produce their audience by capturing and governing them based on platform metrics. However, algorithmic lore videos also moralize content creation by developing moral frameworks which discipline creators and justify the difficulties of their work. In economic terms, experts tell creators how YouTube’s labor markets should operate.

Algorithmic lore videos enroll creators in what Cotter (2018) terms the “visibility game.” This term refers to the back and forth between the companies which control the technical systems of the platform and users who push against platform policies to develop strategies to attain visibility. Platforms often articulate normative frameworks to discourage certain strategies by framing them as “gaming” (Petre et al., 2019; Ziewitz, 2019). In the context of YouTube, algorithmic experts also play a role in the propagation of moral frameworks about “good” and “bad” content (Bishop, 2020). In the videos I analyzed, experts typically articulated rules about “good” and “bad” visibility practices when discussing how to cope with and who to blame for invisibility. As Eves argues,

But ultimately, I want everyone to acknowledge that they need to do this: they need to blame your content not YouTube for the issues that you’re facing. And I truly do believe this, that the content that you’re creating is the issue of why it’s not getting promoted. Now there are some anomalies like false positives ²⁵ and we can go into a whole discussion on that. I’ve seen false positives for years. I’ve experienced it myself, but really as a whole you know creators need to stop blaming YouTube and start blaming their content because their content’s not performing with the viewer or the types of viewers that are consuming the videos. ²⁶ (emphasis added)

For Eves, this was the central message of his hour-long video: creators are to blame for algorithmic invisibility, not the platform. Creators are encouraged to take an active role in their self-improvement in ways that uphold narratives of platform meritocracy. For example, Eves notes that he finds enjoyment in audience development as it is like “playing a video game.” For him, the platform presents challenges, and the good creator is one who can work within the limits of their rules to solve these challenges as they emerge. Thus, despite appealing to the desire to “beat the algorithm,” algorithmic lore videos instead advocate for long-term experimentation and conformity with the platform. This firmly entrenches the “good creator” as one who toils to learn the algorithm through a continuous, incremental path of self-improvement.

Who then are the “bad” creators in the eyes of algorithmic experts? Like YouTube, algorithmic experts frame certain practices as morally good, and others as off-limits. Consider, for example, how Roberto Blake—a business coach with 489,000 subscribers and 32 million channel views ²⁷ —discusses methods that artificially increase visibility:

Sub4Sub and all these little hacks, they’re not going to help you and they’re not going to help you in the long run and people who do them don’t stick around. It doesn’t work. It’s not going to work going forward. YouTube is relentlessly enforcing their policies and it’s rough and I get it and you might be a small YouTuber and you feel hopeless but the thing is, you just have to close your eyes and put in the work over and over. If this is something you love, then just do it and do it for the love of the game until you figure out, until you figure out how to do it a little bit better and that comes with time and that comes with work and that comes with patience, just like a sport. ²⁸ (emphasis added)

Sub4Sub is a tactic where creators exchange subscriptions to artificially increase their subscriber count. Often, Sub4Sub offers can be seen in the comment sections of popular videos. As seen in this quote, Blake discourages tactics like these by framing frustration as a natural—and necessary—part of creator development. Paradoxically, he employs sports metaphors to argue that creators need to be “doing it for the love of the game”: in his view, the better moral pay off is found at the end of a stable path of progression rather than through quick spurts of relevance and visibility.

Algorithmic lore videos do double duty. First, they enroll creators in “the visibility game” (Cotter, 2018). Experts teach creators how to pursue optimization, and how to distinguish between “good” and “bad” optimization strategies. Through sports and video game metaphors, they give meaning to the toils of content creatorship; while others game the system through methods like Sub4Sub, good creators are putting in the hard work to learn how the platform works and solve the challenges thrown at them. These meritocratic ideals of incremental self-improvement allude to the second duty of algorithmic lore videos: they justify the moralizing force of YouTube’s market infrastructures. Algorithmic lore videos teach creators to “blame their content, not YouTube,” and that their placement in the market is the result of their content not producing the right data. Again, algorithmic experts resemble neoliberal free-market theorists as they describe “good” data as “natural” or “organic,” and “bad” data as “artificial.” By inferring that success follows “good” creators who toil to produce “natural” data, algorithmic lore videos do important work to shape YouTube’s labor markets. They naturalize the aspirational labor (Duffy, 2017) of creators, and encourage continued work to improve oneself and one’s content even when it does not pay off.

Cui bono? How algorithmic lore videos benefit YouTube

Algorithmic lore videos make YouTube easier to govern in the following four ways: by legitimizing platform narratives of algorithmic objectivity, by teaching creators to calculate the value of their content and format it to fit the platform, by encouraging creators to build and govern audiences for the platform, and by giving moral weight to the difficulties of content creatorship. In this section, I examine how the lessons of algorithmic lore videos are economically productive for YouTube.

First, algorithmic lore videos help legitimize narratives of platform meritocracy by framing the platform as a rational and knowable market for visibility. As lay economists of the platform economy, algorithmic experts transform the complexities of the platform economy into abstract models. Models like Crawford’s emphasize that YouTube’s platform economy is composed of rational actors engaging in ordered and calculable interactions, and thus fundamentally knowable. This framing infers that since the algorithm is rational, creators must not be providing the right value to the viewer or demonstrating it to the algorithm. What results is a downloading of responsibility from the platform to individual creators, and an entrenchment of meritocracy as how the platform actually works. YouTube benefits from reduced scrutiny, and by increased levels of conformity with the platform’s data infrastructures.

Second, algorithmic lore videos benefit YouTube by developing calculative agency among content creators. Algorithmic experts encourage creators to learn how to use YouTube’s economic metrics to evaluate and adjust their content, in order to adapt to a market which is constantly changing. This is reinforced through speculation techniques such as Johnson’s notification hack. These videos encourage creators to make their content contingent on the data flows provided to them by YouTube, which ultimately works to the platform’s benefit. It teaches creators to anticipate market trends; when creators emulate popular creators and the traditional media industries, this increases the overall supply of content related to popular trends on the platform. This high-value content can help to attract audiences and keep them on the platform. In turn, this creates more opportunities for YouTube to generate valuable advertising revenue and user data from each user. It also ensures that creators are self-correcting; if they are constantly refining their content in response to YouTube’s data flows, this decreases the amount of work that the platform must do to discipline creators.

Third, algorithmic lore videos teach content creators how to build and govern audiences for the platform. YouTube benefits by having content creators do the work of building, governing, and retaining audiences, to the point that creators recruit viewers off-platform to watch content on YouTube. This has tangible economic benefits for the platform. Well-defined and governed audiences increase the number of viewers and watch-time—and by extension valuable advertising revenue. It also simplifies the task of matching ads with viewers and creators by homogenizing audiences and grouping them by shared interests.

Fourth, algorithmic lore videos moralize the metricized path to content creatorship. Optimization work is affectively charged, and creators are grappling with frustration as they confront the radical uncertainty of the platform and its algorithms. As market devices, algorithmic lore videos teach creators to “blame their content, not YouTube” and to “hack” the algorithm in ways that maintains—rather than undermines—YouTube’s control over creators. Overall, YouTube benefits by shifting the responsibility for coping with uncertainty onto content creators, naturalizing the instability of the algorithm as a necessary part of the job of “YouTube Creator,” and by keeping creators engaged in the translation of eyeballs to tangible monetary value for YouTube even when it does not pay off for them.

Do creators benefit from the advice of algorithmic experts? It is important not to discount creators’ enjoyment of making and uploading content to the platform and their desires for visibility. Many algorithmic experts see themselves as working to empower fellow creators. However, experts’ good intentions do not contradict my argument that the platform stands to benefit the most from the circulation of algorithmic lore videos. On the contrary, one may argue that algorithmic lore videos are so effective precisely because their intent to provide useful advice to fellow creators to help them cope with uncertainty in these algorithmic systems is genuine. However, while creators may sometimes benefit from algorithmic lore videos, it is reasonable to assume that overall, these videos always benefit the platform.

Conclusion

I have examined algorithmic lore videos and theorized the sociotechnical form they take within YouTube’s algorithmic systems. I argue that these videos are market devices: as lay theories of the platform economy, they provide models for economic action on YouTube. Algorithmic lore videos legitimize platform narratives of algorithmic objectivity and meritocracy by theorizing the platform as a rational and knowable market for visibility. By teaching creators to analyze and interpret metrics, they develop creators’ calculative agency in line with platform rules. Through data, creators are taught how to format their content to be marketable, how to capture and govern audiences, and to find meaning in the often frustrating and heavily metricized path to success. Importantly, algorithmic lore videos do this without actually undermining the platform’s power. Thus, while algorithmic experts claim to “know” and merely describe the YouTube algorithm to their viewers, they are actually performative; they help enact and structure the platform in a way which is economically productive for YouTube. A market devices approach allows us to conceptualize algorithmic lore videos as sociotechnical disciplinary devices rather than mere ill-fated attempts to understand the platform’s algorithms. By attending to the videos and theories produced by algorithmic experts, we see that despite the speculative and partial nature of their knowledge, the attempts of algorithmic experts to capture how the algorithm works may in fact play a role in shaping how it actually works.

In closing, three points merit further investigation. First, algorithmic experts are also subject to the same data infrastructures as the creators they pitch their theories to. Future research could examine the production processes of algorithmic lore videos—especially how algorithmic experts “keep up” with the latest changes to the algorithm. Second, this article has examined how the models of algorithmic lore videos can be consequential when their lessons are taken up by viewers. While the videos described here have broad reach, whether users actually take up these lessons is another matter. Future work could draw on interactions between algorithmic experts and users such as video comments to examine the extent to which algorithmic lore shapes the practices of individual users. Third, the use of machine learning to train platform algorithms raises interesting questions about the mechanics of platform governance through user-generated sociotechnical devices. As content creators engage with and adopt the practices espoused by algorithmic experts, the data that they generate in selecting, watching, and engaging with algorithmic lore videos are used to train recommendation and search algorithms. Algorithmic experts then interpret viewer data and feed their insights back into the community through future algorithmic lore videos. Future work could further unpack the mechanics of this contingent, distributed, and recursive form of governance on YouTube and how it is effected through user-generated content.