AI 101 — What is ChatGPT?

ChatGPT is a chat app from OpenAI where you type a request in plain language and get back an answer — text, images, code, a plan, or finished work — powered by a large language model called GPT running behind the screen. ChatGPT is the product; GPT is the model inside it. Almost everything a billion-plus of us now ask AI to do happens through this one text box.

Why it matters right now

ChatGPT is the front door most people use for AI, and OpenAI says more than 1.2 billion people use it every week, according to OpenAI's October 7 announcement. That announcement also moved the app onto GPT-6 for everyone, not just paying users, and added what OpenAI calls Intelligent UI — replies that can arrive as tappable diagrams, charts, and little interactive tools instead of plain paragraphs.

More importantly, the thing has quietly changed species. It started as an answer box; it is now partly an agent. OpenAI launched ChatGPT Work, an assistant that can gather material from your apps and files and produce finished documents, spreadsheets, and web apps, staying with a task for hours — with its coding agent Codex built in, which OpenAI says more than 5 million people use weekly. And in August, OpenAI removed the cap on text chats for free users, the same month the app first crossed a billion weekly users. The free product is now effectively unlimited text, funded partly by advertising OpenAI began selling in October 2026, per OpenAI's ad announcement.

The mental model

Every time you press enter, three things happen. First, your message is stitched together with everything else the model should see right now: the earlier turns of your conversation, and the hidden instructions OpenAI calls the system prompt that set the app's tone and rules. That whole pile is the context window — the model's working memory for this conversation, unpacked in What is a context window?. Second, the large language model underneath — the subject of What is a large language model? — reads it and predicts a reply one piece at a time, which is why answers arrive word by word. Third, if you ask for something that needs the outside world, the app can call tools: run a web search, execute code, generate an image, or open a file, then fold the results back into the answer.

Nothing is looked up in a database and nothing is stored in the model itself. The model's weights were fixed when training ended; your conversation is the moving part.

The analogy: a brilliant researcher with amnesia

Imagine hiring an assistant who has read a huge fraction of everything written — fluent, fast, broadly knowledgeable — but who starts every morning with no memory of yesterday. On the desk in front of them sits a fixed pile of paper: everything you have said so far, plus OpenAI's house rules. That pile is the context window, and it is the only thing they remember. Hand them the reports and they will synthesize them in seconds. Ask what happened last Tuesday and they will guess — plausibly, confidently, and possibly wrong. And they only know what is in the pile unless you let them walk to the library (a web search) or run the numbers (code execution).

Common misconceptions

"It's a fancy search engine." It isn't. ChatGPT's core skill is generating text that fits the pattern of a good answer, and when the answer isn't in the pile in front of it, it will still often produce one rather than admit the gap. That failure mode has a name — what an AI hallucination is — and it is the reason fact-checking still matters.

"It remembers me." The app has optional memory features, but the underlying model never learns from your chat. Start a fresh conversation and the assistant is the amnesiac again, holding only what you paste in or what the app chooses to carry over.

"GPT and ChatGPT are the same thing." GPT is the model family — also sold to developers as an API, under versioned names — while ChatGPT is the consumer app wrapped around it: accounts, history, safety rules, tools, and increasingly, an interface that builds itself around your request. Confusing the two is like confusing an engine and a car.

Where to learn more

Related reading: What is a large language model? on the engine under the hood · What is a context window? on the only memory the app has · What is an AI hallucination? on why confident answers still need checking.

When was the last time ChatGPT convinced you of something that turned out to be wrong? Tell us in the comments.