AI models do not usually process text one character or one word at a time. They break text into tokens, which are small chunks such as words, word parts, punctuation, or spaces.
In practice
Tokens matter because model context windows, API pricing, and output limits are usually measured in tokens. A long prompt, a large file, or a long answer uses more tokens than a short exchange.
What to watch
A token is not exactly the same as a word. Different languages, code, symbols, and formatting can change how many tokens a message uses.