In plain language
Models do not count words; they count tokens, and longer prompts and replies use more of them.
Example
A long email thread plus instructions uses thousands of tokens, which affects cost and whether it fits the model’s context window.
Related terms
Large Language Model (LLM) · Context Window · Prompt · Temperature
Common misconception
A token is not the same as an API access token; the AI term is about text, while access tokens are login credentials.
Implementation guidance
Estimate tokens before sending large documents, trim irrelevant text, set maximum output lengths, and monitor token usage for cost.