How to write job postings with AI that attract the right profile

TLDR: A job posting is marketing aimed at an audience of one: the candidate who fits. AI speeds up the drafting (YesChat to contrast versions, Jasper if the team publishes at volume) and your work concentrates where it matters: defining the real role, removing the clichés that deter and the biases that discriminate unintentionally, and stating the salary. An honest posting filters better than any later screening.

This guide is for anyone publishing postings and receiving what they did not seek: two hundred applications with no fit, or three applications for a role that should attract more. In both cases the problem usually starts in the posting’s text, the piece of the process that gets the least attention and determines the most.

1. Research how your target profile searches

Before writing, look at the market from the candidate’s side: which postings compete for the same profile, what they promise, what vocabulary they use. Kagi performs in that search with its lenses filtering forums and communities, where you also see what real candidates say about your sector’s postings: what attracts them, what scares them off, which promises they no longer believe. From $5 a month with a free trial.

Two typical findings from that half hour: the job title your company uses internally is not what candidates search (and the internally-titled posting never appears in their searches), and the benefits your sector takes for granted do not differentiate, while the missing ones (salary, real flexibility) disqualify you.

2. Define the real role before drafting

AI cannot invent your role, and when it does it produces the generic posting that attracts generic applications. The upfront document that decides quality: what this person does in a normal week (concrete, not “manage strategic projects”), the three genuinely essential requirements (the list of ten is the signature of a badly defined role), what can be learned on the job, and the salary range.

On salary, the practical argument besides the regulatory one (EU pay transparency advances by directive): postings with a range receive better self-filtered applications, and the whole process saves the weeks of discovering at the end that expectations never matched.

3. Generate and refine with your company’s voice

With the definition done, drafting is fast. YesChat lets you generate the posting with GPT and with Claude and compare approaches, from a free plan: useful because each model structures differently and seeing two versions speeds up finding the tone. For teams publishing at volume who need a uniform brand voice, Jasper applies the company profile to every posting, from $59 a month with a trial.

The human refinement has two passes. The cliché pass: out with “dynamic”, “proactive”, “works well under pressure” and the corporate “family”, which inform nobody and deter the good ones. And the bias pass: gendered language, disguised age requirements (“digital native”) and exclusions without need (the degree the role never uses, the availability that is not required). Asking the model itself to review the posting for exclusionary language is a free second opinion, and final responsibility belongs to whoever publishes.

4. Turn the posting into an honest filter

The perfect posting is not the one attracting the most applications: it is the one attracting the right ones and deterring the rest with grace. The honest filter’s pieces: the “what working here is really like” paragraph (the demanding parts included), the description of the first quarter in the role (letting candidates project themselves in or out) and the short, specific question on the form, which costs a genuinely interested person two minutes and discourages the mass applier.

That upfront filter is the best gift to the next phase: the screening candidates with AI guide starts where this one ends, with less volume and more fit. The whole sector lives in AI for human resources.

Common mistakes

Publishing the generated draft without a prior definition. It produces the generic posting that could belong to any company, and attracts exactly that.

Hiding the salary. It multiplies discardable applications and loses the candidates with options, who do not invest in opaque processes. Transparency is also advancing by regulation.

The ten-essential-requirements list. Good candidates self-exclude before impossible lists (especially the profiles that only apply when they meet everything), leaving the ones who apply to everything.

Skipping the bias review. Exclusionary language discriminates unintentionally and exposes you legally. The review pass is short and mandatory.

Frequently asked questions

Can people tell a posting was written with AI?

They can tell a generic posting, human or generated. With a real role definition, concrete facts and your company’s voice, a well-refined generated draft is indistinguishable from a good handwritten one, because it is one.

How much does this flow cost?

From zero on the free tiers: YesChat for drafting, and the research can run on any search engine. The paid set (Kagi plus Jasper for volume) lands around $64, relevant only for teams publishing constantly.

Should I publish the salary range?

Yes: it filters expectations from minute one, attracts candidates with options and pay transparency advances by regulation in the EU. The internal discomfort of publishing it is smaller than the cost of processes dying at the money conversation.

The same posting for every job board?

The content yes, the format no: each board has its fields and customs. Generating variants from the same definition takes minutes and avoids the hand-trimmed posting that loses exactly the filter paragraph.