I've written maybe forty job descriptions, and the ones that attracted the right candidates had something in common: they read like a spec, not a billboard. They said what the person would own, what they needed, and what "good" looked like — and they didn't quietly filter out half the applicants with words like "ninja" or "rockstar."

So when I built a generator for this, the goal wasn't "write faster." It was "write the same structure every time so nothing important falls out."

Structure first, words second

The generator asks for role, level, and must-haves, then produces responsibilities and requirements as two separate lists. That separation sounds trivial and is not. Requirements inflate — every manager adds "and also proficient in our internal tool no one has heard of" — and responsibilities shrink. Splitting them forces a check: is this line a thing they'll do, or a thing they must have? The two should be reconcilable, and when they aren't, that's a signal the role is undefined, not a writing problem.

The salary-range prompt is the part I'd never cut. Even if you don't publish a number, the generator nudges you to decide one before you write, because a description written without a number in mind drifts toward vagueness that good candidates filter out anyway.

The inclusive-language check is a lint, not a rewrite

The inclusivity check flags gendered or coded wording — "he," "guys," "aggressive" — the same way a linter flags an unused variable. It points at the line; it doesn't invent a new description for you. I'm careful to frame it as a check on your draft, because the moment a tool claims to "make writing inclusive," it's promising something only a human editor owns.

What it can't catch is structural exclusion: a requirement list that quietly demands five years for an entry role. That's a judgement call the generator surfaces with a prompt, not an automatic fix.

What I learned shipping it

The hardest part was resisting feature creep toward "also write the LinkedIn post" and "also screen resumes." Those are different jobs. JobDesc AI stays scoped to the description because that's where the structure pays off. A tool that does one document well beats a tool that does five badly.

If you build your own, store templates per role family. The second backend role you write should start from the first, not from a blank page — but keep the must-haves editable, because the role always shifts.

The limit

It's a drafting aid. The output is a structured starting point you post after a human review, not a legally vetted document. I don't claim it makes a posting compliant with any hiring law — it improves consistency and flags obvious wording issues, and the final responsibility for what you publish is yours.

When I need a clean draft fast, JobDesc AI is the one I open — structure first, then I edit the must-haves. The inclusivity lint alone has caught a couple of words I'd have shipped without thinking.

FAQ

What does the generator actually produce?
A structured job description: separated responsibilities and requirements, plus a prompt to set a salary range before you write.

Does the inclusivity check rewrite my text?
No. It flags gendered or coded wording like a lint pass. You decide what to change.

Can it make my posting legally compliant?
No. It improves structure and surfaces obvious issues. Compliance with hiring law is a human and often legal review, not something a generator certifies.

Should I still edit the output?
Yes. Treat it as a clean first draft. The must-haves and responsibilities should reflect your actual role, which only you know.