AI output controls
How Clearly AI keeps Review and Workflow output structured, grounded, and repeatable.
Clearly AI uses LLMs inside structured Review and Workflow paths. These controls make outputs more deterministic, repeatable, constrained, and expected. They are about predictable behavior, not security guardrails.
Admins define the review templates, sources, workflow nodes, and provider settings. Users review generated output before relying on it.
Review controls
- Review Templates define the task. Questions, answer types, allowed options, hidden instructions, and compliance assertions come from the review template.
- Outputs are schema-aligned. Clearly AI uses strict structured output validation, then parses responses into the expected answer, summary, compliance, or workflow shape.
- Answers are grounded in sources. Review answers use project sources and relevant Knowledge Base context. When the available sources do not answer a question, Clearly AI marks the information as missing.
- Citations support review. Answers include citations where the review workflow can tie statements back to source material.
- Compliance is evaluated against configured assertions. Status labels come from the review template's requirements and the generated answer, not from an unbounded free-text judgment.
Workflow controls
Workflows run through configured nodes. A workflow can only perform the actions present in its graph, such as creating a review, commenting on a ticket, sending a Slack message, or writing a document through a connected integration.
External write actions use the integration credentials configured for your organization or user. They do not get arbitrary access to unrelated systems.
Admin controls
Admins can keep output predictable by keeping these inputs current:
- Review Templates and compliance assertions.
- Knowledge Base policies, standards, and runbooks.
- Source categories and trust levels.
- Workflow branches and write-back steps.
- Chat admin controls, including organization instructions and Chat Admin access.
- Model provider settings, including BYOK if required.
Human review
Clearly AI is built for security and risk teams to review, edit, regenerate, export, and route results. Review generated answers with citations before approving risk decisions or sending responses outside your organization.