How to Use ChatGPT to Write Job Descriptions That Attract Great Candidates

Why most job descriptions fail before anyone applies

Pull up the last five job descriptions your team posted. I can guess what they look like. A list of responsibilities copied from the last person who held the role. A bullet list of “requirements” that nobody actually checks. Three paragraphs of company boilerplate that says nothing specific about your company.

The cost of this is well documented. SHRM’s reporting on hiring trends consistently shows that unclear job descriptions are one of the top reasons qualified candidates skip an application, and that posts without a salary range now actively suppress applicant volume in the growing list of US states that require disclosure. [1] Microsoft and LinkedIn’s 2024 Work Trend Index makes the related point that 75% of knowledge workers now use generative AI at work, and 66% of leaders say they would not hire someone without AI skills. [2] Hiring expectations have shifted; most job descriptions have not.

So if your hiring funnel is bringing in noise (or worse, silence), the job description is usually the first thing to fix. ChatGPT can do most of that fixing for you, but only if you stop using it the way most people do.

The wrong way to use ChatGPT for job descriptions

Here is what almost every HR manager I work with does the first time. They open ChatGPT and type: “Write a job description for a marketing manager.”

ChatGPT obliges. You get back something that reads like every other marketing manager job description on Indeed. It mentions “a dynamic individual,” “fast-paced environment,” and “excellent communication skills.” It could be your company. It could be a logistics firm in Düsseldorf. You have no idea.

The problem isn’t ChatGPT. The problem is that you didn’t give it anything to work with. AI tools are mirrors. Generic input, generic output. That’s the single most important rule of using ChatGPT for any writing task, and it shows up clearest in job descriptions.

The 3-part prompt that actually works

Here’s the structure I teach in our corporate workshops. It works for ChatGPT, Claude, or Gemini. The three parts are: role + outcomes + voice.

Role: Tell ChatGPT what kind of writer it is, what kind of company it’s writing for, and who the reader is. Don’t just say “you are an HR professional.” Say something like: “You are a senior recruiter at a 50-person fintech startup. You’re writing a job ad that will be read by mid-career marketers who already get 20 LinkedIn messages a day.”

Outcomes: Tell ChatGPT what the person in this role will actually do, in plain English. Not a bullet list of duties. Outcomes. “In year one, this person will launch our first paid acquisition channel and bring CAC under £80.” That gives ChatGPT something to write toward.

Voice: Paste in 100 to 200 words of your existing brand copy. Your About page, a recent newsletter, anything that sounds like you. Tell ChatGPT to match that tone. This single step is the difference between a JD that sounds like your company and one that sounds like a press release.

You are a senior recruiter at [company name and one-line description]. You are writing a job advertisement for a [role title] that will be read by [target candidate description].
Here is what success looks like in this role:
- In the first 3 months: [outcome 1]
- In year one: [outcome 2]
- The biggest problem this role solves for our team: [problem]
Here is a sample of our company voice. Match this tone:
"""[paste 100-200 words of your brand copy]"""
Write a job description with: a 2-sentence hook, what you'll do, what we're looking for (skills, not years), what we offer (be specific), and how to apply. Avoid corporate buzzwords. Use "you" not "the successful candidate." Keep it under 400 words.

Skills-based requirements, not “5+ years experience”

Here’s a fight worth having with your team. Stop writing “5+ years of experience required.” It’s lazy, and it filters out exactly the people you want.

LinkedIn’s own Economic Graph research found that adopting a skills-first approach expands the candidate pool by up to 10 times globally, and by close to 19 times in the United States, compared to filtering by job title and degree. [3] Most job descriptions still default to time-served requirements. If you want to be in the minority that gets it right, ChatGPT can rewrite requirements as skills in under a minute.

Paste your old job description in and use this follow-up prompt: “Rewrite the requirements section to focus on the skills and outcomes the candidate needs to demonstrate, not years of experience. For each requirement, explain what ‘good’ looks like in one sentence.”

The output forces clarity. “5+ years of email marketing experience” becomes “Can build a segmentation strategy that improves open rates by at least 15% within two campaign cycles.” The second version is testable. A candidate either has done that or they haven’t.

This shift also helps internally. Hiring managers stop arguing about whether someone with 4.5 years of experience is qualified, and start asking whether the person can actually do the work. The same outcomes-first thinking is the backbone of our 4-part prompt formula: tell the AI what good looks like, not just what to do.

Running a bias check on your draft (this takes one minute)

Gendered language in job descriptions is real and it costs you applicants. Words like “ninja,” “rockstar,” “aggressive,” and “competitive” have been shown to reduce the number of women who apply. [4] Words like “collaborative,” “supportive,” and “nurturing” can over-correct in the other direction. The goal isn’t to swing one way or the other. It’s to find neutral language that describes the work.

ChatGPT will do this for you if you ask. Paste in your draft and use this prompt:

Review this job description for biased or exclusionary language. Flag any words that lean masculine, feminine, age-coded, or that might exclude qualified candidates. Suggest a neutral replacement for each one.
Format as a table: original term | what it implies | suggested replacement.

You’ll get back a small table. Read it through. Most of the flags will be useful, some will be too cautious. You stay in charge of the edit. But ChatGPT will catch things you’ve stopped seeing because you’ve written 40 of these.

One caveat: ChatGPT can’t catch industry-specific bias or context. If you’re hiring for a role that has historically been gendered, you still need a human who knows the field to do a final pass. AI is a first-pass tool, not a replacement for your judgment.

What ChatGPT can’t do (and you still have to)

Let’s be honest about the limits. ChatGPT will not:

Check legal compliance in your jurisdiction. Pay transparency laws vary wildly. New York, Colorado, California, Washington and a growing number of US states require salary ranges in postings. [5] The UK and EU have different rules. ChatGPT might know the general shape of these laws, but it won’t know what changed last month. Run every final draft past your legal or compliance contact.

Know your real benefits. If you tell it your company has “great benefits,” it will make up a list. Pension contributions, mental health support, learning budget, parental leave, hybrid policy: these need to come from you, not from ChatGPT’s guess.

Catch role-specific accuracy. A hiring manager once asked ChatGPT to write a JD for a “Senior DevOps Engineer with Terraform experience.” The draft mentioned three technologies the team didn’t actually use. The hiring manager spotted it. You can’t always rely on the candidate to. Always have the hiring manager review the technical requirements line by line.

Replace the conversation with the hiring manager. The best job descriptions come from a 20-minute conversation with the person the candidate will actually report to. ChatGPT can structure that conversation for you (try prompting it: “Give me 10 questions to ask the hiring manager before I write the job description”), but it can’t have the conversation for you.

A full workflow from blank page to posted job in 45 minutes

Here’s the workflow I’d run if I were filling a role tomorrow. Time-box it. Set a 45-minute timer. Don’t perfect it.

Minutes 0 to 10: Open ChatGPT. Use the 10-question prompt above to interview yourself or the hiring manager. Type rough answers. Don’t worry about polish.

Minutes 10 to 20: Paste your answers into the 3-part prompt (role, outcomes, voice). Generate the first draft. Read it once. Note three things you want to change.

Minutes 20 to 30: Use the skills-based rewrite prompt on the requirements section. Use the bias-check prompt on the full draft. Apply ChatGPT’s edits selectively. You’re the editor.

Minutes 30 to 40: Add your real benefits, the actual salary range (or the band you can disclose), and any region-specific legal language. Have the hiring manager sanity-check the technical bits.

Minutes 40 to 45: Post it. Don’t wait for it to be perfect. You’ll get more useful feedback from the first 10 applications than from another hour of editing.

If you want to systematise this across your team, our AI workshops for HR teams walk through this kind of workflow live with your own roles. But the workflow above works on its own, today, with a free ChatGPT account.

Frequently asked questions

Related reading

Your Resume Is Being Read by an AI Agent: What That Means for Your Next Job Application

How to Write Better AI Prompts: The 4-Part Formula Anyone Can Use (With 10 Copy-Paste Examples)

Can a Non-Technical Person Really Learn AI? Yes, and Here’s the 30-Day Path

Sana Mian: Co-Founder, Future Factors AI

Sana is an AI educator and learning designer specialising in making complex ideas stick for non-technical professionals. She has trained 2,000+ learners across corporate teams, bootcamps, and keynote stages. Future Factors offers AI Bootcamps, Corporate Workshops, and Speaking & Consulting for businesses ready to adopt AI without the overwhelm.

More about Sana →