Most JDs are bad. They lead with the boilerplate "we're a fast-paced, mission-driven team," bury the actual role under benefits language, and use the same generic responsibility bullets every other company uses. AI can fix that — but only if you point it at the right inputs and edit the output the right way.
One-time setup. Load:
— 3–5 anonymized JDs for roles you've recently filled successfully (these are your voice exemplars)
— Your company values, in 2–3 sentences each
— What makes your team different from competitors hiring for the same role
— Standard benefits and comp transparency language (if you have it)
Without these inputs, AI defaults to generic JD shapes. With them, you get JDs that sound like you.
Before you prompt Claude, write down (in plain language) what you actually need:
— What problem will this person solve in their first 90 days?
— Who do they work most closely with?
— What does "great" in this role look like vs. "fine"?
— What are 2–3 specific projects already on their plate?
— What's the realistic compensation range?
These specifics are what turn a generic JD into a JD a real candidate would self-select into.
Draft a job description for [Role] at [Company]. Use the voice and structure of the example JDs in the knowledge base. The role brief is below. Lead with what the role actually does, not with company boilerplate. Use specific projects, not generic responsibilities. Include the comp range we've agreed to. Keep it under 600 words. Avoid: "fast-paced," "mission-driven," "team player," "rockstar."
The "avoid" list is critical. Without it, every AI-generated JD includes the same dead phrases.
AI JDs have tells. Hunt them:
— "In today's rapidly evolving landscape" → kill
— "It is important to note that" → kill
— "You will be responsible for spearheading initiatives" → rewrite as a specific project
— "Strong communication skills" → either specify what kind, or kill
— "Bachelor's degree preferred" → either it's required or it isn't; pick one
Read the draft aloud. Anywhere it sounds like a corporate trade-show banner, rewrite.
The strongest JDs include 3–4 things AI struggles with: specific named projects ("In the first 90 days, you'll own the migration from HubSpot to Salesforce"), explicit comp transparency ("$140K–$170K base + 0.2% equity"), what failure looks like, and one piece of evidence that this is actually a good place to work.
These specifics are what separate JDs that get 200 unqualified applications from JDs that get 20 great candidates.
1. Letting AI write the company description first. Generic AI defaults to "fast-paced, mission-driven." Lead with the role, not the company.
2. Listing 12 "responsibilities" that are actually duties. Real candidates can't tell what they'd actually do. Replace with 3–4 specific projects.
3. Skipping compensation transparency. JDs without comp ranges signal you don't know what you're willing to pay. Every great candidate filters them out.
4. Listing 15 "required skills." Two-thirds are aspirational. Real requirements are usually 4–6. Cutting the rest doubles your applicant quality.
Yes — AI does the drafting; you do the judgment. The compensation, projects, and standards are your decisions. AI just helps you write them clearly.
If you skip the editorial pass, yes. If you do it properly, the JD reads like you wrote it carefully — which is true.
Yes — increasingly important. The best candidates in 2026 want to know they're joining an AI-fluent team. Mention it.
Same workflow, with a critical addition: include 1–2 specific AI workflows the role would touch. Generic "build AI features" descriptions don't attract the people who can actually do the work.