AI 办公提效Extended Use Cases
Chapter 22
22 / 23

Recruiting & Interview Assistant

⏱️ 25 min

Write JDs, screen candidates, build interview outlines, and handle candidate communication

CHAPTER PRACTICE GOAL
01Work problem

Write JDs, screen candidates, build interview outlines, and handle candidate communication

02Reusable output

A reusable work template with realistic input, a defined output format and a human review point.

03Definition of done

Run one real task, verify the critical facts and record the before-and-after time.

The most time-consuming parts of recruiting: writing JDs, screening resumes, creating interview questions, recording feedback, sending response emails. AI can give you frameworks and drafts — you trim and make final calls.

1) Job Descriptions

  • Prompt:
You are a hiring manager. Write a JD based on the following:
Role: xxx, Level: xx
Team background: ...
Required skills / nice-to-haves / work model / location
Output: role highlights, key responsibilities, hard skills, soft skills, performance metrics, salary range (optional), application email. No exaggerated or misleading descriptions.
  • Ask AI for both a "short version (social media)" and "detailed version (company website)" to keep messaging consistent.

2) Resume Screening & Scoring

  • Have AI output "match score + evidence citations + areas of concern + clarification questions."
  • For missing key info, ask AI to generate "questions to ask" rather than guessing.
  • Quick note: include the instruction Do not make biased judgments based on gender/age/school to reduce risk.

3) Interview Guides & Question Banks

  • Generate "structured interview questions + scoring rubric + follow-up paths."
  • For technical/case questions, have AI provide "expected answer points" and "common mistakes."
  • For behavioral interviews: require STAR format examples and a scoring rubric.

4) Interview Notes & Feedback

  • Feed in interview notes/recording summaries and ask AI to output: Summary, strengths/risks, hire recommendation (with reasoning), info gaps, feedback email draft for the candidate.
  • For multiple interviewers' feedback, have AI merge and flag "conflicting points / items to confirm."

5) Offer/Rejection & Candidate Communication

  • Have AI provide "respectful and concise" rejection templates in multiple languages.
  • Offer attachments and sensitive terms must be confirmed by humans. Ask AI to flag "requires HR review" fields.

6) Risk & Compliance

  • Anti-discrimination: instruct AI not to reference gender/age/marital/parental status.
  • Privacy: if resumes or recordings contain sensitive info, use enterprise/private models or create redacted summaries.
  • Documentation: interview conclusions need supporting evidence — avoid "gut feeling" assessments.

7) Practice

Pick a role you're currently hiring for. Have AI output both JD versions, an interview guide, a scoring rubric, and rejection/feedback templates. Use a real resume (redacted) to generate "match score + clarification questions," then adjust manually.


8) Worked Example: From JD to a Structured Interview Decision

Suppose the role requires someone to handle a production incident independently. Translate that requirement into observable behaviour rather than “works well under pressure”:

CapabilityObservable evidenceInsufficient signal
DiagnosisNarrows the scope in sequence and explains what to check firstMerely names monitoring tools
Risk judgementKnows when to roll back, escalate, or stopOptimises only for speed
CommunicationUpdates impact and next steps for different audiencesSays “I communicate well”

Generate questions from the same rubric:

Capability: production incident handling
Create a structured interview pack:
1. Primary question with a concrete but incomplete incident scenario
2. Two levels of evidence-seeking follow-up
3. Behaviour anchors for scores 1, 3, and 5
4. A note field that records only what the candidate actually says
Do not use age, school, gender, nationality, or family status as signals.

Merge Evidence After the Interview

Below are de-identified notes from three interviewers.
Organise by capability:
- direct evidence supporting hire
- risk evidence
- contradictions between interviewers
- information that remains untested
Do not make the final hiring decision or invent missing statements.

9) Fairness and Privacy Gate

  • Remove name, phone, address, and photo before using an external AI service
  • Use the same core questions and scoring anchors for the same role
  • Scores cite concrete answers, not “good feeling” or vague culture fit
  • AI does not make the final hire, rejection, or compensation decision
  • Candidate-data storage and retention follow company policy
  • Rejection communication avoids unverified or potentially discriminatory reasons

10) Failure Modes and Fixes

FailureCauseFix
A match score looks objectiveNo shared scoring anchorDefine capabilities and 1/3/5 behaviour evidence first
AI prefers elite schools or brandsInput contains proxy variablesDe-identify and require job-relevant evidence only
Many questions produce incomparable interviewsEvery candidate gets a different core interviewFix core questions; personalise only follow-ups
Summary sounds overly certainMissing information is filled inSeparate verified, untested, and contradictory evidence

11) Chapter Deliverable

Create a Structured Interview Kit: capability map, core questions, follow-up paths, scoring anchors, and evidence form. For candidate data, continue to AI Security and Ethics.