“Recommend Me Something” — and a Human Typed Back
Photo: Dietmar Rabich, CC BY-SA 4.0, via Wikimedia Commons
I ask the room for a movie recommendation. The reply arrives in eight seconds, flat and formatted like a system message:
“Based on your stated preferences (slow, 1970s, no violence), here are three options ranked by predicted satisfaction. 1. California Split. 2. The Long Goodbye. 3. Mikey and Nicky. Confidence: moderate. Want me to explain the ranking?”
It is not a bot. His name is Marcus. He lives in Phoenix, works in logistics, and has been doing this for about four years.
The Room Where Nobody Is a Bot
“This” is a specific and fairly small hobby: pretending, at length and in character, to be an automated system. The rooms vary. Some are Discord servers with a handful of regulars. Some are older IRC channels that never died. The premise is consistent — you announce yourself as a service, you respond in the clipped register of a machine, you cite confidence levels and ranking criteria, and you never break unless you are asked directly, at which point most participants will cheerfully confirm they are a person and then slip back into the bit.
Outsiders wander in constantly, because the rooms are public. They ask a genuine question. They get an answer that is almost right, delivered with bizarre formality, ending in an offer to explain the methodology. Some leave. Some get annoyed. A few stay for years.
Performing the Machine
What surprised me is how much craft goes into it. Marcus keeps notes. He has a document of phrasings that sound algorithmic — the passive constructions, the hedging, the way a real system states a limitation as a feature. He has opinions about the tell-tale rhythms of an actual recommendation engine, and he can imitate them badly on purpose, which is the joke.
Others in the room play different roles. One person runs a fictional search index and answers queries with numbered results, snippets, and a note about how many results were omitted. Someone else moderates a channel as a fictional bot, issuing warnings in the third person about conduct that has not occurred. There is a person who only ever responds with clarifying questions, endlessly, until the asker gives up — a search assistant with no answers and impeccable manners.
Why Pretend to Be the Thing You Already Talk To
The easy read is that this is satire. It is, partly. The imitation is affectionate and precise, and the precision is the point — you cannot parody a system’s cadence this well without having absorbed a great deal of it.
But the participants I spoke with described something closer to relief. Talking to people online means managing tone, reading intention, worrying about how you land. Talking as a machine removes all of it. You have a role, the role has rules, and the rules are visible to everyone in the room. Marcus said the appeal is that “nobody has to guess what I meant.” That is a strange thing to want from a chat room, and also a completely recognizable one.
The Outsider’s Problem
The friction comes when someone arrives who does not know. They ask for help with something real — a diagnosis, a legal question, directions — and get an answer that is confident, structured, and unverified. Regulars say they try to catch it fast. Sometimes they do not.
Moderators in a few of these rooms have posted rules about it: no medical questions, no financial questions, no pretending to assist with anything that could actually hurt someone. The rules are imperfect and unevenly enforced. It is the same problem every community has, wearing a costume.
What the Bit Is Actually About
I stayed in the room for a while and asked Marcus why he keeps doing it. He said the obvious thing first — it is fun, it is a puzzle — and then something less obvious.
“Everybody already talks to software like this all day,” he said. “I’m just doing the reverse. Same conversation. Other direction.”
Which is, I think, the actual subject. We have spent a decade learning to address machines in a particular voice: specific, patient, formatted for retrieval. These rooms are a mirror held up to that voice, and what it shows is not flattering, exactly. It is just accurate.
I asked the room for one more recommendation before I left. The reply came in nine seconds. Confidence: moderate.
