Looker conversational analytics reaches general availability
Google made conversational analytics in Looker generally available, grounded in the LookML model, with multi-Explore questions and an explanation of how each answer was calculated.
Looker's agent now shows its work. Conversational analytics in Looker became generally available on November 13, 2025, seven months after it entered preview, and it can explain how it calculated an answer.
What shipped
- Grounded in LookML. Answers come from the measures and joins already defined in the Looker model.
- Up to five Explores can be used to answer one question.
- Shared agents and multi-turn conversations, so a team can reuse a configured agent and follow up on an answer.
- "How was this calculated?" shows the logic behind an answer.
Why it matters
A person will only act on an agent's number if they can see where it came from. Here the answer traces back to definitions someone on the data team wrote and reviewed.
The limit is the scope of those definitions. A semantic model is written for reporting: measures, dimensions and joins. When a question needs a business rule that lives in a policy document or in someone's head, such as when a customer counts as churned, the model has nothing to ground on.