The job posting that attracts (and doesn't scare off) the right candidate
How to use AI to write and test the right job posting, without the inflated requirement list and the jargon that scare off exactly the people you most wanted to attract.
You write the invitation for the launch event by copying the "who can't miss this" list the director sent over, full of internal acronyms and catchphrases. AI makes the text sound nice in minutes. Except whoever receives the invitation doesn't understand half the acronyms and figures the event isn't for them, and the exact person who most needed to be there doesn't RSVP. The question that matters isn't "how exciting did the text turn out". It's: "who, reading this, will feel it's for them, and who will feel unwelcome".
You ask AI to write the announcement about the new reimbursement policy, copying the legal text finance sent over, full of technical terms and fine-print exceptions. AI makes it read fluently in minutes. Except the team skims it, doesn't understand the real deadline, and keeps sending non-standard expense reports, generating rework for the whole finance department. The announcement was correct and communicated nothing, because it spoke the language of whoever wrote it, not the language of whoever needed to act.
You ask AI to draft the confidentiality clause in the contractor agreement, copying the firm's standard template, full of "heretofore" and legalese. AI hands back a technically flawless text in seconds. Except the contractor, an individual with no lawyer, doesn't understand what they're signing, hesitates, and closing the contract is delayed a week over a doubt simpler language would have avoided. The clause protected the company on paper and scared off exactly the quick signature it needed.
You ask AI to write the ad for the new product by copying the list of technical attributes the engineering team sent over: throughput, latency, benchmark. AI hands back a correct and boring text in minutes. The audience that decides to buy isn't made of engineers, and nobody clicks, because the ad talks about specs and doesn't talk about the problem the product solves in the reader's life. The campaign used the language of whoever built it, not the language of whoever buys it.
You open the mid-level data analyst posting by copying the wish list the manager sent over: seven years of experience, mastery of four programming languages, "young and dynamic profile", "we want a true data ninja". AI drafts a nice-looking text in minutes, with upbeat adjectives and an excited call to action. The problem shows up two weeks later: few applications, and the few that came in are from people who already have a job and are just window-shopping the market. The perfect candidate, who has four of the seven requirements and would be great in the role, never even applied, because the list looked like it was asking for a unicorn, and "young and dynamic" read like an age signal she didn't want to test. The question that separates a posting that attracts from one that scares off isn't "how much to ask for". It's: "what on this list is truly minimum to do the job, and what's jargon and bias dressed up as a requirement".
You ask AI to write the description of the new feature in the changelog, copying the project's internal technical name and the list of flags the team changed. AI hands back a correct text in minutes. The user reads it and doesn't understand what they actually gained, so the feature that cost a quarter's worth of work slides right by, generating no "wow" and no adoption. The changelog documented the code and didn't communicate the benefit.
You ask AI to write the prospecting email by copying internal sales talk, full of product acronyms and metrics that only make sense to someone who's already a customer. AI hands back a fluent text in seconds. The cold lead reads it, doesn't recognize a single word describing their own problem, and archives it without replying. The email sold the product to whoever already understands the product, and scared off exactly whoever didn't yet know they had that problem.
You ask AI to write the new procedure for the field team, copying the technical text that came from headquarters, full of process terms that only make sense in the office. AI hands back a correct, formatted document. The field team skims it, doesn't change their behavior, and the procedure becomes a dead letter in a drawer. The instruction was correct on paper and never reached whoever executes it, because it spoke the wrong language to the wrong audience.
You ask AI to write the annual LGPD training by copying the legal text straight from the law's article, full of cross-references and single paragraphs. AI hands back a technically correct piece of material in minutes. The team goes through the training on autopilot, retains nothing, and the next personal-data incident happens in exactly the area that "passed" the training. The content was correct and taught nothing, because it spoke the lawyer's language, not the language of whoever handles the data every day.
You ask AI to write the documentation for the new API for partner teams, copying the technical text the engineering team itself uses internally, full of acronyms and references to architecture decisions that only make sense to someone who already knows the system. AI hands back a correct, dense text. The partner team reads it, doesn't understand how to integrate, and opens a support ticket for every step, because the documentation spoke the language of whoever built the API, not the language of whoever just needs to use it.
You ask AI to write the error text that shows up when a payment fails, copying the technical error code the backend returns, something like "error 402: payment_required_exception". AI hands back a technically correct text. The user reads it, doesn't understand what to do, and abandons the cart at the final step of the purchase. The message was correct from the system's point of view and scared off someone who was already about to buy.
You ask AI to write the repositioning memo for the company, copying the consulting jargon left over from the last workshop: "synergy", "leveraging intangible assets". AI hands back a fluent text full of nice-sounding terms. The team that needs to execute the change reads it and doesn't understand what actually changes, so nobody acts differently on Monday morning. The strategy was flawless on the slide and never reached the shop floor, because the language spoke to the board, not to whoever executes it.
Let me tell you something that hurts to watch happen every week. Someone opens a "mid-level analyst" posting, pastes the wish list the manager sent over their shoulder, asks AI to make the text sound nice, publishes it, and two weeks later complains that "there aren't any good people out there". Except there were. She just never got around to applying, because the posting, without meaning to, sent a "this isn't for you" signal to exactly the person who was most worth attracting.
The core idea of this lesson. AI massively speeds up the part of drafting and testing language variations for the posting, it does that in minutes. But two decisions remain yours, and they're what decide whether the posting attracts or repels: what the real MINIMUM requirement is to do the job (not the manager's entire wish list), and whether the language carries hidden jargon or ageist terms that exclude for no good reason.
01The two classic mistakes that scare off the right candidate
Let's call it what it is. Two mistakes repeat in almost every poorly written posting, and both have the same effect: they disproportionately scare off exactly the people worth attracting.
The first is the inflated requirement. An HP study, cited by Sheryl Sandberg in the book Lean In, showed something any experienced recruiter has already felt firsthand: women tend to only apply when they meet close to 100% of the listed requirements, while men apply at around 60%. This isn't about competence, it's about how each group reads a list of requirements. When you paste seven requirements into the posting and only three or four are actually indispensable, you're not filtering better, you're effectively scaring off whoever holds themselves to the highest standard before applying.
The second is jargon and exclusionary terms. "Rockstar", "ninja", "warrior" sound fun to whoever writes them and sound like an exhausting work culture to a lot of good people who've already lived through that. "Young and dynamic profile", "digital native" sound harmless and are, in practice, an age signal, the kind of phrase that scares off a forty-five-year-old professional with fifteen years of experience who would be exactly who you need.
The same posting, two ways of writing it, two very different funnels:
The good news is both mistakes are easy to fix, once you stop to look at them. This is exactly where AI comes in.
02AI as a generator of variations and a tester of language
This is where AI genuinely speeds things up. Instead of writing one posting and publishing it on a hunch, you ask it to generate two or three tone variations for the same role: one more institutional, one more direct, one closer to how the team actually talks. You compare and choose whichever best represents the real opening, not whichever sounds the most "salesy".
Also ask it to act as a critical reader of its own language: "review this text and flag any term that might sound ageist, gendered, or like exclusionary culture, and suggest a neutral alternative". This catches "young and dynamic" before it goes out, and swaps "ninja" for "someone with experience in", without losing the text's energy.
And adapt it to the channel. The text that works on LinkedIn, more institutional, isn't the same as what works in an internal referral, more direct, nor the same as what goes into a recruiting platform like Gupy, which calls for objectivity. AI puts together the three versions fast; it only does this well if you already know what you want to test.
03What's still yours: the real minimum requirement, not a wish list
Here's the part AI doesn't solve on its own, because it doesn't know your operation. The list the manager sends is almost always a wish list, the resume of the dream candidate, not the minimum requirement to do the job well day to day. Separating the two is your judgment call.
The right question for each item on the list is simple: "without this, can the person really not do the job, or is this just a welcome bonus?". Seven years of experience can be a bonus when three solid years already deliver the essentials. Mastery of four languages can be a bonus when the real day-to-day only uses two. Cutting the list down to the real minimum isn't lowering the bar, it's stopping the scaring off of people who hold themselves to too high a standard to apply to a list even the current team doesn't fully meet.
04The quick audit before publishing
Before the posting goes out, run a short checklist, it takes less than five minutes:
- Is there hidden ageist language? "Young", "digital native", "recent graduate" when it doesn't need to be, all of that is an age signal disguised as energy.
- Is there exclusionary culture jargon? "Rockstar", "ninja", "working under constant pressure" with no context sound less like a job and more like a burnout warning.
- Is the requirement minimum or is it a wish list? If you crossed off half the list, could the team still hire someone who delivers? If so, the list is inflated.
- Does the tone speak to who I want to attract? Read it out loud imagining the right person reading it. Would they feel the posting was written for them, or for a character that doesn't exist in the market?
05The final picture
Let's close it simply. AI writes fast and tests variations of tone, gender-neutrality, and channel in an instant, that's pure time saved. But deciding what's truly minimum to do the job, and auditing whether the language isn't already sending the right candidate away before they even read the second line, that's still yours. The posting that attracts isn't the prettiest one. It's the one that speaks the language of whoever you want on the other side, and only asks for what actually matters.
Do it now
Take your real task, a real opening you have posted or the next one you're going to open.
- List the requirements exactly as the manager sent them, without editing anything yet.
- For each item, ask: "without this can the person not do the job, or is it a bonus?". Cross out anything that's a bonus disguised as a requirement.
- Ask AI to rewrite the posting in two different tones (one more institutional, one more direct), already with the cut-down list.
- Run the language checklist on both versions: ageist term, exclusionary jargon, minimum requirement, right tone for the right audience.
- Choose the version that passed the checklist and publish it.
Compare, if you can, with the performance of the last similar posting you opened the old way. The difference in the funnel is the argument that convinces the next manager to cut the wish list before publishing.
Practice
1. Why does an inflated requirement in a posting disproportionately scare off women candidates, according to the study cited in this lesson?
2. In the choreography of writing the right job posting, which task does AI speed up well, and which stays yours?
3. A posting includes the phrase 'we're looking for a true data ninja, young and dynamic profile'. What does that phrase do, in practice?
</content>
For the board
On the candidate who never camethe good people were there. They just did not apply, because the posting sent a message that this is not for you.
On the requirementa wish list dressed as a requirement does not filter better. It decides who even applies.
On the languageninja, young and dynamic exclude people without anyone deciding to exclude anyone.
Thanks for the feedback. It helps sharpen the next lesson.