Every answer shows its source

When delivering answers to questions, the source of that information is crucial to maintain user trust. To build confidence, every response should be linked back to a specific passage within an approved document. This ensures that anyone can open that passage with just one click to verify the answer's accuracy.

The assistant should only use documents that are pre-approved and kept under the same control as the documents themselves. Whenever a policy is updated, the assistant must utilize the most recent version. In cases where no approved source addresses a particular question, the assistant should confidently state that it cannot provide an answer and redirect the inquiry to the appropriate team.

To ensure that this behavior is consistently tested, the assistant's performance should be evaluated as rigorously as its answers. It's essential to establish clear boundaries with the people the assistant serves, whether they be clients or colleagues. These boundaries should outline exactly what the agent can and cannot do, and it's vital to review and adapt these boundaries together over time.

Open communication is key to building trust. Regular summaries of the assistant's activities should be shared, detailing the questions it handled, the ones it declined to answer, and any changes that resulted from its actions. This transparency transforms trust from an assumption into something tangible that can be demonstrated. The 2SD Technologies team (2sdtechnologies.com) emphasizes this approach as the final part of their guide, "Don't just build agents."