Coded to Exhaustion: The Hidden Cost of Outsmarting the Algorithm
Welcome back to PsyberSpace. I'm your host, Dr. Leslie Poston. Today's episode is a little different, and I am so excited to share it with you. Usually, I'm reviewing other people's research, but today I'm talking about some of my own. This particular research was the focus of my doctoral dissertation work, And I think many of you will immediately understand why I think it's important to talk about today.
Leslie Poston:Note that this is a high level overview of my research. There are detailed articles out for peer review as we speak. Let's get right into it. If you use TikTok or Instagram Reels, you've probably noticed that people talk strangely about certain topics. They say I'm alive instead of suicide or write the letters SA instead of saying sexual assault, using grape emojis instead of saying the word rape, or saying corn instead of porn and angry mango instead of Donald Trump.
Leslie Poston:One creator in my dataset described the January sixth coup as that kegger that occurred in January, which we'll come back to because it was one of the more sophisticated examples in the entire dataset. None of these are typos and none of them are slang. They're deliberate modifications that creators make to their language to avoid having their content suppressed or removed by automated moderation systems. The practice has come to be called algo speak. Instead of saying a word that might get their video flagged, creators substitute, misspell, fragment, or replace it with something the algorithm won't catch but their audience will understand.
Leslie Poston:I identified six primary types of evasion: lexical substitution, which is swapping one word for another orthographic variation, which is creative misspelling symbolic substitution, or using an emoji in place of words syntactic modification, which is restructuring entire sentences multimodal splitting, which is when you distribute sensitive content across spoken audio, on screen text, and visual elements simultaneously so that no single channel triggers detection And lastly, off platform migration, which is giving up on saying it on the platform at all and just sending people to an external link. That last one is qualitatively different from the rest because it concedes that the platform can't be used for full expression at all. Most writing about algos speak treats it as something the internet spontaneously invented, but it didn't. The communicative logic at the heart of it, saying one thing on the surface while meaning something else to people with the cultural knowledge to decode it has a long documented history. Henry Louis Gates Jr.
Leslie Poston:Called it signifying in his 1988 book, The Signifying Monkey. Geneva Smitherman documented it as a core feature of African American Vernacular English, or AAVE, which she showed as a complete rule governed linguistic system, not slang or broken English and not a departure from standard English. And Sarah Florini bridged that tradition into digital spaces through her research on Black Twitter, showing how those signifying practices moved online. So what mainstream media likes to call TikTok language or Gen Z speak traces largely back to Black American rhetorical tradition. And many people using those terms today don't know that.
Leslie Poston:The racial linguistics framework from Aline Rickford and Ball helps explain why this matters. Language is a site where racial hierarchies get reproduced and contested, and the same innovations get valued differently depending on who uses them. That absence of awareness has consequences that I'll get to later in the episode. I was inspired to study this because it was happening to me, like many research topics. I was trying to have a conversation on TikTok about late diagnosed autism in women, doing something they call a duet where you would talk back and forth side by side.
Leslie Poston:And my content kept getting the sound removed, which is an indicator of suppression. The views dropped from thousands to 10 midway through the conversation, which is not a realistic ratio of who's seeing your content. Then TikTok started adding music over the audio so you couldn't hear me talking. And eventually, the creator I was duetting lost their account entirely because of this topic and had to fight to get it back. What I noticed for myself was that the process of building community despite all of this was making me personally feel more emotionally resilient.
Leslie Poston:So I wondered if other people felt the same way. I'm a creator who discusses sensitive topics across several identities that are frequently targeted for moderation. So I came to this research through indirect experience with the evasion practices I was studying. What I found in my data turned out to be the opposite of my own experience, which meant I had to hold my personal experience completely separate from the empirical findings throughout my entire analysis. To do this study, I combined content analysis with survey based measurement.
Leslie Poston:The content analysis covered publicly available TikTok and Instagram Reels content from a time period including March 2020 through December 2025. My initial scrape of this public content pulled in over 30,000 videos that all matched the criteria of these topics. After randomization, exclusions for brand and bot accounts, and a relevant screening pass, I had a working sample of videos, and I watched every one of them at least twice. Once for coded linguistic analysis and once for behavioral observation, tracking emotional cues and where the same creator appeared across multiple videos, whether those cues changed over time. Some videos I went back to a third and fourth time as the third finding, which I'll talk about later, emerged from the data.
Leslie Poston:The survey covered evasion strategy awareness, perceptions of suppression, and psychological well-being, and a resilience measurement adapted to the platform navigation context. The epistemological positioning is interpretivist and critically realist, which basically means I'm studying people who are already interpreting and theorizing about their own situation, developing folk theories about how platforms work, and acting on those interpretations that they make. What I expected to find and what I actually found were different. And the three findings that came out of my research connect to each other in ways I didn't anticipate. My original hypothesis was that navigating algorithmic constraints creatively would build resilience because people who find workarounds are exercising agency, and that agency should be protected.
Leslie Poston:My data went the other direction. Participants who reported the highest rates of language modification also reported the lowest emotional well-being. So more evasion was associated with lower resilience, not higher. Close to ninety percent of participants reported changing their language always, often, or sometimes. More than eighty percent reported doing it preemptively, even when they weren't sure it was necessary, just in case, which means the algorithm doesn't need to act for its effect to be felt because their surveillance has now been internalized.
Leslie Poston:And over seventy two percent rated their own ability to adapt as only a little good or somewhat good, meaning they're performing a mandatory task constantly, but they never feel competent at it, which is exactly what the chronic stress framework predicts. Robert T. Carter's framework of race based traumatic stress describes what the data reflects. Chronic low level stress that's psychologically costly in proportion to how long it lasts and how compelled it is, regardless of whether any single encounter is acutely traumatic. Thema Bryant Davis's work on the psychological cost of code switching maps the same process onto the experience of perpetually translating yourself for an environment that wasn't built with you in mind.
Leslie Poston:No single instance of typing unalive instead of suicide is traumatic. But the ten thousandth instance, knowing that the next workaround will also eventually get flagged with no recourse and no appeal, accumulates. Strategy learning is predominantly social, which connects to Bandura's social learning theory. The most common pathway to learning these techniques was observing other creators on TikToker reels and adopting their approaches. And that same observational learning process is also what distributes strategies broadly enough to trigger detection.
Leslie Poston:Let me talk about the most extreme example of cognitive demand in my dataset. One creator needed to discuss the January sixth attempted coup on the Capitol, and she couldn't use direct language. So, built an entire parallel narrative that spanned multiple videos. The election became a costume contest. Ballots became lists of tallies.
Leslie Poston:The House of Representatives became all the rule makers. The coup became the kegger that occurred in January. And Kevin McCarthy became California Kevin, for example. That's called syntactic and structural modification, the rarest and highest sophistication strategy type. And it's structurally identical to what Gates described as signifying.
Leslie Poston:Two layers of meaning maintained simultaneously, one for the surveillance system and one for the intended audience. Multiple videos of that level of effort with no guarantee any of it works again next month. One more thing on this finding. Neurodivergent participants reported cognitive drain exceeding seventy five percent, substantially higher than the sample average of about fifty five percent. And the extra cognitive layer required to translate authentic expression into algorithmically safe language compounds barriers that already exist for this group.
Leslie Poston:So any practical agenda that ignores that differential is going to underserved the creators facing the greatest burden. Community connection did partially buffer the emotional cost, which is consistent with research on collective resilience through language. About fifty four percent of participants reported that coded language helped them connect with their community. And one person described feeling more proud of our differences, because of the shared struggle. That doesn't reverse the primary finding, but it tells us that evasion that connects costs less psychologically than evasion that fails or isolates, which should inform how we think about interventions for this problem.
Leslie Poston:The cognitive resilience component of the survey didn't produce interpretable results. And my first read was that evasion and cognitive functioning are just unrelated. But that's not what happened. The survey instrument was wrong for the phenomenon. Cross sectional surveys capture how participants feel about their cognitive functioning at a single point in time.
Leslie Poston:And they do work for episodic challenges, things with a clear beginning and end. But the cognitive experience of sustained linguistic evasion isn't episodic. It's ambient, a continuous background monitoring process that runs underneath all platform communication. So, by the time someone takes a survey, they've normalized this act. And the instrument can't distinguish between someone who's cognitively thriving and someone who's normalized chronic depletion.
Leslie Poston:Ara Naza Ukwiko and Vimuri's distinction between cognitive resistance and cognitive resilience is helpful here. The question isn't whether creators avoid impairment. It's whether they maintain functioning despite the ongoing demands. And the null findings suggest those demands are accumulating rather than being absorbed. So, my research, the appropriate method would be diary based longitudinal research, recording cognitive and emotional states close in time to actual evasion events over weeks or months, and identifying that mismatch between the survey instrument and phenomenon itself is a methodological contribution for my future research.
Leslie Poston:Here's where the cultural genealogy from earlier becomes a little more important. A community under suppression pressure develops a new evasion strategy. And that strategy works because detection systems don't recognize it yet, and because its use is limited enough to fly under detection thresholds. Then the technique spreads. First to adjacent communities facing similar suppression, and then to mainstream creators who find it useful, and then on to the general platform population.
Leslie Poston:As the technique spreads, the algorithm learns to detect it through exposure to flagged content. And the strategy loses its specificity as a community signal. It stops working. And the originators of the strategy must develop something new, bearing the psychological cost of that creative labor again while the adopters move on to the next borrowed innovation without recognition or credit. The cycle continually restarts.
Leslie Poston:Creators in my dataset talked explicitly about terms that had gotten too popular and walked their audiences through the reasoning for switching. And that shows metacognitive awareness of what I have named the neutralization lifecycle. They weren't just developing workarounds. Some of them had developed explicit folk theories about the life cycle of their own strategies and the inevitability of their degradation, which connects to Moran et al. Research on folk theories of moderation.
Leslie Poston:Stuart documented the same pattern as an orthographic arms race in pro eating disorder communities on Instagram, where each spelling variant gets detected and requires further modification. And I saw the same dynamic across multiple topic categories and communities in my data. The racial linguistics framework makes visible that this isn't neutral language spread. Prior scholarship has documented the cultural harm of AAVE appropriation, and my research adds an empirical dimension. Appropriation has a functional downstream consequence because it neutralizes the protective value of the appropriated strategies for the people who created them.
Leslie Poston:That moves the argument from cultural ethics, where it could be dismissed as subjective, to measurable harm, where it can be documented and potentially addressed through platform policy changes. My survey data also revealed something I didn't expect. Creators are using coded language not just to dodge the algorithm, but to dodge targeted harassment. Because by communicating in language only their in group understands, they become harder to find for people searching keywords to attack them. Black women, Muslims, queer users face enormous harassment in their comment sections on TikTok and Instagram.
Leslie Poston:And what I'm observing now post study is AlgoSpeak expanding beyond video content into usernames and bios so that people can disguise who they are online and avoid the organized groups trying to cause them harm. I'm not aware of an existing theoretical framework that holds both the algorithmic evasion and the harassment evasion dimension simultaneously. Building one that can hold the political, psychological, and community protection functions all at once is work this research makes possible but certainly can't finish alone. So what needs to change? Moderation systems that can't distinguish or won't distinguish between harmful content and content discussing harm impose the highest cost on the communities they claim to protect.
Leslie Poston:A mental health educator discussing suicide and a post promoting suicide are not the same thing, but current systems treat them similarly, which is a design choice with measurable psychological consequences. Creators currently have no reliable information about why content is suppressed and no functional path to appeal. TikTok only recently added the ability to view full community guidelines and to appeal violations. But by the time you get to the violation stage, you've already been quietly suppressed for weeks or months with no explanation. If platforms are going to exercise this much governance over our speech, there must be public accountability for whether that governance falls equally across communities.
Leslie Poston:The evidence from my research suggests it does not. None of these changes are technically infeasible. They're institutionally inconvenient, which is a different problem and one that regulatory and policy pressure has a role in addressing. On the education side, digital literacy curricula haven't addressed the cultural genealogy of Internet language. Students who understand where terms like unalive and essay and spicy accounting come from, who developed them and why, and what happens to their protective value when mainstream adoption strips their specificity are better equipped to participate thoughtfully in digital culture than students who treat Internet language as a neutral evolving commons.
Leslie Poston:A few things have changed since my data collection ended in December 2025 that I think are relevant. The sale of TikTok to Oracle did finally go through. So starting in January 2026, the new owners have been changing the algorithm much more frequently. Another thing I'm observing, and again, this is observational, not part of my dissertation research, is that Instagram Reels is now starting to get more original evasion content being created natively for the platform, which is interesting because during my study period, most of the Instagram Reels I collected were just reposted TikTok videos with the watermark still on them. The larger algorithmic fight has also moved beyond social media platforms.
Leslie Poston:There's a global push for age gating and digital ID verification that's being driven by a moral panic over youth online, something we've discussed in several episodes. And Australia, The UK, and several other countries are now requiring digital ID to use social media, with some trying to extend that to requiring biometric identification just to use your personal computer. These surveillance technologies will create similar additional harms, including the compounding risks of identity theft, isolation, and putting some users in physical danger on top of creating new barriers to anonymous and pseudonymous expression. The communities my research studied are the ones who will bear the cost of that expansion first. And that dynamic needs to be examined in light of what this research found in my opinion.
Leslie Poston:One survey participant described catching themselves using AlgoSpeak in private conversation off the platform entirely. That's how deep this internalization goes now. My research asked what it costs to speak under algorithmic suppression, and the answer is concrete and measurable. It costs your emotional resilience in direct proportion to how often you have to do it. It costs your cognitive resources that are getting redirected from expression and community building to self monitoring.
Leslie Poston:And it costs the protective value of the strategies themselves as the ingenuity marginalized communities produce gets absorbed into the mainstream, stripped of its origins, and rendered obsolete, leaving the originators to start over. The people who most need to speak and connect are the ones paying the highest price to be heard. And these communities deserve ongoing scholarship that serves them rather than just describes them. Thanks for listening to PsyberSpace. It was a pleasure to share my research with you today.
Leslie Poston:I have more research on different topics ongoing, and I can't wait to share that with you as it develops. I'm your host, Leslie Poston, signing off. If you want to read my research, I'll come back and link it in the show notes as it's published over the next several months. As always, stay curious, and don't forget to subscribe so you never miss an episode.
Creators and Guests
