Appearance
The verification questions layer is one of the layers that feed imper.ai's risk assessment. When enabled, imper.ai asks the user a set of knowledge-based questions during verification and scores their answers to gauge how likely they are the genuine user.
Questions are generated from your connected data sources - for example the user's calendar, directory, or HR records - so a legitimate user can answer them while an impersonator generally cannot.
How answers are scored
Each answer is scored by an LLM that analyzes the response and assigns one of three risk levels:
| Risk level | Meaning |
|---|---|
| ✓ Correct | The answer is completely correct. |
| ✓✗ Partially correct | The answer is close, but not exactly the correct answer. |
| ✗ Wrong | The answer is incorrect. |
The per-question results are combined into an overall questions score, which contributes to the interaction's risk score alongside the infrastructure signals.
Skipped and timed-out questions
Beyond a scored answer, a question can end in one of two other states, both recorded in the transcript:
Skipped - the user chose to skip the question instead of answering it.
Timeout - the user took too long to answer, and the question timed out. A timeout fails the entire verification.
Influences on the score
How much the questions layer contributes, and how sharp its result is, depends on configuration and history:
Verification settings and questions. How many questions are asked and how the policy is configured shape the score - stricter settings and better-covered questions produce a more confident result. See Verification Policies.
Verification history. As a user completes more verifications, imper.ai builds a baseline; later answers are weighed against that history over time.
Viewing the transcript
Every verification keeps a full transcript of the questions asked and the answers given, viewable in the Analysis Report under the Challenge category. The transcript is available to Admin and Member roles; users with the Agent role see the verification outcome and risk score, but not the underlying questions and answers.