Playbook: HR and Recruiting

HR runs on communication and pattern-matching: job descriptions, screening, onboarding questions, policy explanations, difficult emails. AI speeds up all of it.

It's also the area where careless AI use does the most harm. A biased screening step doesn't make one bad decision — it makes the same bad decision hundreds of times, invisibly.

---

1. The line you don't cross

AI can assist hiring and people decisions. It must never be the sole decision-maker for hiring, promotion or firing.

Why:

• Models learn from historical data, and history contains bias — who was hired, promoted, paid more. Well-known cases include a recruiting model that downgraded CVs mentioning women's colleges and clubs.
• AI can't assess context, potential, or the reasons behind a career gap.
• Automated rejection with no human review creates legal and reputational risk, and data protection laws give people rights over how their personal data is used.

Every AI-assisted step needs a named human reviewer and a way to explain the decision.

2. Job descriptions that widen the pool

Unnecessary requirements and coded language quietly discourage good candidates — "rockstar", "aggressive", "young and dynamic", or ten "must-haves" that are really nice-to-haves.

Review this job description for inclusive, clear language.

1. List gendered or exclusionary words and suggest neutral
   alternatives
2. List corporate jargon a junior candidate might not understand
3. Separate genuine must-haves from nice-to-haves; question any
   requirement (degree, years of experience) that isn't essential
   to doing the job
4. Flag anything that could be age, gender or disability
   discriminatory under Indian law
5. Rewrite it: under 400 words, specific about the actual work,
   salary range placeholder included

[PASTE JD]

Clearer, bias-checked descriptions attract a wider pool of qualified candidates — which is the point.

3. Screening with AI, carefully

If you use AI to help review CVs:

1. Remove identifying details first where possible — name, photo, age, gender, address — to reduce bias in both the AI and the reviewer.
2. Score against written, job-related criteria only, agreed before you see any CVs.
3. Require evidence: a quote from the CV for every score.
4. A human reviews every result, including rejections — not just the top of the list.
5. Audit periodically: compare shortlist rates across groups where you lawfully can.

Evaluate this anonymised CV against the criteria below.
For each criterion: score 1-3, and quote the evidence from the CV.
If there's no evidence, score 1 and say "no evidence" — do not
infer. Do not consider anything not in the criteria.

Criteria:
1. [e.g. Hands-on Python for data analysis]
2. [e.g. Managed a project end-to-end]
3. [e.g. Stakeholder communication]

CV: [PASTE]

Check whether your applicant-tracking system's AI features, or your jurisdiction, add further rules.

4. The onboarding knowledge bot

New joiners ask the same questions: How do I claim medical insurance? When is payroll? Who approves leave? A bot answers them instantly and frees HR for the human issues.

1. Gather the current handbook, leave and benefits policies, and IT setup guide. Remove anything confidential.
2. Build it in your company's approved tool (Custom GPT for Teams/Enterprise, Copilot Studio, Gemini Gem, etc.).
3. Instructions:

You answer onboarding questions for [Company] employees using only
the uploaded documents.
- Quote the policy name and section for every answer.
- If the documents don't answer it, say so and direct them to
  [HR contact].
- Never give advice on personal situations (health, disputes,
  performance, complaints) — direct those to a person in HR.

1. Test with 20 real questions from recent joiners before launch. Review unanswered questions monthly to improve the documents.

5. Human warmth in hard messages

AI is useful for structuring a rejection, a policy change announcement or a difficult performance note. The warmth must be yours.

Draft a rejection email for a candidate who reached the final round.
Be kind and specific: mention [2 genuine strengths from the
interview]. Explain briefly that we chose a candidate with more
[specific experience]. Offer to keep their details for future
roles, with their consent. Under 150 words. No clichés.

Then personalise it. For layoffs, grievances, health or performance issues, write it yourself and have it reviewed — never paste the details of these cases into an unapproved AI tool.

---

⚠️ Common mistakes

• AI as sole decision-maker in hiring or firing.
• Screening on criteria that aren't job-related, or letting the AI infer them.
• Uploading employee records to unapproved tools.
• A knowledge bot with outdated policies.
• Sending AI-drafted sensitive messages without human editing.

---

What's next: from people operations to patient care — AI for healthcare and wellness professionals.