Rewrite Job Descriptions for Clarity and Inclusion

Workflow area: HR & Recruiting · Tier: Core · Hands-on time: ~30 minutes

A bloated JD costs you twice: qualified people self-select out (research consistently shows many candidates — women especially — skip roles unless they meet nearly every listed requirement), and unqualified people apply anyway because the wall of bullets says nothing concrete. AI is a strong first-pass editor here because the three fixes — clarity, requirement discipline, bias language — are pattern problems, and patterns are what models see.

When to use — and when not to

• Use it before every posting goes liveeven a "fine" JD usually loses a third of its bullets and gains a real first-line hook.
• Use it on your whole JD librarybatch-rewriting ten stale descriptions in an afternoon standardizes voice and structure across the company.
• Use it when a role attracts the wrong applicantsmismatched pipelines usually trace back to vague duties and inflated requirements.
• Skip the final call on requirementsAI can flag that "MBA required" looks unnecessary for a coordinator role; only the hiring manager can confirm it. Every cut is a human decision.
• Skip legal boilerplate rewritesEEO statements, notice-period clauses, and compensation-disclosure language follow your jurisdiction and your legal team, not a model's preference.

The recipe

1. 1.

   Collect the inputs. The current JD, plus two answers from the hiring manager: "What will this person actually do in a normal week?" and "Think of your best performer in this role — what did they have on day one?" Those two answers outrank everything in the old JD.

2. 2.

   Run the diagnosis pass first — no rewriting yet.

   Copy

   Here is a job description. Do not rewrite it yet. Diagnose it:
   
   1. JARGON AND VAGUENESS: list every phrase a candidate outside our
      company could not act on (e.g. "drive synergies", "wear many
      hats") — with a plain-language alternative for each.
   2. REQUIREMENT AUDIT: for every requirement, classify it as
      ESSENTIAL (job fails without it on day one), TRAINABLE (could
      learn in 3 months), or DECORATIVE (habit, not need). Explain
      each classification in one line.
   3. BIAS SWEEP: flag gender-coded language (e.g. "rockstar",
      "dominant", "nurturing"), age signals ("digital native",
      "recent graduate"), unnecessary physical/education/pedigree
      requirements, and culture-fit phrases that read as "people
      like us".
   4. MISSING BASICS: is salary range, location/remote policy,
      reporting line, or a concrete description of success absent?
   
   Job description:
   [PASTE JD]
   
   Context from hiring manager — a normal week: [ANSWER 1]
   What the best performer had on day one: [ANSWER 2]
   

3. 3.

   Review the diagnosis with the hiring manager. This is the human checkpoint. Typical output worth acting on:

   Copy

   REQUIREMENT AUDIT:
   - "5+ years in a similar role" — DECORATIVE: the weekly duties
     described are achievable at 2-3 years; the number filters
     career-switchers with equivalent skills.
   - "Advanced Excel including pivot tables" — ESSENTIAL: three of
     five weekly duties involve reporting from spreadsheets.
   - "MBA preferred" — DECORATIVE: no listed duty requires it.
   BIAS SWEEP:
   - "aggressive self-starter" — masculine-coded; try "proactive
     and independent".
   

   Accept, reject, or override each finding — the model proposes, the manager decides.

4. 4.

   Rewrite with the agreed decisions locked in.

   Copy

   Now rewrite the JD applying only the changes we agreed:
   [PASTE YOUR ACCEPTED/REJECTED DECISIONS]
   
   Structure:
   - Title: plain and searchable (what candidates type, not internal
     grade names)
   - Hook: 2-3 sentences on what this role owns and why it matters
   - "What you'll do": 5-7 bullets, each starting with a verb,
     describing a normal week
   - "What you'll need": ESSENTIAL requirements only, max 5
   - "Nice to have": TRAINABLE items, max 3, explicitly optional
   - "What success looks like": 3 outcomes at 6 months
   - Salary range, location, reporting line
   
   Tone: direct, warm, no superlatives. Reading level: accessible to
   a candidate skimming on a phone. Keep our legal/EEO block verbatim:
   [PASTE BLOCK]
   

5. 5.

   Run the candidate-eye test. Ask in the same chat: "Read this as a strong candidate with 80% of the requirements. What would make them hesitate to apply?" Fix anything real it surfaces, then have one actual human outside HR read it in under a minute and tell you what the job is. If they can't, the hook failed.

Make it yours

• Batch mode for the JD library: paste your JD template decisions once, then feed ten old JDs in sequence in one chat — "same structure, same tone rules" — and you get a consistent library instead of ten voices.
• Google Docs workflow: keep the diagnosis table in a comment-enabled doc so hiring managers resolve each finding asynchronously — the accepted decisions become your step-4 paste.
• Internal mobility variant: rewrite the same role twice — external posting and internal-transfer version — from one diagnosis; the internal one swaps the company pitch for team context and keeps requirements identical, which keeps internal candidates on equal footing.

Quality check

• Every remaining requirement is defensibly essentialthe hiring manager, not the model, confirmed each one.
• A candidate can picture a Tuesdayduties describe actual work, not aspirations.
• The bias sweep was reviewed, not just acceptedmodels over-flag and under-flag; a human judged each item.
• Salary and location are statedomitting them costs more applicants than any wording fix recovers.
• Legal blocks are untouchedverbatim from your approved language.