Writing a Job Description That Does Not Lose Candidates

A job description fails quietly. Nobody writes in to say "I would have applied but you did not mention pay". They close the tab, and your applicant count simply reads lower than it should have, with no indication of why.

That is what makes job descriptions hard to improve. There is no feedback loop. So the only practical approach is a checklist applied before publishing.

What actually costs you applicants

No pay information. This is the big one, and it is worth treating as more serious than the rest combined. It is among the most common reasons candidates skip a posting outright. It also creates late-stage dropouts -- you both spend three rounds discovering that your band and their expectation never overlapped. And pay disclosure is now legally required in a growing number of jurisdictions.

A wide range beats no range. "12-20 LPA depending on experience" is not a weak answer; silence is.

Not describing the work. A list of required skills tells a candidate what you want. It does not tell them what their Tuesday looks like. A short "what you will do" section is the difference between a posting somebody can picture themselves in and a specification.

Requirement bloat. Twenty undifferentiated bullet points are read literally by exactly the candidates you least want to lose -- the ones who will not apply unless they meet every line. Splitting into "essential" and "nice to have", with essentials kept under about eight, widens the pool without lowering the bar. Nothing about the role changes; only who believes they are allowed to apply.

Missing location or work mode. On-site, hybrid or remote, and from where. It is among the first things a candidate looks for, and its absence reads as on-site by default -- so if you are offering hybrid and did not say so, you are being judged as if you were not.

Hype language. "Rockstar", "ninja", "guru", "fast-paced", "wear many hats", "work hard play hard", "self-starter". These describe a mood rather than a job. Experienced candidates read several of them as a warning about workload and process maturity, which is often what they are.

Age-coded wording. "Young", "energetic", "digital native", "recent graduate", "fresh blood". These discourage experienced applicants, and in several jurisdictions using them as selection criteria is unlawful.

Gender-coded wording. "Salesman", "manpower", "chairman", "he/she", "aggressive", "dominant". That masculine-coded phrasing reduces applications from women is a well-established finding in the hiring literature -- general industry knowledge, not something this product measured. The fixes are trivial: "salesperson", "staffing", "chair", "they".

No sense of what happens next. A line about the stages and rough timeline costs nothing to write. Silence after applying is the most common complaint candidates have about hiring processes, and the posting is the cheapest place to pre-empt it.

Length. Under about 150 words a candidate cannot tell what the job involves, so strong applicants self-select out rather than guess. Past roughly 900 words most people do not finish. Somewhere around 300-600 words is where postings get read.

A free grader that shows its working

Rather than asking you to hold nine rules in your head, there is a free tool at /tools/jd-grader. Paste a job description, get a score out of 100, and see every deduction with the specific fix and the exact words that triggered it.

Three things about it are deliberate:

It is not AI. It is a rule-based checklist. If you are deciding whether to rewrite a posting, you need to see which rule caused a deduction so that you can disagree with it. A number produced by a model you cannot inspect is not actionable -- it is just a number that makes you feel bad.

Nothing is uploaded. The check runs entirely in your browser, so an unpublished requisition never leaves your machine. There is no sign-up and no account.

It tells you what it got right too. The passed checks are listed alongside the failures, because a rewrite that fixes one problem and breaks two others is a real and common outcome.

What a score is not

Worth being blunt: this is a checklist, not a prediction. It does not know your market, your compensation band, your employer brand, or how many similar roles are open within ten kilometres of your office. A posting can score 95 and attract nobody, because the salary is below market or the role is genuinely hard to fill in your city.

What the grader removes is avoidable friction. It does not create demand. Anybody selling you a job description tool that claims otherwise is selling you something else.

After the posting

Once the description is decent, the rest of the funnel is where candidates get lost next: an apply form that asks for twelve fields, a careers page that lists fifteen roles with no way to find the two in your city, and a fortnight of silence after applying.

ShortlistAI covers that side -- a public apply link that needs no candidate login, a company careers page you can group by department, AI screening against the actual job description, and stage emails so people are not left guessing. But the grader is free and standalone, and fixing the posting is the cheapest improvement available to you today.

Paste your next requisition into it before you publish. It takes a minute, and the applicants it saves you are the ones you would otherwise never have known you lost.