AI Job Descriptions: ChatGPT Prompts That Actually Produce Usable JDs

A blank document. A fuzzy hiring manager intake. A deadline before the role goes live. Every recruiter knows the JD writing tax: staring at a template, copying language from old postings, second-guessing whether you remembered the salary band.

ChatGPT can shorten the work substantially, but only if your prompt is right. A vague prompt like “write me a job description for a software engineer” produces vague output: generic responsibilities, unrealistic requirements, the same cookie-cutter “About Us” paragraph every other listing has.

Rich from the SkillDeck channel walks through a technique that fixes this in his Talent Acquisition series. The trick is that you don’t write the prompt yourself. You use a prompt generator to write the prompt, then run that prompt through ChatGPT. Two tools, one usable JD.

This post covers the prompts that work, the structure they produce, and where writing prompts manually starts to fall apart compared to using a JD-aware tool.

Why direct prompts to ChatGPT produce weak JDs

In short: ChatGPT will produce a generic JD from a one-line prompt, but the output is missing structure, role-specific responsibilities, and the elements hiring managers actually care about. The fix is either a stronger prompt or a prompt generator that builds the structure for you.

Here’s what most recruiters do the first time they try ChatGPT for JDs:

write a job description for an SEO intern, remote, India

ChatGPT will give you something. It’ll have a job title, a few responsibilities, a few requirements. But compare it to what a hiring manager actually wants to publish:

Is there a company summary that sells the role?
Are the responsibilities specific enough that the candidate can self-select?
Do the requirements separate must-haves from nice-to-haves?
Is there a benefits section that matches what your company actually offers?
Is there an application process or call-to-action?

A one-line prompt produces output missing most of this. You spend a chunk of editing time getting it back into shape, which is barely faster than writing from scratch.

Rich’s point on the SkillDeck channel is that asking ChatGPT directly for a JD does technically work, but you won’t get a proper job description that way. The accurate JD comes from running a prompt through a prompt writer first, then pasting the result into ChatGPT.

The prompt is the product. The model is the engine.

The prompt generator workflow

In short: Use a prompt generator like PromptHackers to translate your plain-language requirements into a structured ChatGPT prompt. Copy the generated prompt into ChatGPT. The output is a JD with company summary, responsibilities, requirements, benefits, and application process in the right order.

Rich’s recommended workflow uses two tools:

PromptHackers (prompthackers.com): turns your layman-language requirements into a structured prompt
ChatGPT: executes that prompt

You don’t try to be a prompt engineer. You let a prompt-engineering tool do the engineering.

Step 1: describe the role in plain language

Open PromptHackers. In the input box, write what you want the way you’d say it out loud:

Play a role of JD writer.

Write a job description. I want to hire interns who can handle SEO,
digital marketing, do research, should be smart, write blogs.
Job is remote, work from home. Candidate can apply from anywhere in India.

No magic. No “act as a senior recruiter with 20 years of experience.” Just the requirements in normal English. The prompt generator handles the formatting.

Step 2: generate the prompt

PromptHackers takes your input and produces a structured prompt with role context, format requirements, and constraints already filled in. The generated prompt includes things you wouldn’t have thought to specify:

“Include a company summary”
“Separate required from preferred qualifications”
“Output in markdown with H2 section headers”
“Tone should be professional but engaging”

You don’t write any of that. The generator adds it.

Step 3: run it through ChatGPT

Copy the generated prompt. Paste into ChatGPT. The output is a JD with the structure Rich’s first SkillDeck video on JD writing teaches recruiters to use:

Job title
Company summary
Job description (the role overview)
Responsibilities
Requirements
Benefits
Application process

That isn’t random. It’s the canonical JD structure. A direct one-line prompt skips half of these sections. A generator-built prompt forces ChatGPT to produce all of them.

What makes a strong JD prompt vs a weak one

In short: Weak prompts give ChatGPT a topic. Strong prompts give it a role assignment, an output structure, and constraints. The difference is consistency. Strong prompts produce the same quality output every time. Weak prompts produce a coin flip.

If you want to skip prompt generators and write your own prompts, the difference between weak and strong comes down to four elements:

Element	Weak Prompt	Strong Prompt
Role assignment	Missing	”You are a senior technical recruiter writing JDs for SaaS startups”
Output structure	”Write a job description"	"Output: 1) Title 2) Summary 3) Responsibilities 4) Requirements 5) Benefits”
Constraints	None	”Tone: professional but conversational. Length: 400-600 words. No clichés.”
Context	”for a backend engineer"	"for a Senior Backend Engineer at a 50-person Series B fintech, Berlin hybrid, €80-100k”

Here’s a strong prompt template you can adapt:

You are a senior recruiter writing job descriptions for a [INDUSTRY]
company at [STAGE] stage with [TEAM SIZE] employees.

Write a job description for the following role:

- Title: [TITLE]
- Location: [LOCATION + REMOTE POLICY]
- Salary range: [RANGE]
- Reports to: [MANAGER ROLE]
- Team: [TEAM CONTEXT]

Must-have requirements:
- [REQUIREMENT 1]
- [REQUIREMENT 2]

Nice-to-have:
- [REQUIREMENT 1]

Output structure:
1. Compelling 2-sentence company hook (specific, not generic)
2. Role overview (3 sentences max)
3. Responsibilities (5-7 bullets, action verbs only)
4. Requirements split into "Required" and "Preferred"
5. Benefits (4-6 bullets, real not aspirational)
6. Application process (1 paragraph)

Constraints:
- Tone: confident, specific, no corporate clichés
- Length: 400-600 words total
- No phrases like "rockstar", "ninja", "fast-paced environment"
- Use "you" not "the candidate"
- Avoid gendered language

Save this as a snippet. Replace the bracketed placeholders for each new role. The output stays consistent across roles because the structure is locked in.

SpotGPT: the all-in-one TA prompt

In short: SpotGPT generates a JD plus the artifacts that usually come after (KPIs, screening questions, STAR-method interview questions, LinkedIn post copy, candidate outreach emails) from a single prompt. Skip its Boolean string output; it’s unreliable.

If you want everything a TA process needs in a single prompt, SpotGPT covers more ground than ChatGPT alone.

You log in with your Gmail account and write the same kind of layman-language prompt you’d use in PromptHackers:

I need a job description for HR Manager in SkillDeck knowing
Talent Acquisition, payroll, L&D, etc., with 7 to 9 years of
experience in Bangalore.

The output isn’t just a JD. It’s a TA package:

Roles and responsibilities (the JD itself)
Performance objectives, what the role is expected to achieve in the first 90/180/365 days
KPIs, the measurable success criteria for the role
Screening questions, what to ask in the initial recruiter call
Scenario-based STAR questions for situational interviewing
LinkedIn post copy, formatted for direct posting
A candidate outreach email template for passive candidates

Rich’s take on the SkillDeck channel is that SpotGPT bundles together the artifacts a Talent Acquisition expert typically needs (JD, screening questions, STAR questions, posting copy, outreach email) in a single output. His one caveat: skip the Boolean string output. Boolean is the one thing the tool doesn’t do well. Use a dedicated approach for that — our guide to Boolean search strings for 15 common roles covers it.

The trade-off versus ChatGPT plus PromptHackers: SpotGPT gives you more artifacts at once but less control over each one. ChatGPT lets you iterate on a single output. SpotGPT hands you eight outputs you can pick from.

Before and after: a generic JD rewrite with AI

In short: A weak recruiter-written JD leans on generic adjectives and placeholder responsibilities. The same role, run through a prompt-generator workflow, comes back specific and scannable. Below is a sketch of the difference using bracketed placeholders. Adapt it with your real numbers before publishing.

Here’s the kind of starting JD a recruiter copies from an old posting:

Backend Engineer

We are a fast-paced startup looking for a dynamic backend engineer to join our growing team. The ideal candidate will be a self-starter who thrives in a collaborative environment.

Responsibilities:

Develop and maintain backend systems

Work with the team to deliver features

Write clean code

Requirements:

3+ years of experience

Strong communication skills

Bachelor’s degree preferred

Three problems with that draft: every adjective is generic, the responsibilities are placeholders, and the requirements would match almost any backend role.

Now run the same role through a strong prompt. Fill the bracketed slots with details from your hiring manager and let ChatGPT produce structured output:

{Seniority} Backend Engineer, {Domain or Product Area}

You’ll own the {system area} at a {company size} {industry} company. You’ll work with {team composition} and report to {manager role}.

What you’ll do:

{Concrete deliverable 1: what ships, what it replaces}

{Concrete deliverable 2: performance or scale target}

{Concrete deliverable 3: cross-team collaboration}

{Operational responsibility: on-call, ownership scope}

Required:

{N}+ years building production backend services in {languages}

Direct experience with {domain-specific systems or patterns}

Strong {database / infra} background

Preferred:

{Tooling-specific experience}

{Adjacent technology experience}

Compensation: {salary range} {+ equity if applicable}, {location and remote policy}

The second version is shorter in adjectives but heavier on specifics. Candidates can self-select. Hiring managers can defend it. ChatGPT, with the right prompt and your real role data filled into the brackets, produces this kind of structure by default.

Where the prompt-driven workflow breaks

In short: Prompt-driven JD writing falls apart at agency-style volume, when hiring managers send fragmented intake notes, or when every JD also needs to be parsed back into structured search criteria for sourcing. Past a certain volume, you need JD intake built into your CRM, not a ChatGPT tab.

The PromptHackers + ChatGPT workflow is great for one-off JDs. It breaks at scale. Three failure modes:

Volume. A staffing agency running many active roles ends up running the prompt cycle every time a new role opens. Each cycle means copying intake notes into a prompt generator, copying the generated prompt into ChatGPT, then copying the result somewhere it can be edited and shared. The context switching adds up.

Intake fragmentation. Hiring managers don’t send neat bullet-pointed requirements. You get a Slack thread, a Zoom transcript, an old JD with comments scribbled on it. Reformatting that mess into the kind of clean input PromptHackers wants is the actual work, and the prompt generator can’t do it for you.

Downstream parsing. A finished JD is only the start. You then need to extract the requirements back out of it for sourcing: must-have skills, years of experience, location, salary band, search keywords. Doing this twice (once to write the JD, once to parse it) is wasted work.

Tools like Recrudoc handle the second half of the pipeline natively. Paste a JD (or even raw intake notes), and the JD Parser extracts structured requirements, generates Boolean search strings, and pushes everything into your candidate matching pipeline. The same parsed JD then powers AI scorecards, screening scripts, and outreach messages, with no copy-paste between tabs.

For broader context on where AI is changing recruiter workflows, see our overview of AI recruiting trends in 2026 and the best AI tools for recruiters this year. Once your JD is written, the next problem is converting it into a candidate search, which our guide to translating JDs into AI-powered candidate searches covers.

Quick reference: prompts to save

In short: Three prompts cover most recruiter JD-writing scenarios: a structured intake-to-JD prompt, a JD-to-LinkedIn-post prompt, and a JD-rewrite prompt that fixes generic language without rewriting from scratch.

Save these in a notes app or snippet manager. Adapt the bracketed sections per role.

Prompt 1: intake notes to full JD

Act as a senior recruiter. I have intake notes for a new role.
Convert them into a complete job description with:
1) 2-sentence company hook
2) Role overview
3) 5-7 responsibility bullets (action verbs)
4) Required and preferred qualifications (separated)
5) Benefits (only what the company actually offers)
6) Application process

Tone: confident, specific, no clichés like "rockstar" or "fast-paced".
Length: 400-600 words. Use "you" not "the candidate".

Intake notes:
[PASTE RAW NOTES HERE]

Prompt 2: JD to LinkedIn post

Convert this job description into a LinkedIn post that drives applications.

Format:
- Hook line (1 sentence, attention-grabbing, no emoji)
- 3-line role summary
- 4-5 bullets on what the candidate gets (not what we want)
- Clear application CTA
- 5 relevant hashtags

Length: under 1300 characters total.
Tone: human, not corporate.

JD:
[PASTE JD HERE]

Prompt 3: generic JD rewrite

This job description is too generic. Rewrite it with the same role
and seniority but make it specific.

Replace every cliché ("dynamic team", "fast-paced environment",
"strong communication skills") with concrete details.
Add team size, tech stack, or business context where relevant.
Keep the same structure but cut filler.

Original JD:
[PASTE JD HERE]

These three prompts cover most JD-writing situations. Combine them with PromptHackers for the harder cases, or skip the manual workflow entirely with a JD-aware CRM.

Tired of copy-pasting prompts between PromptHackers, ChatGPT, and your ATS? Try Recrudoc CRM free. Paste any JD or intake notes and get structured requirements, Boolean strings, and candidate matches built in.

Sources

The insights in this article are based on the following industry expert discussion:

“AI in Recruitment | Crafting Job Descriptions in Seconds — ChatGPT, SpotGPT, PromptHackers” — Rich, SkillDeck, YouTube