The context window holds the information a language model can use when producing a response. It includes the supplied conversation and material plus the generated output. Anthropic distinguishes this working context from training data and notes that a larger window does not automatically produce better results. Source: Anthropic: Context windows

Keep the current brief easy to find

For an itinerary task, provide the latest dates, traveller requirements and confirmed conditions in a clear brief. Identify which earlier instructions have been replaced. This gives the reviewer a precise set of requirements against which to check the response.

Test long conversations

Ask the supplier to demonstrate what happens as a conversation grows, including how older material is summarised or removed. Check that essential booking conditions remain available. This tests the workflow’s handling of information instead of judging it solely by its advertised capacity.

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