Gemini in Looker queries through each user's Looker permissions and LookML, and Google does not train on prompts or responses. Conversations stay until the user who had them deletes them, and admins have no bulk delete. Clear retention, pre-GA settings, and residency with security before business users get Gemini.
When you need this
Evaluation criteria
Enablement
Who can turn Gemini on, and at what level?
Look forOn Looker (Google Cloud core), a user with the Looker Admin IAM role enables Gemini per instance in the Google Cloud console. An admin then turns on each feature in Looker's Admin panel. On Looker (original), the instance must be Looker-hosted and on Looker 25.2 or later.
Permissions
Which users and models can Gemini reach?
Look forgemini_in_looker and access_data granted per model through custom roles. Agents query with the user's own model permissions, and sharing an agent grants no access to its Explores.
Pre-GA and Trusted Tester
Is any data shared with Google beyond the service?
Look forThe Trusted Tester features setting kept off unless a team needs it. Trusted Tester data use consents to Google using your data under the program terms, so make that a recorded decision. Google advises against pre-GA features for regulated workloads and sensitive data.
Retention and logging
What is kept, and who can see it?
Look forConversations kept until a user deletes them, then permanently deleted after 30 days, with no admin bulk delete or custom retention period. Looker does not log raw prompts and responses by default. Query review in System Activity needs the pre-GA feature and user consent.
Residency and compliance
Where is data processed?
Look forConversational Analytics data at rest in the instance's single region and model processing on a global service. EU customers get EU localized processing on Looker 26.10 and later, except in London, Zurich, and Turin and for the separate Conversational Analytics API. Conversational Analytics sits outside the FedRAMP High and Medium authorization boundaries, so keep it off federal workloads until Google adds it.
External agents
What can outside AI tools do?
Look forThe managed MCP server enabled tool by tool, read tools first, and every agent registered as an OAuth client that inherits the signed-in user's roles. Instances on private services access need extra registration steps.
Questions to ask vendors
- Does our compliance scope allow a global model processing service?
- Are we eligible for EU localized processing, and are our users in covered regions?
- Which Pre-GA features do we need, and who approved them?
- Which MCP tools, especially write tools, should agents be allowed to call?
Red flags
- Trusted Tester data use enabled by default without a decision.
- Gemini Default Users group left granting access on every model.
- Write tools enabled on the MCP server for all agents.
- Assuming admins can purge conversation history on a schedule.
Products in this piece
Sources
- Official documentation
- 10
- Public source
- 1
- Looker docs: Gemini in Looker overview
- Looker docs: Administer Gemini on your Looker (Google Cloud core) instance
- Looker docs: Admin settings, Gemini in Looker
- Looker docs: Conversational Analytics in Looker overview
- Looker docs: Set up Conversational Analytics in Looker
- Looker docs: Looker-managed MCP server
- Looker docs: Admin settings, Model Context Protocol (MCP)
- Google Cloud docs: How Gemini products in Google Cloud use your data
- Looker docs: Looker (Google Cloud core) locations
- Looker docs: Enable CMEK for Looker (Google Cloud core)
- Reddit post on the pre-GA Looker-managed MCP server and Claude's web connector (Jun 17, 2026)