What are tokens in AI?

What is a token in AI?

A token is a small chunk of text — often a whole word, but sometimes part of one — that an AI model uses as its basic unit of language. Models don't read letter by letter or sentence by sentence; they break text into tokens, then predict the next token over and over to produce a response. "Tokenization" is just the step of slicing text into those pieces.

How tokenization works

Common words are usually a single token, while longer or unusual words get split into several. "Cat" might be one token; "unbelievable" might be two or three. Spaces and punctuation count too. A handy rule of thumb for English: about 0.75 words per token, so roughly 1,000 tokens covers about 750 words. Both your input and the model's output are measured this way — they're the unit behind a model's context window.

Why tokens matter

• Cost — AI APIs are usually priced per token, so longer prompts and answers cost more.
• Limits — context windows and rate limits are defined in tokens, capping how much a model can take in at once.
• Speed — models generate one token at a time, so longer outputs take longer.

This is why a concise, well-targeted prompt is often cheaper, faster, and better — see why your prompts aren't working.

Do you need to think about tokens?

For everyday use, no — you don't need to count them. But understanding tokens demystifies the things that confuse beginners: why a chat "forgets," why an API bill is what it is, and why pasting a whole book doesn't work. That kind of mental model is exactly what AI fluency is made of, and you can build it in about 5 minutes a day. See where you stand with the free AI IQ test.