A compliance-focused insurance company is building a Copilot Studio agent to answer customer questions about policy coverage. The legal team requires that the agent only answer from pre-approved knowledge sources, that it never infers or synthesize answers beyond what is explicitly stated in those sources, and that all AI-generated content can be traced back to a specific document. Which Generative answers configuration best meets these requirements?
Select an answer to reveal the explanation.
Short Explanation
Classic mode with citations is the compliance-safe setup—it's like having a lawyer who only quotes exact contract paragraphs, with footnotes, rather than paraphrasing from memory. You keep the AI grounded in approved documents and can trace every answer back to a source.
Full Explanation
Copilot Studio's Generative answers node in Classic mode offers more controlled, predictable behavior than Orchestration mode. In Classic mode, the developer explicitly specifies which knowledge sources the agent may draw from, and the AI generates answers grounded exclusively in those documents. When citation display is enabled, the agent surfaces references to the specific source documents used in its answer, enabling traceability required for compliance environments.
The configuration that uses Classic mode in the Generative answers node with specific knowledge sources configured, and enables citation display so users can see the source document for each answer, is correct because Classic mode restricts the agent to configured, approved knowledge sources (for example, specific SharePoint libraries, public URLs, or uploaded files) rather than allowing broad source selection. Combined with citation display, every answer is traceable to a specific policy document, satisfying the legal team's requirements for source attribution and scope control.
The configuration that uses Orchestration mode with multiple knowledge sources so the agent can select the most accurate source for each question is wrong because Orchestration mode's dynamic source selection is the opposite of what compliance requires. While it produces more flexible answers, the AI may draw from multiple sources or make inferences beyond explicit document content, which violates the never infer beyond stated content requirement. Traceability to a specific document is also harder in Orchestration mode.
The configuration that disables the Generative answers node entirely and uses only manually authored Question and Message nodes for all responses is wrong because it eliminates the scalability benefit of AI-powered question answering. Large insurance policy corpora would require thousands of manually authored question-answer pairs to cover all possible customer questions—this is not operationally feasible and is an exam anti-pattern.
The configuration that uses Orchestration mode and configures a custom content moderation policy to block responses that reference non-approved sources is wrong because content moderation is a reactive filter for harmful or inappropriate content, not a source restriction mechanism. It cannot determine whether a response came from an approved versus non-approved source, nor can it enforce grounding in specific documents.
Exam tip: Classic mode equals controlled, scoped, grounded in specified sources (compliance-friendly). Orchestration mode equals dynamic, flexible, potentially broader source selection (innovation-friendly). For regulated industries or strict legal requirements, Classic mode with citations is the correct answer pattern.