LEVEL 02 · LESSON 11
Grounding
Grounding connects an answer to evidence you can inspect.
What it is
Grounding means anchoring an answer in relevant evidence supplied to the system, such as a policy, a database result, or an observed tool response.
Why it matters
A model’s general training cannot reliably represent your current business facts. Grounding makes an answer easier to check and update.
A business example
A customer asks about returning a discounted item. The assistant receives the current return policy and the order details. It explains the applicable exception and points to the policy section rather than guessing from typical retail practices.
How it actually works
Select evidence that is authoritative for the question, include its source and date, and instruct the model to distinguish supported statements from missing information. Then check whether each consequential claim follows from the evidence. Grounding can use a pasted document, retrieval, or a live tool; it does not require a vector database. If sources disagree, the system needs a rule for which source takes precedence or a route to human review.
A common misconception
Providing a source does not guarantee that the model uses it correctly. A cited answer can still contradict the source or overlook an exception.
ONE SENTENCE TO REMEMBER
Grounding connects an answer to evidence you can inspect.
CHECK YOUR UNDERSTANDING
A policy assistant is given the current travel policy with each request and cites it in its answers. What does this grounding guarantee?
Show the answer
Answer: B.
Grounding connects an answer to evidence you can inspect, which makes it easier to check and update. It does not guarantee correct use: a cited answer can still contradict the source or overlook an exception. Grounding can use a pasted document, retrieval, or a live tool.
Take it into your next meeting
Which source should win when two documents disagree about the same policy?
One more concept, understood.
Mark this lesson complete to keep your place.