A retail company deploys a customer service chatbot built on a pretrained large language model. Support agents report that the bot confidently answers questions about store return policies but frequently states rules that were changed six months ago and no longer apply. The prompt already includes clear instructions and formatting examples. Which approach would most directly resolve the problem of outdated policy answers?