People write “unalive” instead of “die” — not to be clever, but because the algorithm removes the real words.
Literal meaning: Algospeak is a coded language built to do one specific job: stay legible to the human reader while staying invisible to the algorithm scanning for it. It is a self-censorship phenomenon: users adopt these coded expressions to evade content moderation that may be real or only imagined — the adaptation happens whether or not the algorithm was actually going to flag the word, because the cost of guessing wrong is high enough to shape language pre-emptively. Three techniques recur: replacing letters with numbers or symbols (“seggs” or “s3x” instead of “sex”), substituting code words or euphemisms for sensitive topics (“unalive” for suicide or death), and deploying specific emojis to signal restricted concepts without triggering keyword filters. “Le$bean” for lesbian. “Mascara” for massacre. It only works as long as both conditions hold at once: readable to people, unreadable to the machine.
Origin: The term was coined by internet culture researchers and popularised through journalism from around 2022, as TikTok’s content moderation became a documented shaping force on language — most visibly in Taylor Lorenz’s Washington Post reporting, which catalogued substitutions like “unalive” for suicide, “SA” for sexual assault, “corn” for porn, and “le dollar bean” for the clitoris. Recent studies on TikTok show how these coded expressions allow creators to remain understandable to human audiences while navigating opaque algorithmic systems. Linguists such as Lauren Squires provide the broader framework for understanding how internet language becomes socially recognizable, while recent TikTok studies document algospeak directly as a strategy for evading algorithmic moderation.
Language actively shaped by content moderation — where the vocabulary in use on a platform reflects the boundaries the algorithm enforces, not just what users want to say.
The Appeal: Algospeak enables communication about suppressed topics — mental health, LGBTQ+ identity, sexuality, political content — in contexts where direct expression is prevented. For marginalised communities in particular, it preserves access to platforms they depend on: “unalive” allows discussion of suicidality without triggering suppression of crisis support content itself. It creates community through shared code: knowing the substitution signals membership. It is also creative: the language is playful, inventive, and evolves faster than moderation can follow.
The Friction: Algospeak reveals what the algorithm suppresses, and the suppression is not neutral. Content Moderator decisions — and the training data of automated systems — determine which topics require circumvention. LGBTQ+ vocabulary, disability language, and discussion of mental health are documented to require more algospeak than, say, financial or political content, revealing whose speech the moderation architecture treats as risky. The euphemism has a cost of its own: mental health crisis language and harm reduction information, once coded, become harder to find for the people who need them and harder to study for the researchers who track them. Recommender Systems shape the conditions: demotion rather than removal is often the mechanism, so creators code their language to avoid the penalty without triggering outright removal. The code also fails: when moderation catches up with a substitution, the community must invent another — an ongoing linguistic arms race that ends, for some, in Deplatforming anyway. AI Literacy is what the vocabulary amounts to in practice: a working model of the classifier, built collectively from what it punishes rather than from anything the platform discloses. That is also the limit of it, because a model inferred only from effects is always one moderation update behind.
Why This Matters: Algospeak is a living record of what platforms suppress. Once you can read it, the vocabulary in use on a platform is also evidence about the architecture’s values — a map of what kinds of speech it treats as normal, and what it pushes to the margins.
Related terms: Content Moderator · Recommender Systems · Deplatforming · Ragebaiting · Coded -Pilled · AI Literacy · Brain Rot
Read more:
- Internet ‘algospeak’ is changing our language in real time, from ‘nip nops’ to ‘le dollar bean’ — Lorenz, T. (2022). The Washington Post
- How Social Media Platforms Shape Language — Squires, L. (2022). Social Semiotics
- You Can (Not) Say What You Want: Using Algospeak to Contest and Evade Algorithmic Content Moderation on TikTok — Steen, E., Yurechko, K. & Klug, D. (2023). Social Media + Society
- How Algorithm Awareness Impacts Algospeak Use on TikTok — Klug, D. et al. (2023). CHI Conference on Human Factors in Computing Systems
- “They Edited Out her Nip Nops”: Linguistic Innovation as Textual Censorship Avoidance on TikTok — Calhoun, K. & Fawcett, A. (2023). Language@Internet
- Enregistering Internet Language — Squires, L. (2010). Language in Society
- Algospeak: How Social Media Is Transforming the Future of Language — Aleksic, A. (2025). Penguin Random House
- The Internet Is Making Us Fluent in Algospeak — Dingfelder, S. (2025). Scientific American