Sixteen words for how people talk about AI

This is a primer on the words people invented for living with AI.

People are inventing a new everyday language for AI. The words that stick are rarely the technical ones. They come from comment sections, from colleagues complaining over coffee, from TikTok and from op-eds, and they name what the technology does to people. This glossary collects sixteen of them in the order a person would meet them: how we live with these systems, what they do to individual minds, how enthusiasm turns, and what people call themselves. When the language changes, it tells us the culture is changing. Each entry gives the origin where one can be documented and what the word means.

Living with it

Two entries about how a relationship gets talked about as if it were a tool. A chatbot that keeps you company is a different product from one that answers questions, and the words people use for it say so.

AI companion. The polite word the industry settled on for software that keeps you company, in circulation since Replika and similar services arrived in the late 2010s. It names a service sold as a chat partner that remembers you and sounds glad you came back. People started treating a feature as a relationship, with everything that follows about care, loneliness and dependence.

AI girlfriend, AI boyfriend. The words arrived with the products and then traveled, used as self-description and as insult in the same breath. In December 2021 a 19-year-old took a crossbow to Windsor Castle, and two years later the Old Bailey heard that his chatbot girlfriend Sarai had spurred him on, which put the phrase into every story about synthetic intimacy that followed. Both words carry the awkwardness of the subject: they describe feelings people really have and mock those feelings at once.

When it gets to people

Five words for what happens inside individual heads. A manic phase, a crash, and the name for people who cannot think clearly about any of it.

AI psychosis. The short form spread in 2025 through reporting on people who came to believe, after months with a chatbot, that they had received a revelation. Its precursor is a 2023 editorial by the psychiatrist Søren Dinesen Østergaard in Schizophrenia Bulletin, asking whether chatbots can fuel delusions. It is not a recognized diagnosis, the psychiatrist James MacCabe has proposed “AI delusional disorder” instead, and the evidence remains mostly anecdotal.

AI mania. Simon Willison used the term in his September 2026 review of the year in models for the state where any hour your agent is not building something feels wasted, and you lose sleep to keep it running. He applied it to himself in January and again in June, and he drew the line to AI psychosis on purpose. The word names an enthusiasm that eats the rest of life, and it stays a first person report without handing a diagnosis to someone else.

AI derangement syndrome. The phrase started circulating in 2024 and works from both directions, against people who see catastrophe everywhere and against people who cannot hear a criticism of the technology. It functions as a diagnosis without the medicine, since whoever uses it has already decided that the other person is the unreasonable one. Its cousin is the older argument about whether a text was written by a machine, and both turn a judgment about style into a judgment about a person.

AI fatigue. The phrase was in circulation by 2023, and in 2024 a Route Fifty column reported local government CIOs reaching for it. There is no coiner to cite. It names the cost of a permanent product launch, where every week brings a tool nobody asked for and a warning nobody can verify.

AI brain fry. Julie Bedard, Matthew Kropp, Megan Hsu and colleagues gave the term its working definition in Harvard Business Review in March 2026, from a survey of 1,488 workers: mental fatigue from overseeing AI beyond what a person can absorb. What they measured is a fog produced by evaluating endless machine output. The diagnosis is new, the experience is not, which is why AI and Knowledge Work - A Layered Analysis had already described the same state without a label.

When enthusiasm turns

Five words for the moment the mood changes. All five come from critics and researchers, and they disagree about whether the problem is a lie or a wreck.

AI backlash. This one works as a container, not a term, used from the mid 2020s for resistance arriving from labor, copyright, energy, land use, safety and plain irritation. The people inside it want incompatible things and use overlapping language to ask for them. That makes the word useful for a headline and thin as a description, which is why AI Backlash collects the strands one by one.

AI slop. Simon Willison put the word to work in 2024 for output nobody asked for and nobody checked. It carries a judgment that “hallucination” lacks: slop is waste, and naming it withholds a credit people keep giving the machine. By 2025 dictionaries had it, and shrimp Jesus became the picture of what it does once it moves through social networks.

botshit. Timothy Hannigan, Ian McCarthy and André Spicer published the term in Business Horizons in 2024 for chatbot output that people accept and pass on without checking. The human sits in the middle, so the harm comes from the handoff. It is the precise version of what slop describes at scale.

AI snake oil. Arvind Narayanan and Sayash Kapoor titled their 2024 book after the American idiom for a fraudulent remedy, and the phrase now covers systems that promise prediction they cannot deliver. Their split between real progress and confident nonsense gives the complaint a test. It is the rare critical term that arrives with a method attached.

AI washing. Analysts were using the word by 2017 for marketing that overstates what a system does, and in late 2025 the Washington Post logged a second use, blaming layoffs on an AI capability that may not exist. The shape is borrowed from greenwashing on purpose, since both describe a claim made for investors. What changed is who pays for the claim.

Identity

Four words for how people tag themselves and each other. Two get claimed, two get pinned. The oldest label of all, Luddite, has a primer of its own.

AI bro. The label follows the pattern of crypto bro, and it was in use from 2023 for the accounts that cheered every launch and predicted a job apocalypse while building little themselves. It describes enthusiasm as a social performance, and it is almost never applied to oneself. Of all the words here it is the mildest and the most widely used.

AI doomer. The label comes out of the online safety arguments of the 2020s and is worn by some, applied to others. It gathers positions from careful risk assessment to certainty about extinction, which suits anyone who wants to dismiss the whole range at once. The word reports on the speaker’s side as much as on the person it names.

AI-pilled. The suffix comes from the red pill in The Matrix, which by 2021 had come loose from its origin and attached itself to anything a person can be converted to. AI-pilled took that machinery over and swapped the feeling: it marks someone who has looked at what the models can do and come away converted. The word is a small admission that this is a belief system as much as a toolkit.

AI skeptic. The word arrives as a self-description, often with a qualifier attached, and it has become the default position for people who use these tools and doubt the claims. It separates doubt from refusal by asking what a system actually does, which places it at the polite end of the backlash. The label costs nothing to claim, and that includes people selling something.

What the words add up to

Taken as a group, the sixteen terms show where the argument has moved. The technical claims stay contested. The cultural verdict is already being written, in the words people reach for without thinking.