How to Write a Job Description Using AI — Detailed, Inclusive, and Effective

Why AI Job Descriptions All Sound Alike

AI job descriptions sound identical because most people hand the model a job title and nothing else. With no constraints, it fills every gap from the average of every posting it has ever read — and the average job posting is bloated, hedged, and padded with filler.

Watch what happens to one opening line when the input changes and nothing else does.

Same tool, seconds apart. The difference is not the AI. It is that the second prompt told it what the job is for.

That distinction has a cost attached. The longer and more absolute your must-have list, the more good candidates read themselves out of the running before they reach the bottom of the page. Most talent acquisition leads will tell you the same thing: a role with twenty requirements is not a high bar, it is an unclear one. If you are building the wider hiring system rather than fixing one posting, our guide to using AI across the recruitment process covers where this step sits.

How to Write a Job Description Using AI

To write a job description using AI, run the work in two passes: one to build, one to break. Pass one converts your inputs into six fixed blocks. Pass two hands that draft back to the model as a critic and asks it to find coded language, requirement creep, and missing pay detail. Then you review it and post it.

The Draft-Then-Sweep Method is a two-pass writing system: the first pass drafts to a fixed six-block structure, and the second pass audits that draft for biased wording, requirement bloat, and missing pay and duties information.

Splitting the work matters more than it sounds. Ask for a great job description and you get an enthusiastic one. Ask separately for the flaws in a draft, and the model stops defending the text it just produced and starts finding fault with it. Same tool, different job, much better result.

The Six Blocks a Job Description Needs

A job description needs six blocks, in this order: Role Impact, First 90 Days, Core Responsibilities, Must-Have Qualifications, Nice-to-Have, and Pay and Working Pattern. Fix that order once and every posting your organisation publishes stops looking like it came from six different companies.

1. Role Impact — three sentences on why this job exists and what changes because someone does it.
2. First 90 Days — three measurable outcomes the new hire will be judged on by month three.
3. Core Responsibilities — five to seven bullets, each one an outcome the person owns.
4. Must-Have Qualifications — five maximum, every one genuinely non-negotiable.
5. Nice-to-Have — three optional strengths, clearly marked optional.
6. Pay and Working Pattern — salary range, benefits summary, location and remote policy.

Two of the six are where drafts reliably fall apart, and both failures come from the same habit: describing activity instead of ownership.

Block three is the one candidates actually read. Here is the difference a single rewrite makes.

Pass One: Draft the Six Blocks

Pass one is a single prompt that turns five facts into a full six-block draft. The quality of what comes back depends almost entirely on what you put in, so gather the inputs before you open the chat window.

You need these, and nothing more:

• Job title, team, and who the person reports to.
• Why you are hiring — backfill, growth, or a new function.
• What success looks like at 90 days, however roughly you can describe it.
• The skills that are genuinely non-negotiable, written as a short list.
• Salary range, location, and remote policy.

If the hiring manager sent you a wall of unstructured notes instead, let the model do the extraction — but forbid it from inventing the gaps.

You are helping me prepare inputs for a job description. Below are unedited notes from a hiring manager.

Extract and list only these six things:
- Job title
- Team and reporting line
- Why the role exists
- What success looks like in the first 90 days
- Skills stated as genuinely non-negotiable
- Anything about pay, location, or working pattern

For anything the notes do not cover, write MISSING and move on. Do not infer, do not fill gaps, and do not add industry-standard requirements.

Notes:
[paste the raw notes here]

This is the workhorse. It sets the role, the structure, and — most importantly — the constraints. The negative constraints at the bottom are doing more work than anything else in it.

You are a senior talent acquisition partner writing an internal job description.

Write it in exactly six blocks, in this order:
1. Role Impact — three sentences on why this role exists and what changes because of it.
2. First 90 Days — three measurable outcomes the person is judged on.
3. Core Responsibilities — five to seven bullets. Each states what the person owns, decides, or produces. Not what they assist with.
4. Must-Have Qualifications — maximum five, drawn only from the non-negotiable list below.
5. Nice-to-Have — maximum three, clearly marked optional.
6. Pay and Working Pattern — salary range, benefits summary, location and remote policy.

Here is everything I know about the role:
Job title: [title]
Team and reporting line: [team, manager]
Why we are hiring: [backfill / growth / new function]
Success at 90 days: [your notes, however rough]
Non-negotiable skills: [list — these are the only ones allowed in block 4]
Pay, location, working pattern: [range, location, remote policy]

Rules:
- Do not add any qualification, tool, or years-of-experience figure I did not give you.
- If something is missing, write MISSING: [what you need from me] instead of guessing.
- Plain professional English. Do not use rockstar, ninja, guru, world-class, fast-paced, dynamic, or synergy.
- No sentence longer than 25 words.

Two details make this prompt work where a generic request fails. The MISSING instruction stops the model papering over thin inputs, which is where invented requirements come from. Capping block four at five forces a real decision about what the job needs, rather than letting the model pad the list on your behalf.

If your drafts still come back generic, the problem is usually upstream — our guide to giving AI enough context covers what to add. The prompt’s shape follows the Role, Goal, Context, Format structure we use across the site.

If you want the prompts in this article as a saved file rather than scrolling back to copy them, they live in the free AI work templates library.

Pass Two: Sweep the Draft Clean

Most people learning how to write a job description using AI stop at the end of pass one, which is exactly where the avoidable damage gets published. Pass two is a separate conversation with a different instruction: stop writing, start criticising. You hand the model the draft it just produced and ask it to find four specific categories of problem — not to improve the text, and not to rewrite it.

You are an inclusion and compliance reviewer. Do not rewrite the document.

Review the job description below and return a table with four columns:
Phrase | Problem | Suggested replacement | Which block it is in

Check for exactly these four things:
1. Masculine-coded or competitive wording — aggressive, dominant, relentless, crush, warrior, and similar.
2. Age-coded wording — digital native, recent graduate, high energy, young team, and similar.
3. Requirement creep — any must-have that is not genuinely needed to do this job in month one.
4. Missing information — no salary range, no location or working pattern, no equal opportunity statement.

Then, below the table, list every responsibility bullet that describes a task rather than an outcome.

Do not soften your findings and do not pad the table. If a line is fine, leave it out.

Job description:
[paste your pass one draft here]

The table format matters. Ask for prose feedback and you get three paragraphs of encouragement with the real problems buried in the middle. Ask for a table and every flagged phrase has to stand on its own line with a justification next to it — which also makes it obvious when the model is reaching.

Run on a real draft, the change is less subtle than people expect.

Pay, Duties, and the Legal Layer

Two things in a job description carry legal weight in ways the wording alone does not: whether you disclose pay, and whether the duties you list actually match how the role is classified. AI can find the gaps in both. It cannot tell you what your obligations are.

What the wording rules cover

The wording question is the more visible one. The US Equal Employment Opportunity Commission’s guidance on prohibited employment policies and practices is direct about job advertisements: an ad may not show a preference for, or discourage, applicants on protected grounds — and it gives “recent college graduates” as an example of phrasing that can discourage older applicants. That is exactly the category pass two is built to catch.

Why the duties block matters

The duties question is quieter and gets missed more often. The US Department of Labor’s Fact Sheet #17A on white-collar exemptions states the principle plainly: job titles do not determine exempt status, and an employee’s specific duties and salary have to meet the regulations. A vague responsibilities block is not just weak writing, then — it is the part of the document that would have to do the work if the classification were ever questioned.

Check this job description for two things only.

1. Disclosure gaps. The role is based in [location] and can be performed from [locations or remote scope]. Tell me whether the posting states a salary range, a benefits summary, and the pay basis. Flag anything absent. Do not tell me what any law requires — I will confirm that separately.

2. Duties clarity. Rewrite block 3 so every bullet states what the person decides, owns, or produces, rather than what they help or assist with. Be specific about the level of independent judgement involved.

Return only the flagged gaps and the rewritten block 3.

Job description:
[paste your draft here]

Before any of that, decide what is going into the chat window at all. Compensation detail is one of the most sensitive categories most HR professionals handle, and consumer AI accounts are not the place for all of it. Our guide to using AI safely with work data covers the account-level differences.

When AI Job Descriptions Go Wrong

Almost every failure in this workflow traces back to one of five causes, and each has a fix that takes under a minute.

That last row is the one that costs people the most time. When a single block is wrong, the instinct is to regenerate the document — and you lose three good paragraphs to fix one bad one.

Rewrite block [number] only. Leave every other block exactly as it is and return only the rewritten block.

The problem with it: [describe the specific issue in one sentence].

Keep the same length and the same constraints as before. Do not reintroduce buzzwords.

This is the same diagnostic habit that applies to any AI output — identify which part of your input caused the gap, then fix only that part. Our guide to refining AI output without starting over works through the method properly.

What AI Cannot Decide for You

The sweep is good at finding things that are demonstrably in the text. It is useless at the judgement calls that actually determine whether this hire works out, and being clear about that line is what keeps the workflow honest.

There is also a limit worth naming directly, because it gets oversold elsewhere: running a bias sweep does not make a job description unbiased. It catches the wording patterns a language model can recognise. It cannot see that your must-have list quietly encodes a career path only one kind of candidate tends to have. That is a conversation with the hiring manager, not a prompt.

Once the description clears that gate, it becomes the source document for everything downstream. The public advert is a separate piece of writing with a different job to do, and turning the description into a job posting is worth treating as its own step rather than pasting the internal version onto a job board. The same six blocks also feed straight into interview questions built from the role, which is how you end up assessing candidates against what you actually published.

Write the six blocks once for a role you hire for repeatedly, and pass one becomes a two-minute job forever after. That is the whole of how to write a job description using AI without publishing something you would not apply to yourself — and the sweep is the part you never skip.