A token is the basic unit of text an AI language model works with. It is usually a whole short word, part of a longer word, a number or a punctuation mark. In English, one token is often around three-quarters of a word, though this varies by model and language.
Before a model reads your message, a tokeniser splits the text into tokens and converts each into a number. The model then predicts the next token, one at a time, to build its reply. Everything the model takes in and gives back is measured in tokens, which is why both limits and prices are expressed that way.
For a retailer or café group, tokens matter in two practical ways. First, AI providers usually charge per million tokens of input and output, so a long product catalogue pasted into every chat costs more than a short question. Second, every model has a ceiling on how many tokens it can handle at once, which caps how much information you can give it.
Watch for hidden token use. System instructions, conversation history and retrieved documents all count, not just the customer's message. A chatbot that resends a long menu on every turn can burn through budget quickly. Non-English text and unusual product codes can also use more tokens than they appear to.
Understanding tokens lets you estimate running costs, set sensible limits and compare providers fairly. It is the unit behind almost every AI bill and every "the model forgot" complaint.