A word like automation might be split into two or three tokens rather than counted as one unit, and a short sentence typically runs 15 to 20 tokens once punctuation and spacing are included. Most AI providers charge by the token and count both the text sent to the model and the text it generates back.
Tokens matter beyond cost because a model's context window is also measured in tokens, so a long document can use up most of the available space before the model even reaches the actual question. The common mistake in cost planning is estimating usage in words rather than tokens, which tends to understate the real token count and the real bill.