AI Integration with Lever Interview Question Generation

The integration surfaces within two primary functional areas of Lever: the Job Requisition configuration and the Interview Kit builder. When a recruiter or hiring manager creates or updates a requisition, an AI agent can be triggered via a Lever webhook to analyze the job description, required skills, and internal hiring guidelines. This agent generates a structured set of behavioral, situational, and technical questions, which are then posted back to Lever's API as a draft Interview Kit, ready for review and assignment to specific interview stages like 'Technical Screen' or 'Hiring Manager Round'.

The workflow is designed for precision and context. The AI model is grounded using your company's historical data—successful hires, role-specific competencies, and past interview feedback—ensuring questions are predictive and role-aligned, not generic. For example, for a 'Senior Software Engineer' req, the system can pull from past scorecard data to emphasize system design and leadership questions over basic syntax. This happens in minutes, turning a manual, day-long research task into a consistent, auditable process that populates Lever's native interview planning tools.

Rollout is typically phased, starting with a pilot team where generated kits require a human-in-the-loop approval step within Lever before they become active. Governance is managed through Lever's existing Permissions and Audit Log, tracking which user triggered generation and which questions were edited or approved. This ensures control and continuous improvement, as rejected or modified questions feed back into the model for refinement. For a deeper look at orchestrating these custom workflows, see our guide on AI Integration with Lever Custom Workflows.

The integration is built on Lever's webhooks and REST API. When a candidate reaches a specific stage (e.g., "Phone Screen" or "On-site Interview"), Lever fires a stage_change webhook. This event payload, containing the opportunity_id and job_id, triggers an AI agent. The agent fetches the full job requisition details—including title, department, required skills, and hiring team notes—via the GET /jobs/{id} API. It also retrieves historical data on successful hires for similar roles using the GET /opportunities endpoint with appropriate filters, creating a context-rich prompt for the question generator.

The core AI workflow executes in a secure, isolated environment. The agent uses the retrieved job and historical data to generate a structured set of questions: behavioral (based on required competencies), technical/skills-based (drawn from the job's requirements), and role-specific situational questions. The output is formatted as a JSON payload and posted back to Lever as a note on the candidate's profile via POST /opportunities/{id}/notes, tagged for easy filtering. Alternatively, for a more integrated experience, questions can be written to a custom interview kit object or appended to the interview guide section of the job posting, making them instantly available to all interviewers scheduled for that role.

Governance is designed-in. All generated questions are logged with a full audit trail linking them to the source job, candidate, and the specific AI model version used. A human-in-the-loop approval step can be inserted before questions are committed to Lever, allowing a recruiter or hiring manager to review and edit. This architecture ensures compliance, maintains data sovereignty (PII never leaves your controlled environment), and scales to support high-volume hiring by processing webhook events through a managed queue. Rollout typically begins with a pilot team, using a sandbox Lever environment to refine prompts and workflows before enabling the integration across the organization.

Production implementations connect to Lever's REST API and webhooks to read job requisition data (role, department, required skills) and write generated questions back to custom fields or note fields. All AI processing occurs in a secure, isolated service layer—candidate PII is never sent to external LLM APIs. Instead, the system uses de-identified job data (e.g., sanitized job descriptions, competency frameworks) to generate questions, which are then associated with the Lever candidate record via the candidate ID. Audit logs track every generation event, linking the job requisition, the user who triggered it, the model version used, and the final output for compliance review.

Rollout follows a phased, role-based approach to manage change and validate quality:

• Phase 1 (Pilot): Enable AI question generation for a single recruiting team and role type (e.g., Software Engineers). Questions are generated as drafts in a custom Lever field, requiring a recruiter or hiring manager to review, edit, and approve before use in an interview guide.
• Phase 2 (Expansion): Expand to additional departments, using feedback from Phase 1 to refine prompts and question templates. Introduce automated quality checks, such as flagging questions that are too generic or that may introduce bias, for human review.
• Phase 3 (Automation): For trusted role profiles, allow auto-approval of generated questions into the official interview plan, while maintaining a human-in-the-loop override. Integrate with Lever's interview scorecards to correlate question quality with hiring outcomes and interviewer feedback.

Governance is built into the workflow. A centralized prompt library manages question-generation templates for different role families (e.g., behavioral for managers, technical for engineers), ensuring consistency and allowing for updates based on hiring team feedback. Access controls (RBAC) determine which users or teams can trigger generation and approve questions. The system is designed for explainability: for any generated question, you can trace back to the source job requisition data and the specific prompt template used, making it easy to audit and refine the process over time.