Available nowGuide

Tokens: what they are and how to use them well

What tokens are in AI models, why they matter, and how to write more token-efficient prompts.

3 min read · AI & Automation

What a token is

A token is the smallest unit of text an AI model processes. It's not exactly a word or exactly a letter: it's usually a word fragment. Short, common words tend to be a single token, while long, uncommon, or foreign-language words can split into several tokens.

Simple example: the word "hello" is probably one token. A compound or technical word can turn into two or three tokens.

Why they matter

Tokens are the unit a model uses to measure three things at once:

  • Cost: AI model usage is generally measured and billed by the number of tokens processed, not the number of questions.
  • Context limit: every model has a maximum number of tokens it can "keep in mind" in a single conversation (what you wrote, plus what the model answers, plus any document you uploaded). If you go over that limit, the model starts losing track of what was said first.
  • Speed: the more tokens a model has to process and generate, the longer it takes to respond.

How to write more token-efficient prompts

Less efficient

Pasting an entire 40-page document when you only need the model to review one specific section.

More efficient

Paste only the relevant section, or first ask the model to identify which part of the document you need before analyzing it in depth.

Less efficient

Repeating the entire conversation context in every new message, "just in case".

More efficient

Trust that the model remembers the conversation (within its context limit) and just clarify what changed.

Less efficient

Asking for very long answers when what you need is a short decision or a brief list.

More efficient

Be specific about the output format: "answer me in 3 lines" or "just give me the list, no explanation".

In short: think of tokens as the space and time you give the model to work with. Give it only what it needs, in the most direct format possible, and you'll get faster, more focused answers.

Note: we don't include pricing figures or exact token limits here because they change depending on the model and provider. Check the official documentation of the tool you use for those specific numbers.