AI glossary for businesses
Token
In short
A token is the small unit of text that an AI language model reads and produces. A token can be a whole word, part of a word, a number or a punctuation mark. Models count everything in tokens: the length of your input, the size of the context window and, when you use an API, the price you pay.
How tokens work
Before an AI model reads your text, it cuts it into pieces. Think of Lego bricks: a common short word may be one brick, while a long or rare word is built from several. "Shop" is likely one token. A long product code or a rare surname may be four or five.
For English, a common rule of thumb from OpenAI is that 100 tokens are roughly 75 words. Greek usually needs more tokens for the same meaning, because many tokenisers were built mainly on English text and Greek words are often split into more pieces. The exact ratio depends on the model, so test it with your own texts. Most providers offer a free tokeniser tool or show token counts in the API.
In practice for a small business
An e-shop with 2,000 products wants new descriptions in Greek and English through an API connection, perhaps automated with Make or Zapier. To estimate the cost, the team:
runs 20 products through the model and records the tokens used per product, input plus output;
multiplies by 2,000 and by the provider's current price per token;
adds a margin for retries and corrections.
This small test turns a vague worry about cost into a number they can approve or reject. It also shows whether Greek output costs noticeably more than English. For a wider view on value, see how to measure ROI on an AI project.
If you use a monthly subscription such as ChatGPT or Microsoft 365 Copilot, you do not pay per token. Tokens still matter, because they set how much text fits in one conversation.
Why it matters and what to watch out for
Limits: the context window is measured in tokens. Very long Greek documents can hit the limit sooner than you expect.
Cost surprises: output tokens are often priced higher than input tokens. Asking for shorter answers can reduce cost.
Hidden tokens: system instructions and chat history are counted too, on every request.
Prices change: providers update prices and models often. Recheck before a large job.