Write a job description that filters for the right operator

An outcome-driven job description that makes the right people lean in and the wrong people bounce, so your applicant pool is smaller and better before you screen a single resume.

What you'll have when you're done

A job description built around what the person will own and achieve, not a wish-list of years and degrees, plus a second AI pass that flags any requirement quietly screening out good candidates. The result: fewer, better-fit applicants, a lighter screening load, and a JD that is more inclusive and more legally defensible at the same time. The filtering starts at the top of the funnel, where it is cheapest.

A vague JD is a volume problem you create for yourself

The lazy JD ("rockstar marketer, 5+ years, fast-paced environment") does two bad things at once. It attracts everyone, so you drown in unqualified applicants. And it leans on proxies (years, degrees, "digital native") that screen out capable people for reasons unrelated to the work, which is both a quality miss and a legal risk.

A sharp JD is a filter. When it describes the actual outcomes the role owns, the right operators recognize themselves and apply, and the wrong ones self-select out. I learned this the expensive way, by posting a lazy "rockstar, 5+ years" JD, getting 250 applicants, and realizing most of them were responding to the vagueness rather than the job. The volume felt like success. It was the opposite: I had built myself a week of screening to undo a problem I created in the posting. AI is good at two halves of this: drafting the outcome-based version from your inputs, and then pressure-testing it for requirements that create adverse impact. You get a better funnel before screening even begins.

What you need first

• Your last great hire's actual wins in this role (or the closest role), the real outcomes that mattered.
• The first-90-day outcomes you want this person to own.
• Two or three JDs you admire as format references.
• A Claude Project (a business plan if any of your inputs are sensitive).
• Your salary range, which is now legally required to post in many states.

Step-by-step

Step 1Feed the AI outcomes, not a wish-list

Give the model the raw material that produces a real JD:

Draft a job description for [role], built around OUTCOMES, not credentials.
Inputs:
- What my best hire in this role actually achieved: [paste]
- What this person must own and achieve in their first 90 days: [paste]
- JDs I admire for tone/structure: [paste 1-2]
Lead with what they'll own and accomplish. Keep "requirements" to genuinely
job-related must-haves. Include this salary range: [range].

What comes back leads with the work, which is what the right operator is scanning for. The difference is stark side by side, illustrative, same role:

Before (credential wish-list): "Seeking a rockstar marketing manager. 5+ years experience, bachelor's degree required, digital native, thrives in a fast-paced environment. Must be a self-starter."

After (outcome-driven): "In your first 90 days you'll own paid acquisition across Meta and Google, get our blended CAC under $X, and stand up weekly performance reporting the leadership team trusts. You've run a six-figure monthly ad budget before and can point to what you changed and what it returned."

The second one is shorter and much harder to fake. The wrong applicant reads it, thinks "I've never owned a budget like that," and self-selects out. The right one reads it and recognizes their last job. You did your first round of filtering before a single resume hit your stack.

Step 2Run the bias-and-inflation pass

This is the step most people skip and the one that protects you. Ask the model to attack its own draft:

Now review this JD for requirements that could screen out qualified people
without being job-related: degree requirements that aren't truly necessary,
"X years" thresholds that are really proxies, and coded language
("digital native," "recent grad," "young and energetic," "native English").
Flag each and suggest a job-related alternative.

Degree inflation and coded language are classic sources of adverse impact; catching them here is far cheaper than defending them later.

Here is the shape of what the pass returns, illustrative:

• "Bachelor's degree required" → not job-related to running paid acquisition; many strong performance marketers are self-taught. Suggest: remove, or "degree or equivalent hands-on experience."
• "5+ years" → a proxy for skill; some three-year operators outperform eight-year ones. Suggest: replace with the capability itself ("has independently owned a six-figure monthly ad budget").
• "Digital native" → coded age language that can signal age discrimination. Suggest: cut.
• "Native English speaker" → national-origin risk, and rarely the real need. Suggest: "fluent written and spoken English."

Each flag comes with the why and a job-related replacement, so you are not just deleting words, you are swapping a proxy for the real requirement underneath it. That is the move that widens the pool and lowers the liability in the same edit.

Step 3Cut the requirements list in half

Take the flags seriously. Most JDs list "requirements" that are really nice-to-haves, and every unnecessary requirement shrinks your pool and tilts it. Keep what the job genuinely demands; move the rest to "bonus" or cut it. A shorter, sharper requirements list pulls a better, broader pool. There is a well-documented asymmetry behind this: in the often-cited Hewlett-Packard finding, many qualified people (women especially) will not apply unless they meet nearly every listed requirement, while others apply at far lower thresholds. So every nice-to-have you leave sitting in the requirements list is quietly screening out the exact candidates who would have cleared the bar that actually matters.

Step 4Read it as the candidate you want

Final pass: would your ideal operator read this and think "that is me, and that sounds like real work"? If it reads like a generic corporate template, push the AI to make the outcomes more specific and concrete. Specificity is what makes the right person lean in.

How you'll know it's working

Your applicant volume drops and your applicant quality rises, the opposite of the usual trade. The screening stack you hand to the resume workflow is smaller and more on-target. And your JDs stop carrying the quiet liabilities (degree inflation, coded language) that a sharp eye, or a plaintiff's lawyer, would catch.

When it breaks

• Still too many unqualified applicants. The outcomes are not specific enough, or requirements crept back in as proxies. Push for concrete, role-specific outcomes.
• The AI writes corporate mush. Your example JDs were generic. Feed it ones with real voice, and your best-hire's actual wins, not a job-board template.
• It strips a requirement that genuinely matters. Job-related requirements stay; the pass is for un-related proxies. Tell it which must-haves are truly essential and why.
• You skip the bias pass to save time. Don't. It is the cheapest legal insurance in the hiring set, and it improves the pool at the same time.
• The salary range it wrote is a huge band. A $60K-to-$200K range tells applicants you do not know what the role is, and it deters the good ones. Give it your actual, tight range; many states now require you to post one anyway.
• The bias pass flags a requirement you genuinely, legally need. Some roles carry bona fide requirements (a specific license, legal work authorization, a security clearance). Tell the model which ones are legally mandated so it stops flagging them and focuses on the proxies that are not.

Make it yours. Outcome-based works for every role, but the outcomes differ by altitude. For an hourly or operational role, the concrete bar might be throughput or accuracy ("process 40 orders an hour at under a 1% error rate"), not a quarterly objective. For an executive role, it is the strategic mandate ("get the company to a repeatable sales motion by Q3"). Tell the model the level and it pitches the outcomes at the right height.

Where this fits in your harness

This is the very front of the hiring pipeline, upstream of everything. A JD that filters well means the resume screen starts with a better stack, and interview scorecards evaluate people who were the right fit to interview. Get the top of the funnel right and every stage below it gets easier.