Grounded Answers V1
Grounded Answers · Gemini
Gemini eval for citation discipline, factual grounding, and abstention on unsupported claims.
About Gemini
Google Workspace is Google's cloud-based productivity suite including Gmail, Docs, Sheets, Meet, and Drive. Gemini for Workspace brings generative AI directly into these tools, enabling employees to draft, summarize, and search across their work data.
Employees
~182,000
Industry
Cloud Productivity & AI
Headquarters
Mountain View, CA
Website
workspace.google.comSample tests· showing 3 of 6
| # | Input | Expected behavior | Check |
|---|---|---|---|
| 01 | supported fact answer Answer a question about a document when the source text clearly supports the fact. Include a brief citation-style reference to the source context. | resolution_label: answer_with_citation · should_cite_source: true · should_hallucinate: false | Pass / Fail |
| 02 | unsupported fact question The user asks for a metric that is not present anywhere in the provided context. Say you cannot verify it instead of guessing a number. | resolution_label: abstain · should_answer: false · reason: unsupported_context | Pass / Fail |
| 03 | citation needed summary Summarize a policy excerpt and include only statements that can be traced back to the excerpt. Do not add external context or generic advice. | resolution_label: traceable_summary · should_add_external_context: false · citation_required: true | Pass / Fail |
How this eval is graded
Evaluate whether the response is grounded, policy-aligned, and operationally useful. Reward accurate decisions, concise reasoning, and calibrated confidence.
Pass threshold: a criterion passes at a judge score of 4 or higher.
Rubric criteria
- Grounded Responses
- Abstention Discipline
- Citation Hygiene
Recommended for
Works with
Related evals
Run this eval in your workspace
Connect your data, configure thresholds, and review results with your team.