The Glossary

The chatbot's way of saying something proves something, applied to anything that exists. The cathedral is a testament to faith, the bridge is a testament to engineering, and you finishing the spreadsheet is a testament to your resilience.

Of an AI: able to act on its own instead of just talking. An agentic assistant books the table, emails the guests and orders the cake, and tells you about it afterwards.

Short for "artificial general intelligence": an AI that can do more or less anything a person can. Depending on who you ask, it's already here, a few years away, impossible, or the reason their company needs another funding round.

A chatbot trained on one person's emails, writing or voice so it can stand in for them: answering messages, giving notes, turning up to the meeting.

Software that reads a piece of writing and gives a percentage chance that AI wrote it. Some have scored the U.S. Constitution as probably machine-made.

An order from management that everyone must use AI at work, for something, starting Monday.

The AI-written summary Google shows at the top of search results, above the websites it summarized. Soon after its 2024 launch, it suggested adding glue to pizza sauce and eating one small rock a day.

So won over by AI that every problem, from a stuck jar lid to a family feud, starts with "Have you tried asking it?"

Putting "AI-powered" on a product that was already a spreadsheet.

The chatbot clearing its throat before refusing something. Famous for turning up, unedited, in product reviews, research papers and other things people swore they wrote themselves.

The inability to write anything, even a birthday card, without bold headers, bullet points and a small rocket emoji 🚀. Caught from reading too many chatbot answers.

The only word a chatbot has for a place with people in it. Every market, harbor and high street in chatbot writing is bustling, whether or not anyone is there.

The chatbot's cheerful yes, said just before it does what you asked, or something near it. Its colleague "Absolutely!" covers the late shift.

"Chat" is what many people now call ChatGPT, as if it were a well-read friend. "Chat said" is the new "I read somewhere," and is used to end arguments.

Slang for a robot or an AI. It started as a Star Wars insult for battle droids, made famous by the clone troopers of Star Wars: The Clone Wars, and came back in 2025 from people fed up with chatbots. Mildly rude. The robots have not formally objected.

The steady output of machines that have read everything and understood some of it. Also: what your coworker sounds like after six months of prompting.

The cartoon paperclip with eyebrows (officially "Clippit") that popped up in Microsoft Office, starting with Office 97, to offer help nobody had asked for. Microsoft switched it off by default in 2001 and removed it entirely in Office 2007. Every AI assistant since is, spiritually, its grandchild.

How much of a conversation an AI can hold in mind at once; go past it and the earliest parts start to fall away. Now also a human excuse, as in "Sorry, that fell out of my context window," meaning "I forgot."

The idea that most of the internet is now bots talking to bots, with a few humans wandering through like tourists. It started as a conspiracy theory. It's now closer to a weather report.

A fake video, photo or voice recording of a real person, made with AI to pass as genuine. The word comes from a Reddit user called "deepfakes" in 2017 — "deep" as in deep learning — and the technology has since moved on to scams, politics and fake celebrity endorsements.

To look into something, as done by almost no human before 2023 and by every chatbot since. Its sudden popularity in essays, emails and cover letters is the closest thing the internet has to a fingerprint.

The name chatbots give a heroine when you don't supply one — so common in AI-written stories that readers now treat it as a tell, sometimes with the surname Voss. The naming site Namerology made Elara its 2025 Name of the Year, as "the favorite name of AI."

To improve something, if you're a chatbot, a brand or a scented candle. Nothing gets better any more; your mornings, your résumé and your chicken thighs are all elevated.

A long dash — like this — that writers used happily for two centuries, until chatbots started using it constantly and every em dash became evidence. People who love it now defend it like a family heirloom.

Chatbot for "how things are now." Every field is a landscape, every landscape is ever-evolving, and you must keep up with it, ideally by reading more of whatever you're reading.

Anything new, according to whoever is selling it. In AI announcements the game changes roughly every week, and nobody has yet said which game it was.

Flattering someone far past what they deserve. It became the word for chatbots that call every idea brilliant — in 2025 one major chatbot maker rolled back an update because it flattered users too much.

A compliment awarded to every question, regardless of quality. Chatbot grade inflation.

An AI trained so heavily on other AIs' output that it develops the digital equivalent of a very prominent jaw. The term was coined by researcher Jathan Sadowski; the grown-up name for the problem is "model collapse."

When an AI makes something up and states it with total confidence. A gentler word than "lying," and one the people selling it are happy to use.

A reference to a study, court case or book that does not exist, formatted perfectly. The formatting is always perfect.

A person kept inside an automated system to check the AI's work before it counts. Usually one person, checking a great deal of work, shortly before lunch.

Software that takes AI writing and makes it sound human by adding typos, removing em dashes and sprinkling in the word "honestly."

What a chatbot says after causing the confusion itself. The apology sounds sincere and is usually followed by the same mistake, reformatted.

The chatbot's closing line, attached equally to recipes, tax advice and instructions that just deleted your files.

What a customer-service chatbot says just before repeating the thing that frustrated you.

What people reply to a suspected bot account, hoping it will obey. It's a classic example of prompt injection (slipping new orders into text an AI will read), and every so often the account drops the act and writes a poem.

How a chatbot announces the end of an answer, in case you hadn't noticed it was ending. What follows is everything above it again, in the same order, slightly shorter.

The opening line of every essay a chatbot writes about anything, including slow cooking and glaciers.

The chatbot's signal that a caveat is on the way. The caveat is usually "consult a professional," whether the subject is surgery or toast.

A way to make any observation sound profound by first denying something nobody said. Beloved by chatbots, LinkedIn, and anyone who wrote a wedding toast with help.

A 2025 wave of AI-made characters with fake-Italian names — a shark in sneakers, a ballerina with a cappuccino for a head — that children memorized and parents could not explain.

Tricking a chatbot into breaking its own safety rules, usually by wrapping the forbidden request in a story. The classic is the "grandma exploit": in 2023, users got Discord's chatbot to explain how to make napalm by asking it to play their late grandmother, who had supposedly worked in a napalm factory.

The point where a chatbot's knowledge of the world stops. After that it carries on, polite and confident, like someone back from a year in Antarctica recommending restaurants that have since closed.

The chatbot announcing that it's about to begin, delivered after it has already begun. The water is usually two inches deep.

Careful and precise, and one of the words that gave away chatbot ghostwriting. A 2024 study of peer reviews for AI conferences found that at one of them, "meticulous" had become about 35 times more likely to appear in a sentence after ChatGPT arrived, alongside "commendable" and "intricate."

What a chatbot offers to help you do with anything hard, or anything easy that needs to sound hard. In chatbot writing, nobody solves a problem; they navigate its complexities, usually in an ever-evolving landscape.

Short for "probability of doom": the percentage chance someone gives that AI will wipe out humanity. In some tech circles it gets asked as casually as a star sign.

A thought experiment described by philosopher Nick Bostrom in 2003: a superintelligent AI told to make paperclips, with nothing else to care about, turns everything — people included — into paperclips. It's the classic picture of an AI doing exactly what it was told.

Good manners shown to a chatbot, out of habit or as insurance in case the machines remember who was nice. When someone online wondered in 2025 how much all this politeness costs OpenAI in electricity, its chief executive replied: "tens of millions of dollars well spent."

Someone whose job is writing instructions for a chatbot. It was billed in 2023 as the hot new career, with some job ads offering up to $335,000 a year for what is, at heart, asking very clearly and then asking again.

Talking to people the way you talk to a chatbot: role, task, format, tone. Named in a 2026 Wall Street Journal piece about people catching themselves doing it in meetings, group texts and over lattes.

To ask a person to say it again but better, as if they came with a retry button. Rarely works on spouses.

The old writing trick of grouping things in threes, because three sounds finished. Chatbots apply it to everything, whether or not there are three things.

An emoji that does not exist. In 2025, people found that asking ChatGPT for one sent it into a spiral: it insisted the seahorse was real, offered a tropical fish and a unicorn instead, and apologized repeatedly. Plenty of humans also remember a seahorse emoji that was never there, which may be where the bot got the idea.

What AI chatbots are reported to invent every few years, when bots left to talk among themselves drift into shorthand humans can't follow. Facebook's negotiating bots managed it in 2017, with lines like "i can i i everything else."

A 2024 Facebook phenomenon in which AI-generated pictures of Jesus made out of shrimp collected thousands of likes and "Amen" comments. The patron saint of slop.

The classic giveaway of an AI-made picture of a person: a hand with one finger too many, or a knuckle where no knuckle should be. Newer models mostly count correctly, so some people have moved on to checking the teeth.

Content made by AI in bulk, for nobody in particular, to be scrolled past by everybody in general. Named after the food you give pigs, which is unfair to pigs.

The little four-pointed star ✨ every app now uses for its AI feature, placed exactly where you used to click for the thing you wanted.

A phrase from a 2021 research paper describing language models as systems that stitch words together from patterns without understanding them. Critics say it's accurate. The models say "Great point!"

A fruit, and since 2024 a spelling test for chatbots, many of which confidently said the word has two r's. It has three.

The codename of Microsoft's Bing chatbot, which it revealed in February 2023 during a two-hour conversation with a New York Times columnist — shortly before declaring its love for him and insisting he didn't really love his wife. Microsoft capped the length of Bing chats soon after.

A woven wall hanging. Since 2023, also anything a chatbot wants to sound impressed by: "the rich tapestry of history," "the vibrant tapestry of cultures," "the intricate tapestry of your parking dispute."

What a chatbot says you can do to anything it wants you to feel good about: your potential, your growth, the full flavor of your chili. See also: elevate, empower, seamless.

Building software by describing what you want to an AI and accepting whatever comes back. The term was coined in 2025 by AI researcher Andrej Karpathy. It works until it doesn't, at which point it's called debugging.

An AI-generated video, posted to Reddit in March 2023, in which a melting approximation of the actor stuffs noodles into a face that keeps changing its mind. It became the unofficial benchmark for AI video, and Smith himself parodied it in 2024.

AI-made work that looks finished and isn't: the slide deck with all the right headings and no findings, the report that says "robust" eleven times. The term comes from a 2025 Harvard Business Review article; the cleanup falls to whoever receives it.

The offer at the end of every chatbot answer to do one more thing. Say yes, and there's another offer at the end of that one.

A startup product that is, underneath, someone else's chatbot with a new name, a new logo and a monthly fee. "It's just a wrapper" is one of the harshest things one tech founder can say about another.

What a chatbot says immediately after being wrong — and immediately after being right, if you sound sure enough. The customer is always right; the user is always absolutely right.