Language models process text in units called tokens. These can be individual characters, whole words or parts of words. Tokenisation is the process of splitting text into those units. A token count therefore differs from an ordinary word count. Source: Google Cloud: Generative AI glossary

Use the provider’s counting tool

When assessing a supplier proposal, ask for a token count using representative enquiries and replies. Include the instructions and supporting material the application sends. This gives the estimate a basis in the intended task instead of a rough count of visible words.

Keep the assumptions with the estimate

Record the model, sample length and expected response length alongside any usage calculation. Ask which input and output charges apply. This makes the estimate reviewable when the workflow changes and avoids treating a sample conversation as a fixed cost for every enquiry.

Written and updated by Travelgenix. Sources checked on 19 October 2026. Practical advice and examples are Travelgenix editorial guidance.