Your job description is your first interview. Most companies bomb it.

A JD is supposed to do one thing: make qualified people want to apply for your role. That's it. Every other use of it — legal cover, internal alignment, hiring manager's wish list — is secondary, and most JDs get written as if those things were the point. The result is a document that reads like a procurement spec and performs like one: technically complete, strategically useless. You get fewer applications than you should, and the ones you get are disproportionately from people who optimize their resume to match the spec rather than people who'd actually do the job well.

Three specific things kill most JDs. AI catches all three in under a minute, if you ask it the right question.

Failure Mode One: Requirements Inflation

This is the most common and the most expensive.

The pattern: the hiring manager wants "someone great," HR adds requirements for legal and consistency reasons, and the final JD asks for 8–10 years of experience, three specific technologies, a degree, industry background, and "a proven track record" in a role where someone with 4 years and the right trajectory would probably do excellent work. Nobody is lying. Everyone is hedging. The result is a requirements list that nobody would actually hire against — but that qualified candidates read and self-select out of.

Tara Mohr's 2014 research on this is still the clearest thing written on the dynamic. Her survey of over a thousand professionals found that when people didn't apply for a role because they didn't meet the stated qualifications, the most common reason — by far — wasn't low confidence. It was that they took the qualifications seriously. They assumed "required" meant required. They didn't want to waste their time on an application that would be filtered out.

This is the trap. Every "required" item you add that isn't actually required costs you qualified applicants — disproportionately the ones who self-evaluate carefully. You're not raising your bar. You're narrowing your pool to a group that's either exactly on-spec or willing to apply to everything regardless of fit. Neither group contains your best hire.

How AI catches it? You paste in a JD and ask the model to separate true requirements from nice-to-haves. Good output will push back on items like specific year counts, specific technology versions, or specific industry experience, and tell you which ones a great candidate could acquire on the job. Most JDs lose 30–40% of their "requirements" under that review. The role doesn't change. The applicant pool widens meaningfully.

Failure Mode Two: Corporate Jargon

The second killer is softer but compounds.

"Rockstar." "Ninja." "Guru." "Fast-paced environment." "Wear many hats." "Self-starter." "Passionate." "10x engineer." These phrases say nothing. They're verbal decoration that signals "we didn't think carefully about this role." Worse, they actively filter out certain kinds of candidates — senior people who've seen enough to find them grating, introverts who don't recognize themselves in the "rockstar" language, anyone whose working style is thoughtful rather than frenetic.

The fix isn't corporate blandness. The fix is specificity. Instead of "rockstar," say "you'll own our conversion funnel end-to-end, working directly with the CEO on strategy and with two engineers on execution." That sentence does actual work. It tells the candidate what the job is, who they'll work with, and what's expected. A rockstar line tells them nothing except that you copy-pasted from a template.

How AI catches it? Paste in a JD and ask the model to flag every phrase that doesn't communicate specific information about the role. Ask it to propose replacement sentences that describe the actual work. Most first drafts have six to ten of these. Most edited drafts have zero.

Failure Mode Three: Biased Language

The third pattern is subtler, better-documented, and more costly than most people realize.

In 2011, Gaucher, Friesen, and Kay published a study in the "Journal of Personality and Social Psychology" analyzing the language of real job advertisements across a large sample of industries. They found that male-dominated fields used significantly more masculine-coded language — words like 'competitive, dominant, aggressive, driven, leader, ambitious' — and that this language measurably reduced women's interest in those roles. Crucially, the reason wasn't that women thought they couldn't do the job. It was that the language signaled "you probably won't belong here."

The mechanism is not controversial. It has been replicated. It also applies beyond gender — language that reads as in-group specific for any dimension (age, background, culture) creates the same effect.

Most JDs have several of these words without the writer noticing. They're not trying to exclude anyone. The words are just what corporate writing sounds like. But the cost is real: your pool narrows, and you're not even aware which candidates decided not to apply.

How AI catches it? This is one of the cleanest AI wins in hiring. Language models are very good at flagging coded language and suggesting neutral alternatives. "Competitive, driven leader who thrives under pressure" becomes "focused and collaborative — comfortable owning decisions in ambiguous situations." Same role. Broader reach.

The Prompt That Does All Three at Once

You don't need three separate prompts. Here's one that runs the full audit:

"Review this job description for three specific issues. For each, flag the exact text, explain the problem, and suggest a revision.

1. Requirements inflation: Which "required" items are actually nice-to-haves? Propose which should move to "preferred" or be cut, and briefly say why.

2. Vague or jargon language: Identify phrases that don't communicate specific information ("rockstar," "fast-paced," "passionate," etc.) and replace them with concrete descriptions of the actual work.

3. Biased or coded language: Flag masculine-coded terms (competitive, dominant, aggressive, driven) or exclusionary phrasing. Suggest neutral alternatives grounded in the Gaucher et al. research.

Output as a table: Issue Type | Original Text | Suggested Revision | Why.

[paste JD here]"

Run this on your next open role before you post it. The first time feels uncomfortable — you'll see how much of what you wrote doesn't hold up. That discomfort is the value. A JD that survives this review will consistently outperform one that didn't, both in application volume and in who applies.

The Reframe

The JD isn't a spec. It's a recruiting document. Its job is to help the right people recognize themselves in the role and feel welcome to apply. Everything else — requirements, responsibilities, qualifications — is in service of that goal, not above it.

Most companies write JDs backwards. They start with what they want, stack up every possible filter, pad it with corporate language to make it sound aspirational, and then wonder why their pipeline is thin. The fix isn't to write marketing copy. The fix is to write with specificity, honesty, and awareness of who reads these things and what they do when they read them.

AI makes the fix fast. The only remaining question is whether you're willing to see what your JDs are actually doing.

Next: Lesson 09 — Sourcing Candidates Your Competitors Miss