Start with the work, not a list of clever questions
A structured interview asks candidates a consistent set of job-related questions and assesses their answers against criteria chosen in advance. AI can help draft those materials quickly, but it cannot decide which skills matter most for your team. Begin by listing the role’s main responsibilities, the skills needed to carry them out, and the evidence an interview could reasonably reveal. Keep requirements distinct from preferences. For example, explaining a technical decision may be essential for a support role, while experience with your exact ticketing tool may simply be convenient.
Choose a manageable number of competencies, such as problem solving, collaboration and task-specific knowledge. For each one, write a brief description of what successful work would involve. This gives the AI a defined task and gives your team a way to reject plausible-sounding questions that do not test anything important.
Ask AI for a draft you can inspect
Provide the role summary and competencies, then specify the output format. One useful prompt is: “Draft two structured interview questions for each competency below. For each question, state what it assesses, suggest one neutral follow-up, and describe what a strong answer might demonstrate. Use plain language, avoid personal or hypothetical trick questions, and do not assume candidates have used our particular tools.” Include only information you are comfortable entering into the AI system; a generic role description is often enough.
Review the questions for overlap and relevance. Behavioral questions can invite candidates to describe past work, while situational questions can ask how they would approach a realistic task. Either type can be useful, but neither should depend on a candidate knowing unstated details about your workplace. Replace broad prompts such as “Are you a team player?” with questions that invite observable evidence, such as how the candidate handled a disagreement over priorities. Make follow-ups consistent enough to preserve comparability while allowing interviewers to clarify an answer.
Build a rubric around observable evidence
A scoring rubric should describe what interviewers are looking for before they hear any answers. Ask AI to propose a small scale for each competency, but revise vague labels such as “excellent communication.” A more useful description names evidence: the candidate explains their approach, identifies a relevant trade-off, and shows how they would check whether it worked. The rubric should also distinguish a limited answer from one that addresses the question in depth without requiring a single preferred story or personality style.
For example, a problem-solving rubric might use three anchored levels: “limited evidence” when the answer gives no clear method; “sufficient evidence” when it describes a workable method and a way to check the outcome; and “strong evidence” when it also considers constraints and explains how the approach could change if new information appears. Decide how to handle unanswered questions and what weight, if any, each competency receives. Do not let AI assign weights simply because they look precise; those choices should reflect the role.
Test the materials before using them
Run a short review with the people who will interview candidates. Ask whether each question maps to a stated competency, can be answered by someone from a different but relevant background, and fits the time available. Have interviewers score a sample answer independently, then compare which evidence they used. If scores differ widely, the rubric may need clearer anchors rather than more elaborate wording.
Check questions and follow-ups for requests for sensitive personal information or assumptions about a candidate’s background. Replace questions that invite disclosure about family circumstances, health or other matters unrelated to the work. If your organization has hiring policies or applicable requirements, have the appropriate person review the materials. AI-generated text is a draft, not a substitute for that review.
Keep AI out of the final judgment
During interviews, ask the planned questions, take notes on job-related evidence, and score each competency against its rubric before discussing overall impressions. Leave room to record uncertainty: an answer may not provide enough information to score confidently. A structured process supports more consistent discussion, but it does not eliminate judgment or guarantee a fair outcome.
Be cautious about putting candidate names, transcripts or detailed answers into an AI tool. Check your organization’s data-handling rules and the tool’s settings before using it with interview material. AI may help organize de-identified notes if permitted, but a polished summary can omit context or make weak evidence sound decisive. Interviewers should verify any summary against their own notes and remain accountable for the hiring decision.
