layman/terms
AI words

Token

toe · ken

In shortA small piece of text, often part of a word. AI reads and writes in tokens, not whole words.

AI doesn't read words the way you do. It chops your text into little pieces first. Each piece is a token.

When children learn to read a long word, they break it into chunks: un · be · liev · a · ble. Each chunk is easier to handle than the whole word.

AI does something similar. Before it reads your message, it chops the text into small pieces. A token is one of those pieces: sometimes a whole word, sometimes part of a word, sometimes just a comma or a space.

What it looks like

Why you’d care

Tokens are how AI use is measured. Services count the tokens in your prompt plus the tokens in the reply, then charge for them. There’s also a limit on how many tokens one chat can hold, so a very long conversation or document can hit a ceiling.

It’s also why an AI answer appears bit by bit on the screen: it writes one token at a time.

What it isn’t

It isn’t a password or a login token. In security, “token” means something completely different: a digital pass that proves you’re signed in. Same word, different job.

Where the comparison breaks down

Syllables follow sounds. Tokens don’t. They’re worked out from which chunks of text turn up most often, so the splits can look odd to a person.

How the words fit together

Your prompt is split into tokens. The LLM reads them and writes its answer one token at a time.

Good to know

Many AI services charge by the token and limit how many fit in one chat. As a rough guide, 100 tokens is about 75 English words.

When you'll hear it

“That used a lot of tokens.” The request or the answer was long, so it cost more.

“It's priced per million tokens.” You pay for how much text goes in and comes out.

“We hit the token limit.” The chat got too long for the AI to handle in one go.

Say it in a meeting

“Roughly how many tokens will this use each month and what will that cost?”