Job Posts That Pull in Winners

Writing Better Job Descriptions with AI

Part 1: Why Most Job Descriptions Fail, and How AI Fixes the Root Cause

In 2021, Unilever's talent acquisition team ran an internal audit on their job postings. What they found was uncomfortable: descriptions for similar roles across different business units read like they'd been written by completely different companies, because they had been. Each hiring manager had copied from an old posting, tweaked a few lines, and posted it. Some descriptions were four years old. Others listed requirements that had nothing to do with how the role actually functioned. The result? A flood of mismatched applicants, longer time-to-fill numbers, and frustrated hiring managers who blamed HR for sending them the wrong candidates. The postings weren't broken in an obvious way. They were broken in the slow, invisible way that comes from nobody owning the quality of the writing itself.

Unilever's fix wasn't to hire more recruiters. They restructured how job descriptions were created, using standardized frameworks, clearer ownership, and eventually AI-assisted drafting tools to bring consistency back to the process. The result was faster posting cycles and a measurable improvement in applicant relevance scores, which their team tracked by comparing hiring manager satisfaction ratings before and after the change. The lesson wasn't that AI is magic. It's that job descriptions are a writing problem before they're a recruiting problem, and writing problems respond well to structured, repeatable processes, exactly what AI tools are built to support.

This is the tension at the heart of this lesson: most HR professionals and hiring managers are skilled at understanding roles, but they're not trained copywriters. They're asked to produce a high-stakes document, one that represents the company's brand, attracts the right candidates, and filters out mismatched ones, under time pressure, with no template, and often no feedback loop to know whether it worked. AI tools like ChatGPT Plus and Claude Pro don't replace your judgment about what a role requires. They give you a writing partner who can take your rough notes and produce a structured, compelling draft in under two minutes.

The Principle Hidden Inside Unilever's Problem

Here's what Unilever's audit actually revealed: the quality of a job description is almost entirely determined by the quality of the input that goes into writing it. When a hiring manager sits down to write a posting, they usually start with what the last person in the role did, not what the next person needs to do. They copy language from other postings without checking whether that language is accurate or effective. They default to corporate jargon because it feels safe and professional. None of this is laziness. It's what happens when a skilled professional is asked to perform a writing task they've never been trained for, without a clear brief, in a hurry.

Think of it this way: if you asked a brilliant colleague to write a proposal for a client they'd never met, they'd ask you a dozen questions first. What does the client care about? What's the budget? What are they afraid of? What would make them say yes immediately? A good AI tool works the same way, but only if you give it the right brief. The problem most HR teams run into when they first try using ChatGPT to write job descriptions is that they type something like 'write me a job description for a marketing manager' and get back something generic and unusable. That's not an AI failure. That's a brief failure. The principle is simple: garbage in, garbage out, and the fix is learning what a good brief looks like.

A good brief for a job description answers five questions before you ask the AI to write anything: What does this person actually do day-to-day? Who do they work with, and what decisions can they make alone? What does success look like after 90 days and after one year? What skills are genuinely required versus nice-to-have? And what kind of person thrives in this team's culture? You don't need to write paragraphs for each question. Even rough bullet points give the AI enough material to produce a structured, specific draft that sounds like it was written by someone who knows the role, because, through your input, it was.

A Different Industry, the Same Problem: Healthcare Hiring

Intermountain Health, a large nonprofit health system based in Utah, faces a recruiting challenge that looks nothing like Unilever's on the surface, but shares the exact same root cause. Their HR team posts hundreds of clinical and administrative roles per month. The clinical roles require precise language for licensing and compliance reasons. The administrative roles, things like patient services coordinator, billing specializt, or health unit coordinator, often get the least attention and the most recycled language. A 2022 internal review found that several administrative postings had not been substantially updated in over three years, despite the roles changing significantly due to new software systems and expanded telehealth responsibilities. Candidates were applying based on job descriptions that no longer matched the actual job.

When Intermountain's HR team piloted AI-assisted drafting for administrative roles, the most immediate benefit wasn't the writing quality, it was the conversation the process forced. To give the AI a useful brief, recruiters had to go back to hiring managers and ask the questions they'd been skipping: What does this person actually do in week one? What software do they use daily? What's the biggest reason someone fails in this role in the first six months? Those answers hadn't been captured in years. The AI draft was just the output. The real value was the structured input process that made recruiters gather better information before they wrote anything. This is a pattern you'll see across every industry: AI improves the output by improving the discipline of the process that feeds it.

Comparing the Old Way to the AI-Assisted Way

Click to enlarge

A Third Example: Retail Management Hiring at Scale

Using Copilot inside Word, their HR team created a master brief template that regional recruiters filled in with local context, store size, team headcount, any regional specifics. Copilot then generated a consistent draft that regional managers could review and lightly adjust. The core language stayed standardized; the local detail made it feel relevant. This approach. AI for the structure and consistency, humans for the local knowledge and final judgment, is the model that works at every scale, from a five-person startup posting its first operations role to a Fortune 100 team posting five hundred roles a month. The AI handles what it's good at. You handle what only you know.

Putting It into Practice: What This Looks Like on a Real Monday Morning

Say you're an HR manager at a regional accounting firm, and a partner just told you they need to hire a Client Services Coordinator, a role that's been vacant for six weeks and is causing real strain on the team. Your instinct is to pull the last posting from your ATS and update the date. That's the fastest path to a mediocre result. Instead, spend ten minutes with the partner asking those five brief questions: What does this person do daily? Who do they work with? What does success look like at 90 days? What's genuinely required vs. preferred? What kind of person fits this team? Write down their answers in bullet form, don't worry about making them pretty.

Now open Claude Pro or ChatGPT Plus and paste those bullets in with a simple instruction: 'Write a job description for a Client Services Coordinator at a mid-size accounting firm based on these notes. Include a job summary, key responsibilities, required qualifications, preferred qualifications, and a brief culture note. Tone should be professional but approachable, not stiff.' In about ninety seconds you'll have a clean, structured draft. Read it. You'll almost certainly find one or two things that need adjusting, a responsibility that's missing, a requirement that's overstated. Make those changes. Then, before you post it, paste the draft back into the AI and ask: 'Does this description contain any vague language, gendered phrases, or requirements that might unnecessarily limit our applicant pool?' You'll get a specific, actionable list of suggested edits.

The whole process takes thirty to forty minutes instead of two to three hours. But more importantly, the output is better: it's specific to how the role actually works today, it uses language that attracts the right candidates rather than repelling them, and it's been reviewed for equity issues that most manually written postings never get checked for. That's not a minor efficiency improvement. That's a fundamentally better document, produced faster, by a professional who doesn't have to be a trained copywriter to get a professional result. And it's repeatable, every role, every time, with the same structured approach.

Key Lessons So Far

1. Job descriptions are a writing problem before they're a recruiting problem, and writing problems respond to structured, repeatable processes.
2. The quality of your AI output is directly determined by the quality of your brief. Five focused questions about the role produce dramatically better results than a vague prompt.
3. AI tools like ChatGPT Plus, Claude Pro, and Microsoft Copilot don't replace your knowledge of the role, they convert your knowledge into a well-structured document faster than you can write it from scratch.
4. The process improvement is as valuable as the output improvement. Asking the right questions to build your brief surfaces role information that often hasn't been updated in years.
5. A two-step approach, use AI to identify gaps in your brief first, then draft, consistently outperforms going straight to a full draft with incomplete information.
6. Consistency at scale (like Target's approach) and quality at speed (like Intermountain's administrative roles) are both achievable with the same AI-assisted framework.
7. Always run a completed draft through a second AI prompt specifically checking for vague language, gendered phrasing, and inflated requirements, this step takes five minutes and catches problems that manual review routinely misses.

The Hidden Cost of Vague Requirements

In 2022, Unilever's talent acquisition team ran an internal audit of their job postings across 50 open roles. What they found was uncomfortable: the average job description contained 14 requirements listed as "essential", but when hiring managers were asked which requirements actually predicted success in the role, they consistently named four or five. The rest were either inherited from old postings, copied from competitor listings, or added as hedges. Candidates were being screened out for missing qualifications that nobody truly needed. The pipeline was artificially narrow, and diverse candidates, who research consistently shows are more likely to self-select out when they don't meet every listed requirement, were disappearing before anyone even saw their resume.

Unilever's solution wasn't to rewrite every job description from scratch. It was to build a structured review process where hiring managers answered a simple set of questions before any posting went live: What does this person actually do on day one? What does success look like at 90 days? Which requirements are genuinely non-negotiable, and which ones could be learned on the job? When those answers fed into the job description drafting process, eventually supported by AI writing tools, the average number of "essential" requirements dropped from 14 to six. Application rates increased. Candidate quality, by hiring manager ratings, went up. The job descriptions had become honest.

The principle here is deceptively simple: a job description fails when it describes the ideal candidate rather than the actual job. AI tools are exceptionally good at helping you catch this. They can flag when a list of requirements reads like a wish list rather than a role specification, suggest more precise language for vague terms like "strong communication skills," and restructure a bloated posting into something a real candidate can read in under three minutes and understand exactly what they're signing up for. But the AI can only work with what you give it. The quality of your input determines the quality of the output, every single time.

Turning a Brain Dump Into a Structured Posting

Marcus runs a regional sales team for a mid-sized logistics company. When a territory manager position opened up, he sat down to write the job description and produced what most hiring managers produce: a dense, rambling document that was part wish list, part internal org chart explanation, and part boilerplate copied from the last time the role was posted, three years ago, before the company had shifted entirely to a new CRM platform. The posting listed "experience with legacy freight management systems" as a requirement. The company no longer used those systems. Nobody caught it.

Marcus took his rough notes, bullet points about what the role involved, who it reported to, what a good quarter looked like, and what had gone wrong with the last person in the seat, and pasted them directly into Claude. He asked it to reorganize the information into a structured job posting with a role summary, responsibilities, requirements, and a section on what success looks like in the first 90 days. The AI didn't just reformat his notes. It flagged a contradiction: he'd listed "independent self-starter" as a requirement but also described a role involving daily check-ins with a regional director and weekly team calls. Claude suggested the role might better suit someone who "thrives with clear accountability structures and regular team collaboration", which was actually far more accurate.

That kind of structural feedback, noticing when the requirements contradict the described work environment, is something a busy hiring manager rarely has time to catch on their own. AI tools don't get fatigued by the fifth reread of a document. They don't have the organizational context that makes certain assumptions invisible to insiders. They read your text the way a stranger would, which is exactly how your best candidates will read it. Marcus published the revised posting, and for the first time in two hiring cycles, he didn't receive a single application from someone who was clearly confused about whether the role was remote or field-based. The clarity had done its job.

A Different Industry, The Same Problem

Healthcare recruiting faces a version of this challenge that's even more acute. When Northwell Health, one of New York's largest hospital systems, began reviewing job postings for nursing and allied health roles, they found that clinical credential requirements were often listed inconsistently across identical roles posted at different facilities. One posting for a registered nurse position required "BLS certification," another listed "CPR certification," and a third said "Basic Life Support (BLS/CPR) required." These were the same requirement written three different ways. Applicant tracking systems were flagging candidates as unqualified because a keyword wasn't matching, not because the candidate lacked the actual credential.

Using Microsoft Copilot integrated into their document workflows, Northwell's HR team began running all job postings through a standardization prompt before publication. The AI would check for inconsistent terminology, flag credential abbreviations that needed expansion, and suggest language aligned with their internal taxonomy. This wasn't about replacing the clinical expertise of the hiring teams, it was about eliminating the administrative noise that was making identical roles look different to both candidates and recruiting software. In a sector where a single unfilled nursing position can cost a hospital an estimated $40,000 to $60,000 in agency staffing and overtime, getting the posting right the first time matters in dollars, not just principles.

What AI Catches That Humans Miss

Click to enlarge

Small Business, Big Stakes: The Founder's Version

Priya runs a 12-person digital marketing agency in Austin. She'd never had a formal HR function. When she needed to hire her first dedicated account manager, she wrote the job description herself in about 20 minutes, posted it on LinkedIn, and waited. What she got back was a flood of applications from people with backgrounds in customer service and retail management, not agency account management. The posting had described the role's interpersonal demands clearly but said almost nothing about the actual deliverables: managing campaign performance data, running client calls, writing status reports, and coordinating with creative teams on timelines. Candidates with the right instincts had no idea what the job actually required day-to-day.

Priya rewrote the posting using ChatGPT Plus. She described the role out loud in a voice memo, transcribed it, and pasted the transcript as her prompt. She asked the AI to identify what was missing from a candidate's perspective, what would someone need to know to decide whether this role was right for them? The AI's response surfaced five gaps: no mention of the reporting structure, no clarity on client-facing expectations, no description of the tools used (the agency ran everything through Asana, HubSpot, and Google Looker Studio), no indication of whether the role was hybrid or remote, and no information on how performance would be measured. Priya added all five. Her next round of applications was smaller in volume but dramatically better matched. She made a hire in three weeks instead of eight.

Matching Tone to Your Actual Culture

One of the subtler things AI can help with is tone calibration, making sure the language of your job posting actually reflects what it's like to work at your organization. This matters more than most hiring managers realize. A startup that uses formal, corporate language in its postings will attract candidates who expect formal, corporate environments. A law firm that uses casual, breezy language will attract candidates who are surprised to find a structured, hierarchical workplace. The mismatch isn't just an aesthetic problem, it's a retention problem. When people arrive expecting one culture and find another, they leave. The job posting is the first honest signal you send about what working there is actually like.

AI tools handle tone calibration well when you give them clear direction. You can paste in two or three examples of internal communications, a team email, a Slack announcement, a paragraph from your employee handbook, and ask the AI to match that voice in your job posting. You can also go the other direction: paste in your drafted job description and ask the AI to assess what kind of work environment the language implies. Does it sound collaborative or hierarchical? Formal or casual? Urgent or measured? The AI will give you a clear read, and you can adjust accordingly. Notion AI and Claude are both particularly good at this kind of tonal analyzis because they process context across longer documents without losing track of the overall register.

There's also the question of inclusivity. Research from Textio, a company that has analyzed tens of millions of job postings, found that certain word patterns consistently correlate with lower response rates from women and underrepresented candidates. Words like "dominant," "aggressive," and "ninja" skew response demographics measurably. Words like "collaborative," "develop," and "support" tend to attract more diverse applicant pools. AI tools trained on large language datasets have absorbed some of these patterns and can flag potentially exclusionary language when prompted to do so. This isn't about political correctness, it's about not accidentally narrowing your talent pool through word choices you weren't even conscious of making.

What This Section Has Shown

1. Inflated requirement lists are a structural problem in most organizations, not a writing problem, and AI helps surface the gap between what's listed and what's actually needed.
2. Pasting rough notes or voice-memo transcripts into an AI tool is a legitimate and effective starting point. You don't need a polished draft to get useful output.
3. AI catches contradictions between stated requirements and described work environments, a type of internal inconsistency that's nearly invisible to writers who know the role well.
4. Standardizing job posting language across an organization, especially for credential requirements in regulated industries, is a high-value, low-effort AI use case.
5. The 'candidate perspective' prompt is one of the most practical techniques in AI-assisted job description writing. It shifts the AI from writer to evaluator.
6. Tone calibration using internal communication samples helps ensure your posting attracts candidates who'll actually fit your culture, reducing early attrition.
7. Inclusivity review through AI is about expanding your talent pipeline, not about compliance. The language of a posting shapes who applies before any human ever reviews a resume.

From Generic to Irresistible: Making Your Job Description Work as a Sales Document

In 2019, Textio published research showing that job descriptions written with gendered or exclusionary language received up to 25% fewer applications from qualified candidates. Unilever noticed this firsthand when their recruiting team audited postings for entry-level supply chain roles. The descriptions were accurate, every required skill was listed, but they read like legal disclaimers. Words like 'dominant,' 'aggressive growth targets,' and 'high-pressure environment' were quietly repelling the very candidates Unilever needed. Their talent acquisition team rewrote the postings using more inclusive, specific language, and application rates from underrepresented groups climbed significantly within two hiring cycles. The job hadn't changed. The words had.

The principle buried in Unilever's experience is one most hiring managers miss entirely: a job description is not a compliance document. It is the first conversation you have with a candidate who doesn't know you yet. Every phrase either opens a door or quietly closes one. Generic requirements like 'strong communication skills' or 'team player' tell a qualified candidate nothing useful, and signal to them that you haven't thought carefully about who you actually need. AI tools can help you replace that vagueness with specificity, and replace exclusionary defaults with language that accurately reflects your culture without narrowing your candidate pool unnecessarily.

This is where AI earns its place in the recruiting workflow, not by writing job descriptions for you wholesale, but by acting as a sharp editor and strategic thinking partner. You bring the role knowledge, the team context, and the hiring goals. The AI helps you stress-test your language, sharpen your clarity, and catch patterns you've stopped seeing because you're too close to them. The result is a posting that works harder in every direction: attracting more qualified applicants, filtering out poor-fit candidates earlier, and giving your hiring managers a cleaner signal from day one.

Click to enlarge

How a Retail Chain Rewrote 200 Postings in a Week

A regional retail chain with 47 stores across the American Midwest faced a recurring problem: their store manager postings were attracting candidates who looked great on paper but burned out within six months. The HR director, working with a consultant, used ChatGPT to audit the existing postings. They pasted each description into the chat and asked the AI to identify requirements that were vague, skills that were duplicated, and any language that implied a specific cultural archetype. The AI flagged phrases like 'must thrive in chaos' and 'relentless focus on metrics', accurate descriptions of the role, but ones that selected for a very narrow personality type that tended to peak fast and crash hard.

The team used those flags as a starting point for a real conversation about what the role actually demanded. They weren't trying to make the job sound easier, they were trying to describe it more honestly. 'Comfortable managing competing priorities across a fast-moving floor' replaced 'thrives in chaos.' 'Uses daily sales data to coach your team' replaced 'relentless focus on metrics.' Same job. Cleaner signal. Applications from candidates with longer tenure histories at comparable retailers increased by roughly 30% over the following quarter. The AI didn't solve the hiring problem. It helped the humans see the problem clearly enough to solve it themselves.

A School District That Stopped Losing Teachers to the First Interview

A suburban school district in Ohio was experiencing a painful pattern: strong candidates were accepting interviews, then withdrawing after the first conversation. Exit surveys revealed the problem, the job descriptions had made the roles sound straightforward, but the interview revealed complexity the posting had glossed over. A fifth-grade teaching position described as requiring 'differentiated instruction experience' didn't mention that the school used a specific literacy framework, that the class sizes averaged 28 students with a 20% IEP rate, or that the role included co-teaching with a special education specializt three days a week. Candidates felt misled, even though nothing in the posting was technically false.

The HR coordinator used Google Gemini to help rebuild the postings from the ground up, starting with structured interviews with current teachers in each role. She fed those interview transcripts into Gemini and asked it to extract the five most important things a candidate should know before applying. The AI surfaced specifics the coordinator had taken for granted, the co-teaching model, the literacy framework, the IEP percentage. Adding those details to the posting didn't shrink the applicant pool; it shrank the dropout rate. Candidates who applied knew what they were walking into, and the district stopped losing people between application and offer.

The Nonprofit That Started Competing with Corporate on Talent

A mid-sized nonprofit focused on workforce development in Chicago had always struggled to attract experienced program managers away from the corporate sector. Their postings led with mission, which mattered to them, but buried the role details and soft-pedaled the professional growth angle entirely. A development director used Claude to rewrite three key postings, specifically asking the AI to help balance mission appeal with professional credibility signals. Claude suggested leading with the scope of the role, budget managed, team size, stakeholder reach, before introducing the mission context. The postings started reading like opportunities for serious professionals, not volunteer positions with a salary attached.

Applications from candidates with private-sector program management backgrounds doubled within 60 days. More importantly, the quality of conversations changed. Candidates arrived knowing the role was substantive, which meant interviews could focus on fit rather than convincing people the job was real. The nonprofit hadn't increased salaries or changed their benefits. They had simply written a description that took the candidate seriously. That's what AI-assisted editing can do when you use it as a thinking partner rather than a content machine, it helps you see your own organization the way an outsider does, and close the gap between what you offer and what you're communicating.

Putting It All Together in Your Next Hire

The most effective workflow isn't to hand a blank page to an AI and ask for a job description. It's to gather your raw material first, talk to the hiring manager, interview someone currently in the role, pull the last performance review for the position, and then use AI to organize, sharpen, and pressure-test what you've collected. Claude and ChatGPT are both excellent at taking messy, conversational notes and turning them into structured, readable drafts. Gemini can cross-reference your draft against similar postings to flag gaps. Grammarly AI can catch tone inconsistencies. Each tool plays a different position in the workflow.

Once you have a draft, the editing phase is where AI adds the most value. Ask it to flag jargon that an outsider wouldn't understand. Ask it to identify requirements that are actually preferences, and separate the two clearly in your posting. Ask it whether your 'About Us' section sounds like a real team or a corporate brochure. These are the questions a skilled editor would ask, and most HR teams don't have one on staff. AI doesn't replace editorial judgment, but it gives you something close to it at no extra cost and in about ten minutes.

Finally, treat your job description as a living document with a feedback loop. After each hiring cycle, ask your new hires what the posting got right and what surprised them about the role. Feed those answers back into your next revision. AI can help you analyze that feedback at scale, paste in a dozen responses and ask for themes. Over time, your postings get more accurate, your candidate pools get better calibrated, and your hiring managers spend less time filtering noise. That's not a technology story. It's a discipline story that technology makes easier.

Click to enlarge

What the Best Job Descriptions Have in Common

1. They treat the candidate as a decision-maker, not an applicant, giving them the information they need to self-select accurately.
2. They replace vague adjectives ('dynamic,' 'passionate') with specific behaviors and measurable expectations.
3. They separate must-haves from nice-to-haves, ideally in labeled sections, so qualified candidates don't disqualify themselves over a preference.
4. They describe the work environment honestly, including the hard parts, so candidates aren't surprised after accepting an offer.
5. They include real numbers wherever possible: team size, budget, quota, average deal size, class size, patient volume, whatever is relevant.
6. They have been reviewed by at least one person currently doing the job, not just the manager hiring for it.
7. They have been stress-tested with an AI tool for clarity, inclusivity, and missing information before being posted publicly.

Takeaways

• A job description is a sales document and a filter simultaneously, it needs to attract the right candidates and discourage the wrong ones.
• AI tools work best as editors and auditors of your draft, not as blank-page generators, bring your own role knowledge first.
• Vague language doesn't just fail to attract candidates; it actively misleads them, which creates dropout and early attrition.
• Gendered and exclusionary language patterns are often invisible to the people who wrote them. AI catches what familiarity hides.
• Specificity (real numbers, real scenarios, real expectations) is the single highest-leverage edit you can make to any posting.
• The feedback loop matters: what surprises new hires after joining is exactly what your next posting needs to include.
• ChatGPT, Claude, Gemini, and Grammarly AI are all free or low-cost tools that can meaningfully improve your postings this week.