Job Descriptions With AI: A Practical Guide for Hiring Teams

AI can turn a hiring manager’s rough notes into a clean, well-structured job description in minutes. What it cannot do is know your team, your pay range or what the job really involves. The drafts that work come from a person who supplies those facts in writing, gives the model a clear structure, and then reviews every line before it goes live. This guide shows how to do that, step by step.

TL;DR

• Give the AI real inputs: top outcomes, daily work, must-have skills, reporting line, location and pay range. Without them it fills gaps with plausible filler.
• Prompt with a role, a template, a tone and negative instructions (no jargon, no clichés, no inflated requirements).
• Review the draft for invented facts, biased wording and padded “must-have” lists before publishing.
• Writing the description is low risk compared with using AI to screen or rank candidates, which is where laws such as NYC Local Law 144 apply.
• Save your best prompt and style rules so every new posting sounds like your company.

Table of Contents

• What AI is good at in a job description (and what it is not)
• What a good job description should accomplish
• Step 1: Collect the inputs before you prompt
• Step 2: Write a prompt that gives the model structure
• Step 3: Review the draft against your notes
• Step 4: Check for bias and exclusionary wording
• Step 5: Know where legal rules start
• Our perspective: treat the job description as a shared source of truth
• Build a repeatable workflow with AmmarAI
• A reusable checklist before you publish
• Common mistakes to avoid
• FAQ
• Sources

What AI is good at in a job description (and what it is not)

A job description has two jobs. It tells a candidate what the work is, and it tells your own team what you are hiring for. AI helps most with the part people dread: turning scattered notes into a readable document with a consistent structure. SHRM’s guidance on AI for job description drafting frames it the same way: AI can convert a supervisor’s notes into a clean description, and you review the final version to make sure it reflects your organization, audience and purpose.

Where it falls short is facts. A writer at Rework describes AI models as strong on writing and structure but weak on facts, often filling gaps with plausible filler, which is why intake notes need to be written down before you prompt. If you ask for a “Marketing Manager job description” with no other detail, you will get a generic document that could belong to any company. It may also list software, certifications or years of experience nobody on your team asked for.

A useful rule: let AI own the wording, and let humans own the content. The model decides how a sentence reads. You decide what is true.

What a good job description should accomplish

Before you open any tool, decide what the finished page has to do. Rework’s guidance says an effective description covers the top outcomes of the role, the real daily work, the must-have skills and the pay range. That is a good test for your own draft.

In practice, a strong posting does five things:

1. Says what success looks like. “Grow qualified leads from webinars” tells a candidate more than “responsible for lead generation.”
2. Describes the actual week. Meetings, tools, handoffs, who you work with.
3. Separates must-haves from nice-to-haves. Short lists attract more applicants than sprawling ones.
4. Shows the pay range and location setup. Candidates screen on these first.
5. Sounds like your company. A posting that reads like every other posting does not help you stand out.

If the AI draft misses any of these, the problem is usually in the input, not the model.

Step 1: Collect the inputs before you prompt

This is the step most people skip, and it decides the quality of everything after. The research behind this guide points to the same list of details to give an AI tool: top outcomes, real daily work, must-have skills, reporting line, work location, pay range, and company style rules or a sample posting (Rework). A short, plain guide to how to write a job description covers the same ground if you want a second checklist for your hiring managers.

Here is an intake template you can paste into a shared doc and send to the hiring manager:

Notice what is not in the table: boilerplate. Equal opportunity statements, benefits language and legal text should come from your HR or legal team’s approved wording, not from the model.

Step 2: Write a prompt that gives the model structure

SHRM’s approach is to prompt with the job title, the organization, a template and the supervisor’s notes. Indeed’s hiring guidance adds that strong prompts define a role, specify responsibilities, outline qualifications, set the tone, and use negative instructions to keep out jargon, clichés and unrealistic requirements.

Put those together and you get a prompt like this:

You are a recruiter writing for [company], a [one-line description]. Write a job description for a [title] reporting to [manager]. Use these sections: summary, what you will do, what you bring, what we offer, pay and location. Use only the facts below. If something is missing, write [MISSING] instead of guessing. Tone: friendly, direct, plain English. Do not use buzzwords, “rockstar” or “ninja” language, or requirements that are not in my notes. Here are the notes: [paste intake].

The “write [MISSING]” instruction is the single most useful line. It turns invention into a visible gap you can fill, instead of a believable sentence you might miss.

If you want to get better at this skill in general, our guide on writing better AI prompts walks through the same principles of role, context, constraints and format. And if you would rather not build the prompt from scratch each time, our AI prompt generator can help you structure one.

Step 3: Review the draft against your notes

SHRM’s guidance is direct on this point: before publishing, check facts and details against the information you supplied, and confirm that nothing was invented or assumed. AI can assist, but the human makes the final call.

A fast review pass looks like this:

• Read it beside your intake notes. Every duty and requirement in the draft should trace back to a line you wrote. Highlight any that do not.
• Delete unrequested requirements. Common culprits are a degree, a specific number of years, a tool nobody uses, or “strong leadership skills” for a non-management role.
• Check the numbers. Pay range, team size, location, travel and schedule are things a model may soften or invent.
• Read it aloud. If a sentence sounds like a brochure, rewrite it in the way the manager would explain the job to a friend.
• Have the hiring manager sign off. The person who will manage the hire should confirm the daily work is accurate.

A worked example

Say a hiring manager’s notes read: “Need someone to run our email newsletter and social, 2 posts a week plus weekly email, works with designer, reports to me, remote in US time zones, $55,000 to $65,000.” An unguided AI draft might add “3+ years of experience in B2B SaaS marketing,” “proficiency in HubSpot, Hootsuite and Google Analytics” and “a bachelor’s degree in marketing.” None of that came from the notes. In your review you strike all three, or ask the manager whether any are genuinely required. The resulting posting is shorter, more honest and likely to attract a wider group of applicants.

Step 4: Check for bias and exclusionary wording

Models learn from text written by people, and that text carries patterns. Rework notes that models can reproduce those patterns, so you should review wording and insert equal opportunity statements from approved text, not model output. The UK government’s guidance on inclusive job descriptions makes the broader point that AI models can repeat biases found in historical recruitment data, so employers should review and adjust outputs to keep recruitment non-discriminatory. That guidance is from the UK, but the practice it describes applies anywhere.

A practical bias check has four parts:

1. Run an inclusive-language checklist. Look for gendered, ageist or ability-coded terms. Examples include “digital native,” “young and energetic,” “must be able to stand for long periods” when it is not essential, or “aggressive” and “dominant” as praise.
2. Prefer skills to credentials. Compare each requirement against objective, skill-based criteria rather than years of experience or degrees (Rework).
3. Shorten the must-have list. The GOV.UK guidance recommends reducing long must-have lists and using behavior-based criteria to attract a broader pool of applicants. If you have ten must-haves, ask which three would make you reject a candidate who lacked them.
4. Swap in approved statements. Paste your company’s equal opportunity and accommodation language from HR. Do not let the model improvise it.

A useful trick is to ask the AI to critique its own draft: “List any phrases in this posting that could discourage qualified applicants from applying, and explain why.” Treat the answer as a prompt for your own judgment, not a verdict.

Step 5: Know where legal rules start

Writing a job description with AI is a different activity from using AI to evaluate people. Most of the hiring-specific AI regulation targets the second one. But the lines can blur if the same tools write the posting, screen the applications and rank the candidates, so it helps to know the landscape.

According to a law firm briefing on AI in recruitment, the EU AI Act classifies recruitment and selection systems as high risk and attaches obligations including technical and organizational measures, while New York City’s Local Law 144 requires bias audits and candidate notice. The same briefing notes that UK employers must comply with existing discrimination law, including the Equality Act 2010. A recruiting-industry summary adds that the EU rules reach beyond Europe’s borders, so agencies placing candidates into the EU can be caught, with high-risk breaches carrying fines of up to €15 million or 3% of global turnover.

Across jurisdictions, the common themes are transparency, fairness, testing for algorithmic bias and meaningful human oversight in selection (Eversheds Sutherland).

What this means in practice for a US small business:

• Drafting copy with AI and reviewing it yourself is generally the low-risk end of the spectrum.
• Screening or ranking applicants with automated tools is where audit and notice rules may apply, depending on where you hire. If you recruit in New York City or the EU, talk to an employment attorney before you switch such a tool on.
• Keep a human in the decision. Whatever the tool does, a person should make and be able to explain every hiring decision.

If you do add AI to screening, it is worth learning how to audit AI resume screening with simple tests so that you can see whether it treats similar candidates consistently.

Our perspective: treat the job description as a shared source of truth

We build AmmarAI as a single workspace where text, images, video and voice work live together with one history and one brand voice, and our view on job descriptions comes from that way of working. A posting is not a one-off document. It is the starting point for the ad copy, the careers page, the interview scorecard, the recruiter outreach and the onboarding plan. If each of those is written from a different draft, they drift apart, and candidates notice.

So we suggest three habits. First, write the intake notes once, in plain language, and reuse them for every downstream asset. Second, save your style rules, such as tone, banned words and approved boilerplate, so each new posting starts from your standards instead of a blank prompt. Third, resist the temptation to let the tool decide what the job is. AI is excellent at making a clear idea readable. It is poor at deciding what your company needs, and that decision should stay with the hiring manager.

The same discipline applies beyond hiring. Speed is the easy part. The value comes from inputs you trust and a review you do not skip.

— Ahmed

Build a repeatable workflow with AmmarAI

If you write postings more than a couple of times a year, a repeatable setup saves more time than any single clever prompt. In AmmarAI you can use our job description generator to turn your intake notes into a first draft, keep your brand voice and instructions in one place, and then reuse the same details for job ads, candidate emails and careers-page copy. Our page on AI for recruiting teams shows how job ads, screening admin and communications fit into the same workspace. If you want to reuse your best instructions each time, AI personas let you save them.

AmmarAI has a Free plan at $0 per month, so you can try it on a real posting first. Paid plans are Starter at $9.99 per month, Professional at $29.99 per month and Ultimate at $59.99 per month. You can compare what each includes on the pricing page.

A reusable checklist before you publish

Print this, or paste it at the bottom of your hiring template:

• [ ] Every duty and requirement traces back to the hiring manager’s notes.
• [ ] No “[MISSING]” markers left in the text.
• [ ] Must-have skills are limited to what you will actually screen on.
• [ ] Degree and years-of-experience requirements are justified, or removed.
• [ ] Pay range, location and schedule are accurate and match what the recruiter will say on a call.
• [ ] Language checked for gendered, ageist or ability-coded terms.
• [ ] Equal opportunity and accommodation statements come from approved HR text.
• [ ] Hiring manager and HR have both signed off.
• [ ] The final version is saved as the source for ads, outreach and interview questions.

If you also use AI to write or tidy any other candidate-facing material, apply the same rule: check claims before you trust the output. Our piece on checking AI content claims explains why verifying specifics matters more than any score a tool gives you.

Common mistakes to avoid

Prompting with only a title. The result is generic by design. Add the notes.

Accepting the first draft. The first output is a starting point. Ask for a shorter version, a plainer version, or a version with the nice-to-haves removed.

Letting requirements grow. AI is happy to add “strong communication skills” and five more bullets. Every extra line narrows your applicant pool.

Skipping the pay range. Candidates look for it. Leaving it out, or letting the model make one up, damages trust.

Copying the same posting for different roles. Similar jobs still have different outcomes. Start from a fresh intake each time.

Confusing drafting with deciding. The tool can write the posting. It should not decide who gets hired.

FAQ

Sources

• How to Use AI for Job Description Drafting and Refinement (SHRM)
• Writing Job Descriptions With AI: Inclusive, Accurate and More (Rework)
• How To Write Job Descriptions Using ChatGPT (Indeed)
• Make job descriptions inclusive (GOV.UK)
• AI and recruitment: five top tips for protecting the integrity of your global hiring processes (Eversheds Sutherland)
• Global regulatory trends in AI recruitment (Eversheds Sutherland)
• AI Act for recruiters: where the rules apply in 2026 (Recruitly)

Recommended

• How Does an AI Resume Builder Work?
• AI for Recruiting: Job Ads, Screening Admin, Comms
• AI Resume Builder: Create a Polished Resume
• Job Description Generator: Write Postings Fast