The word AI never actually sees
When you type a question into an AI chatbot, you see words and sentences. The model itself never sees any of that — it sees tokens. Understanding what a token is explains a surprising number of everyday AI quirks: why a long document gets rejected, why pricing is measured in strange units, and why a response sometimes stops mid-sentence.
What a token actually is
A token is a chunk of text — sometimes a whole word, sometimes part of one, sometimes just punctuation — that an AI model treats as a single unit. Before any processing happens, a “tokenizer” breaks your input into these chunks. The word “token” itself might be one token, while a less common word like “tokenization” might get split into two or three pieces such as “token” and “ization.” As a rough rule of thumb, one token is about four characters of English text, or roughly three-quarters of a word, so 100 words usually works out to around 130-140 tokens.
Why this matters for context limits
Every AI model has a maximum “context window” — the total number of tokens it can hold in memory at once, covering your entire conversation plus its response. This is why pasting a 50-page document sometimes triggers an error or gets silently truncated: the document alone can exceed the token budget before the model even generates a reply. Once you know your model’s context window, you can estimate in advance whether something will fit, rather than finding out the hard way mid-task.
Why tokens, not words, drive cost
API pricing for AI models is almost always quoted per million tokens, split between input tokens (what you send) and output tokens (what the model generates), with output typically costing more. This is also why the same request can cost differently depending on the language you write in — some languages tokenize less efficiently than English, meaning the same sentence can use noticeably more tokens and therefore cost more to process.
A practical habit worth building
If you regularly work with long documents or code files in an AI tool, get in the habit of trimming irrelevant sections before pasting them in, summarizing earlier context instead of repeating it in full, and watching for a model’s stated context window before assuming it can “just read the whole thing.” These small habits prevent the two most common token-related failures: hitting a hard limit mid-conversation, and paying for far more processing than the task actually needed.
