>be me
>be 13
>hear the word ‘4chan’ in a youtube video
>open google
>ohboyherewego.jpg
I’ve recently spent a lot of time thinking about my relationship to 4chan. Probably because one of the biggest summer blockbusters of 2026 started there, with a single 640x480 image and a few lines of flash fiction. I don’t have much to say about the movie Backrooms that hasn’t been said better elsewhere, but the imageboard that inspired it is a different story. Between 2008 and 2016, I spent more time on 4chan than I did with any teacher, friend, or family member.
The posts that started it all (2019)
While I lack the kind of big, embarrassing usage numbers available on platforms like Steam and iOS to prove it, I have all the evidence I need in the way I’ve learned to act online. Just as I can trace certain turns of phrase and body language back to my parents, I know that many aspects of my digital and real-life selves come from my years on /b/, /v/, /mu/, /tg/, /co/, and /r9k/. I bumped thousands of threads, I saved thousands of memes, and, most importantly, I read thousands of greentexts.
An exemplary piece of greentext (2014
Greentext is green text. This is thanks to a set of rules in 4chan’s source code and accompanying stylesheet, which cast every line of text preceded by a > in a slightly desaturated green. On paper, that’s all it is—but to this day, few strings of symbols and letters can slot me into a specific way of reading, writing, and thinking as fast as ‘>be me’.
At a time when many writers fear that ‘AI’ is coming to flatten human language into code, it’s tempting to read this as a fairly on-the-nose inversion. Over the last 15-odd years, as machines have gotten good at writing like humans, a subset of human writers have gotten good at writing like machines. They’ve been doing it through a form invented on a 4chan board with the word ‘Robot’ in its name, and one that is now easily copied by machines meant to write like humans writing like machines. However, as a human writer, I don’t think that’s the most interesting thing going on here. In the same way an empty safe is theft-proof, greentext could represent one of the first AI-proof forms of writing we’ve got.
>quotation
During the 2026 Venice Biennale, the art collective New Models presented several pieces in the exhibition ‘Strange Rules’—among them, a poster reproducing the several hundred lines of code on which 4chan was built. The exact provenance of this code is the stuff of internet creation myths, but the generally-agreed-upon story goes like this: once upon a time, there was a Japanese imageboard called Futaba. At some point, someone put the Japanese code for that site through a piece of software called Babel Fish, producing a rough English-language translation called Futallaby. This is what 4chan creator moot used to build his site’s engine, Yotsuba.
With ‘Strange Rules’, curators Mat Dryhurst, Holly Herndon, Hans Ulrich Obrist, and Adriana Rispoli set out to highlight art that operates at the protocol layer—treating things like algorithms, LLMs, and platform infrastructure as subjects to explore ‘the invisible rule systems that shape how culture is produced and perceived’. In the context of this show, Lil Internet’s 4chan source code poster makes a provocative but soundly defensible claim: everything that has ever come out of 4chan began with those several hundred lines of PHP, and can thus be read as a single ‘work’. Within this code is the infrastructure that divides the site into boards, anonymous users, OPs, and repliers—and the foundation for everything that has flowed out of it since it went live in 2003, from Anonymous to QAnon to Zyzz. Critically, it’s also responsible for greentext.
The actual code in question: $com = eregi_replace("(^|>)(>[^<]*)", "\1<div class=\"unkfunc\">\2</div>", $com);
For those like me who never learned to code sans-vibes, the block above is a rule for labeling text—detecting any line that begins with > and wrapping it in an element. This element doesn’t promise the color green or any color at all; that’s decided elsewhere, in a style sheet. The only thing happening here is classification. From a purely mechanical standpoint, the load-bearing part of any greentext isn’t the green, but the > that comes first.
Within the conventions of early online networks like Usenet, > is used to indicate that any subsequent text is someone else’s words. It’s a somewhat inelegant solution to the problem of quoting someone, since it still leaves the user free to write whatever they want. Subsequent generations of internet platforms have developed much cleaner, less ambiguous mechanics. Fortunately, 4chan is no such platform—and in that sloppiness and lack of definition, greentext was born.
The utility of greentext, as distinct from simple Usenet-style quotation, is best understood in terms of a phrase I’ve already invoked. In English class terminology, ‘>be me’ is first-person narration formally marked as a quotation. The frontloaded deniability of > gives the writer a way of quoting themselves—or, more interestingly, a version of themselves held at arm’s length. Things get even more interesting when we consider the words that often follow.
As a direct statement, ‘>be me’ is an imperative — ‘hey, you there. be me’ — and most classic greentext lines are written in a similar grammatical mood. After the > allows the writer to step away, any subsequent imperative verbs push the text squarely into the reader’s arms. Between these two rival grammatical readings — greentext as quotation versus greentext as imperative — the form never arbitrates. This is probably why it lends itself so well to confession, shame, cringe, and outright hate—a perfect form for a site where every user is ostensibly anonymous.
So, when someone gets the ability to type something into a keyboard without actually owning it, what does it do to them? People have been pondering this subject for as long as we’ve had keyboards.
>agency
The Malling-Hansen Writing Ball holds a special place in the history of mechanized writing as one of the first commercially produced typewriters, but it’s more familiar to many people today thanks to the much-debated impact it had on its most high-profile user: Friedrich Nietzsche. By several accounts, after Nietzsche bought the machine in 1882 to cope with failing eyesight and debilitating headaches, it produced an immediate change in his writing—chopping his previously sweeping handwritten prose into the sharp, punchy, ‘telegraphic’ aphorisms he’d later be famous for. When his friend Heinrich Köselitz remarked on the shift, Nietzsche offered what would become one of his most-quoted lines about technology: ‘our writing tools are also working on our thoughts.’ Few topics continue to beguile media theorists more than the question of whether the above statement has any truth to it.
A writing ball from 1878
The most hardline determinist answer, that our tools don’t merely shape our writing but write us, is widely credited to another German Friedrich. Dissenting from humanists like Marshall McLuhan — who granted media the power to reshape human faculties at a sensory level but never stopped treating humans as the thing being extended — Friedrich Kittler cast the writing ball as the moment media stopped being a prosthesis, setting off a cascade of events that would eventually leave us deprived of any agency at all. In Gramophone, Film, Typewriter, `Friedrich Kittler argued that while pre-typewriter handwriting had been an extension of the soul or the individual, the writing ball turned writing into a mechanized process. Rather than expanding our capabilities to write, it imposed a new set of constraints for us to write within—turning the formerly expressive human into an operator only capable of executing what the rules permit. (RIP FK, you would have loved Claude.)
Kittler gets flattened into a caricature often enough that leaning on the hard-determinist version of his views risks being unfair, but it presents the exact allegations greentext has to beat. If the tool really does write us, greentext should be a closed loop: conditioning whoever uses it, running on whoever reads it, and claiming any output as its own. This may give hardware too much credit. In a 2015 feature for Texte zur Kunst, pointedly titled ‘Do Media Determine our Situation?’, Isabelle Graw, Reinhold Martin, and André Rottmann converge from different directions toward a new perspective on Kittler’s legacy: hardware is influential, but not dictatorial. We might extend this to say that on its own, hardware can’t account for what people use it to generate—and greentext is a near-perfect demonstration of that gap.
In theory, greentext shapes the people who use it to write in more or less exactly the ways Kittler imagined, and its imperatives turn the reader into a certain kind of operator every time it’s read. However, this happens on two distinct levels. First is greentext, the > mechanic—which, as far as conditioning hardware goes, is pretty weak. Where a typewriter makes hard demands of its user — type like this, not too fast, rewrites will cost you — > demands nothing and offers only one thing in return: it will indent whatever you choose to write. Second is greentext, the convention—which has a lot more demands, and more teeth to back them up. Just as the typewriter enforces its terms mechanically — write my way, or I will jam — greentext, the convention, is enforced socially: one beat per line, imperatives and declaratives only. Posts that don’t adhere to the form get mocked, or worse, don’t get read at all. This is the aspect of greentext that really checks Kittler’s boxes.
From this, it becomes clear that the real trial isn’t whether greentext conditions the people who engage with it — clearly, it does — but rather how the convention that conditions them came to be. Here — rather poetically, in the source code of greentext itself — the Kittlerian loop springs a leak.
"\1<div class=\"unkfunc\">\2</div>"
The people who wrote Futallaby and the Yotsuba engine carried the same >-quotation mechanic forward—and across two iterations of code, they left the class name the same: unknown function. To be clear, this is almost certainly a junk placeholder phrase and not worth projecting any authorial intent onto; in the sense that matters to a coder, > was known to mean ‘quote’ long before the first chan site. But the thing we now know as greentext has another function—and in 2003, that function was genuinely unknown because it hadn’t been invented yet.
>inheritance
The strongest evidence that greentext is more human than hardware can be found in the mechanic’s history, starting with an accidental near-controlled experiment that the internet seems to have run for us. Thanks to the shared code that makes 4chan work, Japanese imageboards had the mechanic’s ingredients first.
While sites like Futaba Channel featured green quote lines as early as 2001, and earlier textboards like 2channel used > as a quote convention before the turn of the millennium, even without the styling, there’s a conspicuous absence of ‘>be me’-style posting on these platforms before 2010. Given the same potential to hold versions of themselves at a distance, Japanese imageboard posters developed a completely different visual-textual culture—one with much more emphasis on elaborate text drawings and faces built from Shift-JIS characters. The few attempts at capital-L Literature to come out of this space in the 2000s — like Densha Otoko (Train Man), a 2004 book built from a thread on 2channel’s Single Men board — don’t lean on >-prefixed lines at all. There’s also a noteworthy absence of imperatives: while the Shift-JIS tradition of Japanese imageboards asks you to look at a figure someone else painstakingly assembled, greentext commands you to be the figure.
This is not to say that 4chan users are a bunch of Very Special Geniuses who cracked the meta of internet flash fiction in a way that Futaba users simply couldn’t, nor does this constitute a controlled experiment in any serious sense. Y2K-era Japan and post-2008 North America and Europe are very different environments, and any of those differences could be doing some heavy lifting here. The only thing we know to be constant across both platforms is the source code—which, by Kittler’s hardliner logic, suggests that greentext should have appeared on 4chan and Futaba. It didn’t.
Of course, a true Kittlerian might contend that Japanese and English are also hardware that conditions their human users—but Usenet had English and > for two decades and never yielded anything like greentext, either. This is probably because commitment deniability works even better with identity deniability as a foundation. Every Usenet post came with a name and email address, and the deniability of > is worth little when everyone can see exactly who’s doing the denying. On 4chan, the > was never operating alone—it stacks a second layer of deniability on top of the one the platform grants every poster by default. But before anonymity gets promoted to the real hero hardware of this story, it’s worth remembering our alleged control group: Futaba had full anonymity and the >-mechanic, and still produced no greentext. Even the complete stack of affordances is not enough to dictate what people build on top of them.
It’s also worth revisiting the folk history of 4chan itself. Most people place the origin of greentext around 2010, roughly six and a half years after the >-to-unkfunc mechanic shipped with the site. This lends even stronger credence to the theory that greentext is more human than hardware; if it were a technical inevitability, it’s unclear why it would have taken the better part of a decade to emerge. Returning to Graw, Martin, and Rottmann: the original >-mechanic is the hardware, but the years of posts that made greentext into >greentext are the human contribution that brought the form into being.
>code
Like many rigid literary forms, it’s possible to write a good deal of greentext’s rules down as a list—and in the mid-2010s, a GitHub user named jz-feng attempted just that. ‘Greentext’ is an esoteric programming language built around the key terminology of greentext, allowing users to write software in the language of 4chan posts. Terms like >implying and >inb4 act as control structures, and the entry point — the instruction that boots the program up and sets it running — is ‘>be me’.
A block of Greentext code looks like this:
Everything in this program hinges on the initial number after ‘be replies like’. This is used to test which of the two conditional messages the program’s user receives. If the program is run with the number left at zero (>implying replies is 0), it will print (>mfw) ‘no one replied’ and stop—but if the number is anything other than 0 (>or not), it will print ‘someone replied’ instead. It’s activated with >be me, and ended with >thank mr skeltal. If it weren’t for the visible code scaffolding, it would almost work as a readable greentext.
While this project was clearly meant as a joke, it’s also a rather literal realization of Kittler’s fears: the transformation of a human-defined expressive style into a rigid, algorithmic syntax. Still, the fact that this transformation was done by humans is worth noting—and this isn’t limited to the one human behind the GitHub account. Long before jz-feng’s code, people were treading statements like >inb4 with a curiously computational logic. To post >inb4 is to anticipate and handle a predictable reply before that reply arrives—something programmers call ‘exception handling’, a construct used to manage unexpected events. jz-feng’s keywords weren’t arbitrary mappings of board slang onto programming concepts. Rather, the code transcribed a set of protocols the board had already defined for itself. The posters behind greentext didn’t just let the tool work on their thoughts; they had been feeding the form back into the machine all along. The only thing jz-feng did was spell it out.
This cycle — from human experience to digital form and back to executable code — is where my mind goes whenever someone starts talking to me about the death of writing in the age of AI. If a form like greentext can bloom on an allegedly overdetermined platform like 4chan, it suggests a meaningful human contribution that the hardware can condition but never fully dictate.
Up until now, we’ve mostly been discussing greentext in terms of structure—whether the form shapes the people who engage with it, where it came from, and who deserves more credit for what it’s used to make. However, focusing only on the theory of the form misses the messy, human reality of how that form is actually used. It’s one thing to ask whether greentext conditions the way you write, and another thing to ask whether it leaves you able to mean what you write.
>post-irony
The earliest forms of greentext, which began appearing on the 4chan board /r9k/ (short for ROBOT9001, itself a play on the board’s anti-spam mechanic, ROBOT9000) around the turn of the 2010s, were framed as confessionals. The people behind those confessionals are worth acknowledging: by the late 2000s, the board was the de facto home for an undersocialized, downwardly mobile cohort of young men who’d been trained to see their own views and habits as alien. Fueled by subject matter too raw to be expressed anywhere else, the form quickly became notorious for facilitating some of the most naked writing on the internet—and what made every post postable was the patina of deniability afforded by starting with >. The knowledge that any real confession will sit side by side with shitposts probably didn’t — and still doesn’t — hurt.
These two modes, greentext as a way of working through a poster’s real successes and failures, and greentext as a medium for joking with digital strangers, are inseparable. Sometimes they even show up in the same post, with something ridiculous played straight immediately followed by something straight played ridiculous—and in the hands of a truly skilled poster, it can be difficult to tell which is which. This adds yet another tonal VPN to whatever the writer chooses to write about. Not only are they quoting themselves, but they might be doing it ironically. They could even be pretending to do it ironically.
In ‘E Unibus Pluram’, adult bandanna guy and posthumous internet punching bag David Foster Wallace argued that the exit from the cultural dead end of irony isn’t more irony but a new sincerity that’s willing to risk looking goofy. In this context, greentext offers some striking appeal. Rather than presenting pure detachment or pure sincerity, > is all about the undecidability between them—and whatever else it does, the form’s clearest function is to drop you straight into the writer’s point of view, possibly exaggerated warts and all. This is a literary position with a history that runs much longer than futaba.php, but the advent of greentext has taken it in several fascinating new directions.
With most forms of greentext, deniability is unfalsifiable by design: you can never prove a confession was sincere rather than a bit, and that’s the point. Many greentext posts are designed to be funny, which can produce an involuntary bodily response — laughter — which the > can’t take back. However, laughter is still spectatorial: the reader can always tell themselves they’re laughing at the idea of something, or some unintended aspect of it. The tension between sincerity and deniability reaches its sharpest boundary test in a genre that no one — at least to my knowledge — has done any serious writing about.
Greentext erotica is the one place where the form runs into something that the writer can’t control and the reader can’t dismiss. My reasoning here is crude but simple: out of all the base sensations that make up the lived human experience, arousal is hard to reach from a place of pure ambiguity. Deniability is a property of the writing, and governs what the words commit the writer to—but it has no jurisdiction over what the words do to the person reading them. Somewhere under the bit, in a way that no amount of >-prefixing can obfuscate, a real, un-disclaimable bodily response either happens or it doesn’t. And as the one feeling it, you’re always implicated.
Yes, for erotica this is tame. No, I don’t feel the need to provide any more visceral examples.
In this sense, greentext erotica offers some curious clarification. The > still offers the writer a pass to admit to wanting something the world won’t let them want with a straight face. The form is still what makes the desire sayable, and the writer’s deniability is never once in danger. What it shows is deniability’s limit—the point at which the response crosses out of the deliberately ambiguous text and into hard, biological reality.
Here, again, we must contend with greentext’s use of imperatives—the way it hands itself to the reader to execute. ‘>be me’ is not a standard call for empathy, asking you to feel for the poster the way you’d feel for a character in a story. Rather, it’s a command; for the length of the post, you are ordered to be the writer. This is not to say that empathy never enters the equation—the command is how it gets extracted from a population incapable of asking for it. For the average /r9k/ poster, sincerity-as-request is a non-starter—humiliating out in the world of normies, and social suicide under the board’s famously feral norms. Thus, the form skips the asking entirely and installs the poster in the reader by force.
This is where the feeling of cringe that many people get from reading greentext — and the arousal some people get from reading greentext erotica — develops another level of ambiguity: whose feelings are you really feeling? Through the magic of the form, the writer is able to wash their hands of whatever they’ve provoked in you. They get to keep their deniability, and you’re the one left holding the response, in a physical body that the > has no way to reach into and retract it from.
All of which invites yet another question: is greentext actually good for its writers and readers?
>irony poisoning
In the least-loaded sense of the word, greentext is liberatory. Having a form to say things you don’t have to mean is a meaningful freedom—the kind that the internet’s early utopians sought to extend to everyone. But on a long enough timeline, it’s also a reliable way to lose the ability to mean anything at all.
Irony poisoning is what happens when someone can no longer tell their real beliefs from the ones they profess ironically—reductively, when ‘just kidding’ and ‘not kidding’ curdle into something totally unrecognizable as either. Here, the built-in deniability that makes greentext greentext comes back to collect.
With a literal VPN or proxy network, a 4chan user can conceal their identity from an outside audience. With greentext, they can do the same trick in reverse—the concealment faces inward, hiding things from yourself. Every ‘>be me’ you write allows you to potentially say something true, while keeping open an exit of having only been doing a bit. Done once, this is a joke—but a few thousand times, it becomes something else. For some people, the form drills the reflex of not-quite-meaning-it until it’s on by default even when away-from-keyboard.
On the other end, because reading ‘>be me’ drops you into the same evacuated first person that writing it does, the training runs on the reader as much as the writer. If arousal is the case where the form does something to your body the > can’t retract, this is the case where it starts doing the same thing to your mind. If you occupy enough borrowed first persons, ‘arm’s length’ can become the only length you’ve got.
Granted, this is all theoretical; irony poisoning is an internet coinage, and post-irony is a critical frame. But the underlying machinery, something with the potential to free and hollow out the same people one unowned sentence at a time, is all too real.
>conclusion
When a language model produces passable greentext, it’s executing rules that a generation of anonymous human posters already wrote out by hand. This is all well and good; language models excel at working within far more abstract sets of rules to mimic a given form. But most pieces of greentext start with some variation of ‘>be me’—and no matter what the ELIZA effect might have you think, there is no ‘me’ behind any LLM.
Greentext generated by X user @teortaxesTex using DeepSeek R1 (2025).
This bulwark of meaning only holds if we demand the existence of a ‘me’ behind every ‘>be me’. But even in the years before LLMs, the experience of reading greentext was always defined by anonymity and ambiguity. We’ve mostly discussed this ambiguity in terms of sincerity versus irony—but one level down, the same question takes on a much more existential register, especially for anyone worried about automation: has there ever been a >me?
Like a lot of things on our increasingly Dead internet, the tidiest-looking answer may still live on GitHub. Where 4chan greentext launders any verdict away by design, the Greentext programming language takes a different approach: ‘>be me’ isn’t a declarative that promises interiority, but merely the line where execution begins. But if ‘>be me’ is the line that starts the program running, what machine is it running on?
As a literary form, greentext has no processor of its own—only the person reading it. Grammatically, it executes in your head. The > empties the origin, and ‘be’ sources a destination. Together, they conscript a ‘me’ from the reader at first contact, with or without the reader’s consent—and because you still get to answer from cover, obedience costs nothing and friction is minimal. An anonymous reader stepping into the subject position of an anonymous writer, with no one locatable at either end of the transaction.
The first thing this essay did, before a single line of argument, was to put a ‘>be me’ in front of you—and for the length of a few lines, you were a thirteen-year-old you’ve never met, feeling around inside a first person that was never yours. This turns you, the reader, into something very different: something that executes instructions and gets to disclaim authorship of the result. In loaded 2020s terms, greentext transforms the reader into an agent. I was just quoting a version of myself, and you were just following orders.
I started thinking about this essay shortly after reading Vilém Flusser’s Does Writing Have a Future?, hoping to arrive at my own answer to his titular question—and the one greentext offers isn’t exactly comforting. Its creators are incontestably human, and they built a form that does not require humans to work.
There’s consolation to be found in this ambiguity. While it’s impossible to read a post anywhere else online without wondering whether whatever you’re looking at came from a person or a machine trained to pass for one, that question has never had any purchase in the world of greentext. The form is AI-proof not because its writers never had anything worth taking — the naked confessions of /r9k/ say otherwise — but because nothing is ever stored on the page, where a counterfeiter could reach it. What the > withholds is verification of a ‘me’, not the existence of one—and you can’t counterfeit a thing that was never certified to begin with. Greentext ensures that ‘me’ is always you, and the people who read and write it have spent fifteen-odd years training themselves to never look for anyone else in the first place. By reading without reaching for a writer, and writing at enough of a distance that whether we’re there stops being relevant to our potential readers, we take turns being unlocatable. This act of reading-without-reaching and writing-without-holding is not the same as being human.
>implying it’s also not the same as being not human
References
‘Dead Internet Is Here: Bot Traffic Tops Human Traffic for First Time’, PCMag, https://www.pcmag.com/news/dead-internet-is-here-bot-traffic-tops-human-traffic-for-first-time.
Vilém Flusser, Does Writing Have a Future?, trans. Nancy Ann Roth, Minneapolis: University of Minnesota Press, 2011 [1987].
Isabelle Graw, Reinhold Martin, and André Rottmann, ‘Do Media Determine our Situation?’, Texte zur Kunst 98 (June 2015), https://www.textezurkunst.de/en/98/do-media-determine-our-situation/.
‘Greentext Stories’, Know Your Meme, https://knowyourmeme.com/memes/greentext-stories.
‘How Typing Transformed Nietzsche’s Consciousness’, The MIT Press Reader, https://thereader.mitpress.mit.edu/how-typing-transformed-nietzsches-consciousness/.
jz-feng, ‘Greentext’, GitHub, https://github.com/jz-feng/greentext.
Friedrich Kittler, Gramophone, Film, Typewriter, trans. Geoffrey Winthrop-Young and Michael Wutz, Stanford: Stanford University Press, 1999 [1986].
Marshall McLuhan, Understanding Media: The Extensions of Man, New York: McGraw-Hill, 1964.
Hitori Nakano, Train Man: Densha Otoko, trans. Bonnie Elliott, New York: Del Rey, 2007 [2004].
New Models, ‘The Online Marketplace of Ideas w/ Joshua Citarella’, New Models, 10 June 2026, https://newmodels.substack.com/p/the-online-marketplace-of-ideas-w.
‘Strange Rules’, e-flux Announcements, 2026, https://www.e-flux.com/announcements/6786944/strange-rules.
Teortaxes (@teortaxesTex), post on X, 25 January 2025, https://x.com/teortaxesTex/status/1883290697228562867.
David Foster Wallace, ‘E Unibus Pluram: Television and U.S. Fiction’, Review of Contemporary Fiction 13.2 (1993): 151–194.
‘Why Do A.I. Chatbots Use “I”?’, New York Times, 19 December 2025, https://www.nytimes.com/2025/12/19/technology/why-do-ai-chatbots-use-i.html.
Biography
Myles Badger is a writer and researcher based in Boston whose work explores internet culture, media theory, and the systems that capture our cognitive labor. He is one of the organizers of Boston Tech Poetics, a collective of artists drawn to technology and technologists inspired by the arts. His writing has been published by Do Not Research and can also be found on his newsletter, Myles Mode.