Are Censorship Algorithms Changing TikTok’s Culture?

How social media users push back against the policing of speech

When I was a kid, my mother would speak in pig latin whenever she needed to have an adult conversation in front of my sister and me. Because I didn’t understand pig latin, I viewed these exchanges with childlike wonder, as a secret adult language or code that I could never replicate or comprehend.

In watching TikTok videos, I have started to notice another strange type of language use proliferating on the app. Unlike the freedom of expression that I saw in previous months and years, many users have begun to perform a kind of doublespeak, in which specific words and topics are discussed using abbreviations, euphemisms, and even omissions and pauses, where audience members are meant to fill in the blanks using context clues.

Like parents and adults talking in front of a group of small children, these combinations of nonverbal communication, symbolic writing, and coded language are commonly used to discuss topics that might be considered taboo or NSFW. And, although TikTok is known as an app for children and teens, this use of coded language by users of the site is not just an attempt to protect the eyes and ears of the youth. This use of complex communication styles and techniques is one of many ways that users are able to fly beneath the radar of algorithms that flag videos containing certain words, tags, or popular talking points.

While many users express their disdain for having to perform these verbal gymnastics when sharing important, or even just impetuous, information and data, I find these videos to be fun and fascinating examples of how humans have adapted to computers. Similar to my research on YouTube, this shift towards using coded language for the most mundane of words and phrases is quite common online, and it seems that most social media platforms get to a stage in their development when concerns about safety are reflected in the policing of content and creators. Through natural language processing and other methods, social media platforms are then able to streamline the censorship process by teaching algorithms which words and phrases do not meet the established community guidelines.

But, of course, the irony of using algorithms to police the language of users is that algorithms don’t speak the way that humans do. Even with all of the artificial intelligence and machine learning models that have been developed, computers still take a very literal approach to language, which is easily subverted by human users’ development and deployment of coded language. Through clever word play, and counterplatforming methods, social media users are able to use their awareness of censored words and phrases to continue sharing their message, thereby questioning the politics of what topics are considered taboo across cultures.

A Brief Internet History: Digital Platforms and the Policing of Words

In working through the history of the web, and the rise of user-generated content, concerns about the type of content allowed online is a regular topic of discussion. Consequently, social media platforms rely on documents or codes which outline the rules and guidelines for being a part of the community of users within a specific app or platform. However, these guidelines tend to be quite general, with only the most egregious of violations or content being explicitly stated as banned, viz., harassment, graphic violence, nudity, etc.

Therefore, many users have little information about what words they are, and are not, allowed to use online. But, many content creators have done the work of reverse engineering the reason why certain types of content or profiles are removed, through testing the limits of those guidelines. Within platforms like YouTube, creators have even gone the extra mile of compiling documents with the consequences of using certain words and phrases, with punishments ranging in severity from flagging and demonetization to outright removal and suspension from the site.

Similarly, TikTok users have also come up with their own tips and tricks for working around the unwritten guidelines of the apps. Over time, I have observed several different ways that users get around the problem of having a word banned or heavily policed on a platform. One method is the use of abbreviations to talk about a topic; this is quite common for more serious subjects. For example, instead of using the terms for different types of trauma and abuse, “domestic violence” might become “DV,” and rape or sexual assault is discussed as “SA.”

Another method is the use of a synonym or adjacent word that has the same meaning of the original word, but perhaps puts a different spin on it which can trick the algorithm. For example, instead of using the phrase “commit suicide” on the app, many users commonly use the phrase “un-alive myself” instead. These types of phrases and abbreviations also act in the same way as trigger warnings, allowing certain topics to be discussed without the larger community understanding what the creator is discussing.

This creative use of speech and culture is even more interesting within Black and/or Queer communities on the site, many of which have already mastered the coded language of African American Vernacular English (AAVE). Through AAVE and other forms of unique slang, topics such as gun violence and drug use can be freely discussed on TikTok without being noticed by the algorithm, which is more than likely not trained in such culturally specific language use.

The Dark Side of Subverting Censorship

Despite the fact that this online censorship of specific words is employed in order to decrease the amount of video content reflecting topics that are deemed unsafe or inappropriate (which may or may not be true from a quantitative perspective), what also happens is the creation of communities that coalesce around the newly coded language or communication methods created to represent, or stand in, as a signifier for the banned word.

From the rise of incels and the “red pill” rhetoric to “anti-vax” conspiracists, there are many examples of how words and phrases are created and used in a way that allows for the growth of subversive or underground communities. Many of these communities reflect problematic beliefs, but are able to fly under the radar of censorship algorithms and community guidelines before their language use gains traction and/or critique within the public sphere.

Even before the rise of digital media platforms, these same practices were being employed by internet users who were attempting to subvert the censorship of keywords and search engines. For example, the popular usage of the word “ana” as a stand-in for “anorexia” launched the popularity of “pro-ana” websites, offering resources glorifying and encouraging the culture of eating disorders during the early 2000s.

Even today, abbreviated words and phrases like “thin-spo” are still used to reflect subcultures that spread problematic information online under the guise of self-help or encouragement. This pro-ana rhetoric continues onto platforms like TikTok, where censorship algorithms have a harder time policing this type of content.

The Cycle of Censorship Glitches

In addition to censorship algorithms being less than successful in stopping, or even recognizing, the problematic or harmful information and data they were designed to keep from spreading online, censorship algorithms are notoriously glitch-prone. In the article “Welcome to TikTok’s endless cycle of censorship and mistakes,” Abby Ohlheiser interviews Casey Fiesler, an assistant professor of technology and ethics, to analyze the cycle of TikTok’s corrections to their algorithm related to content censorship. Whether it be removing videos that mention the word “Black” or other forms of racialized terminology, the algorithms on video-sharing platforms regularly put creators of color at a disadvantage when it comes to creating and sharing content about their own culture and identity.

However, as scholars like Ruha Benjamin have stated, much of what technology companies and techno-utopianists would like to describe as glitches are in fact examples of algorithmic systems acting exactly as they should. Similar to the role of racism and other oppressive codes within modern day society, many algorithms are reproducing models of discriminatory design which don’t take into account the unique issues and needs that come with having a diverse and intersectional grouping of communities engaging in online discourse.

What’s more, lawsuits such as the case of LGBTQIA+ content creators who unsuccessfully sued YouTube for discriminatory practices (due in part to its machine learning algorithms) also demonstrate that not all censorship is accomplishing the good that it was created to do. Instead, many communities that are already intersectionally marginalized within offline spaces are experiencing the additional threat of marginalization from algorithms that have not been taught to be comfortable with topics or language outside the norms of White and cisheteronormative culture.

I will also note here, that despite the fact that TikTok continues to censor content about topics of racial justice and social equity, it rarely censors popular music or sounds (which can be full of inappropriate and graphic language and imagery in English and many other languages). In this sense, while topics such as Black Lives Matter and Gender Affirmation may underperform on the site, as the algorithm can block those topics from appearing on the For You page, songs about topics that break the platform’s written and unwritten rules can do quite well. For example, while a user wouldn’t be able to talk freely about using marijuana or other drugs, songs about Mary Jane and dances or trends which reference partying and drug use are allowed to thrive on the site.

Exemplifying many of the frustrations that Black creatives have with TikTok, and social media platforms in general, censoring the sharing of political or culturally relevant content while supporting more lighthearted content is similar in method to tone policing. While Black creatives are free to use any words to express themselves when entertaining users through song and dance, more serious topics tend to garner more oversight and policing by algorithms and the politics of the platform. In this sense, Black creators can only be seen and heard in ways that the algorithm (thereby, the societal norm) deems to be acceptable, otherwise their content is shadow-banned or removed.

Human Overseers and the Complexity of Language and Culture

With that being said, the impact that censorship algorithms have had on TikTok culture and content also demonstrates the importance of human overseers when it comes to the content that is allowed on a social media platform. Although many people do not believe human overseers are an effective check on machines, they can do the work of ensuring the decisions made by machines reflect the realities of how a human would respond to that same content. For example, if a machine flags a video or content as violating community guidelines, a human overseer would also review that recommendation to analyze the machine’s reliability when it comes to censorship.

At the same time, we know that human oversight does not always make things easier for intersectionally marginalized communities based on the way that specific cultures and traditions are interpreted by those outside of those communities. Therefore, it will be important to not only train algorithms to better understand humans, but to also train the humans who oversee algorithms to have a better understanding of different cultures and language use. Then, perhaps, content creators can spend more time sharing important information and making great videos, and less time trying not to get flagged.

As social media is an especially popular tool for activism and resource sharing, the fact that many important topics of cultural and social concern can only be discussed using coded language and nonverbal communication says a lot about how we view these causes within society. As many people already walk on eggshells when discussing issues of race, gender, class, and sexuality within society, the deployment of social media algorithms reflects those same views. Yet, despite these checks on freedom of expression and justice-oriented posting and resource sharing, TikTok creators demonstrate their commitment to countering the platform by creating safe spaces or niches within the site to have serious discussions in addition to dancing.