Build an Interview Evidence Scorecard With AI

AI can turn a job description into a polished list of interview questions in seconds. That speed is useful—and dangerous. A job description often mixes real requirements, vague ideals and inherited wording. If the source is weak, AI simply makes the weakness look organised.

The safer HR workflow is narrower: start from a validated job analysis, use AI to draft a structured scorecard, then have people confirm every competency, question and evidence anchor before interviewing anyone. AI helps with the blank page. It does not select the candidate.

Start with the job, not the candidate

The U.S. Office of Personnel Management defines a structured interview as a systematic way to assess job-related competencies using questions about past behaviour or hypothetical situations. Candidates receive the same predetermined questions in the same order, and responses are evaluated against the same scale.

OPM’s guidance says competency selection should begin with job-analysis data, the critical tasks and the capabilities needed on entry, then be confirmed by subject-matter experts. That sequence gives AI a legitimate input. A CV collection or a model’s guess about “ideal personality” does not.

Worked example: handling a customer trade-off

For a customer success manager, the job analysis may show one critical task: protect a customer relationship while setting a clear boundary when a request conflicts with product policy. “Good communication” is too broad. The scorecard needs observable evidence.

A behavioural question might be: “Tell us about a time you had to say no to a valuable customer while protecting the relationship. What was the request, what constraint applied, what did you say and what happened next?” The evidence anchors help interviewers listen for the same things. They do not guarantee an unbiased decision; they make the reasoning more explicit and reviewable.

You are helping an HR team draft a structured interview scorecard. Use only the approved job-analysis notes below. Propose four job-related competencies that are required on entry. For each, draft one behavioural question and three observable evidence anchors: weak, adequate and strong. Do not infer personality, culture fit, protected characteristics or candidate suitability. Flag any requirement that is vague or unsupported. Return a draft for human review, not a final assessment tool.

[Paste approved, non-sensitive job-analysis notes]

A table with four competencies, one consistent question per competency, three behaviour-based anchors and a list of unclear source requirements for HR to resolve.

The validation meeting matters more than the prompt

The EEOC maintains guidance on the Americans with Disabilities Act and AI used to assess applicants and employees. The practical lesson reaches beyond one jurisdiction: technology does not remove an employer’s responsibility for the assessment process. Local law and company policy still need qualified review.

Bokili helps HR teams practise safe, concrete AI workflows like this one. The transferable skill is not “generate interview questions”. It is knowing how to constrain the tool, inspect the draft and keep consequential judgement where it belongs.