Build a Structured Interview Question Bank per Role

Workflow area: HR & Recruiting · Tier: Core · Hands-on time: ~35 minutes

Structured interviews are one of the best-evidenced predictors of job performance — far ahead of free-form conversations — yet most teams skip them because writing good behavioral questions with scoring anchors is tedious. That tedium is exactly the work AI does well. Your job is to define the competencies (that's a hiring decision) and pressure-test the anchors; the model drafts the fifty questions you'd never find time to write.

When to use — and when not to

• Use it when a role opens repeatedlyanything you hire more than once a year deserves a bank; the investment amortizes immediately.
• Use it when multiple interviewers cover the same loopthe bank plus anchors is what makes their scores comparable.
• Use it after a bad hiremost misses trace to a competency nobody actually probed; rebuilding the bank exposes the hole.
• Skip inventing competencies from the job title aloneif you can't state what the role requires, fix the job description first (pb-29); a question bank built on a vague JD tests vague things.
• Skip anything resembling automated candidate evaluationthis playbook structures questions humans ask and score. Do not paste candidate responses in for the model to grade; interview evaluation stays with the panel.

The recipe

1. 1.

   Fix the competencies first — humans only. From the JD and the hiring manager, agree on 4–6 competencies, each with one sentence on what it means in this role. Example for a customer support lead: case-handling judgment, coaching ability, process improvement, stakeholder communication, composure under load.

2. 2.

   Generate the bank, one competency at a time.

   Copy

   Create interview questions for this competency. Role and context:
   
   Role: [TITLE, LEVEL, TEAM]
   Competency: [NAME] — [YOUR ONE-SENTENCE DEFINITION]
   What good looks like in this role: [1-2 CONCRETE EXAMPLES]
   
   Produce:
   1. Three BEHAVIORAL questions ("Tell me about a time...") that
      probe past evidence of this competency
   2. Two SITUATIONAL questions ("What would you do if...") using
      realistic scenarios from our context: [ONE LINE OF CONTEXT,
      e.g. "B2B SaaS support, 4-person team, 200 tickets/week"]
   3. For each question: two follow-up probes an interviewer can use
      when the answer stays vague
   4. SCORING ANCHORS on a 1-4 scale for the competency:
      1 = concerning, 2 = below bar, 4 = strong — describe the
      OBSERVABLE evidence at each level, not adjectives.
   
   Rules: no brainteasers, no trick questions, no questions about
   protected characteristics or anything adjacent (family plans,
   age, health, religion, nationality). Questions must be answerable
   by candidates from any company background, not just our industry.
   

3. 3.

   Pressure-test the anchors — this is where quality lives. A weak anchor says "2 = poor communication." A usable one is observable:

   Copy

   Competency: Coaching ability — scoring anchors
   1 — Blames team members when describing failures; no example of
       developing anyone; probes reveal they did the work themselves
       instead of coaching.
   2 — Gives feedback examples but only corrective and ad hoc; no
       evidence of a development plan or follow-through.
   3 — Concrete example of diagnosing a skill gap, agreeing a plan,
       and following up; outcome stated.
   4 — Multiple examples across different people; adapts approach per
       person; can cite a measurable improvement they coached
       someone to.
   

   Ask the model: "For each question, what would a well-rehearsed but weak candidate say that could fool an interviewer? Add a probe that exposes it." Then delete any question you couldn't answer well yourself from a past role — that's the realism test.

4. 4.

   Run the legal and fairness sweep.

   Copy

   Review the full bank. Flag any question that: could elicit
   protected-characteristic information even indirectly; assumes a
   traditional career path; requires unpaid work ("design us a
   plan"); or is culturally specific to one region. Suggest a
   compliant rewrite for each flag.
   

   Treat the output as a checklist for your HR/legal review, not a substitute for it.

5. 5.

   Assemble the interviewer kit in Google Sheets. One row per question: competency, question, follow-up probes, anchors 1–4, and a blank evidence-notes column. Add a cover row with the rules: same core questions per candidate, score independently before the debrief, evidence quotes required for any 1 or 4.

Make it yours

• Panel splitting: ask the model to divide the bank across a 3-interviewer loop so each competency is owned by one person and nothing is asked twice — "assign competencies to interviewers, 25 minutes each, and produce a per-interviewer one-page script."
• Level calibration: generate the same bank at two levels ("associate" vs "senior") and diff the anchors — the deltas make useful promotion criteria too.
• Debrief template: have the model produce the matching debrief sheet — per-competency scores, evidence quotes, hire/no-hire recommendation last — so the structure survives past the interview itself.

Quality check

• Competencies came from humansthe model expanded them; it did not choose them.
• Anchors describe observable evidenceno adjectives standing alone; two interviewers reading anchor 3 would recognize the same answer.
• Every question passed the "could I answer this?" testrealistic for a good candidate with normal experience.
• The fairness sweep went to a human reviewerHR or legal saw the flagged items and the final bank.
• The kit enforces processindependent scoring before debrief is written into the sheet, not assumed.