Token

A token is the unit of text an AI language model reads and generates — a chunk of characters that is usually a word, part of a word, or a punctuation mark. In English, one token averages about three-quarters of a word, so 1,000 tokens is roughly 750 words.

Why it matters

Tokens are the meter that everything AI runs on. API pricing is per token, in and out. Model limits — how much context fits, how long a response can be — are measured in tokens. Speed scales with token count. For marketing teams operating AI at volume, token economics quietly decide what's affordable: a workflow that stuffs an entire website into every prompt costs multiples of one that retrieves only the relevant passages, for often-identical output quality.

How it's used

You'll meet tokens in three practical places. Budgeting: estimating the monthly cost of a content pipeline or agent means estimating tokens per run times runs per month. Prompt design: trimming boilerplate from prompts that run thousands of times is real money — this is part of why prompt ops emerged as a discipline. Debugging: when a model "forgets" earlier instructions or truncates output, you're usually hitting a token limit — see context window.

Related terms

Context window · Prompt engineering — the context window is measured in tokens; prompt engineering is partly the art of spending them well.