Here is the awkward finding HR teams keep running into. SHRM’s 2025 Talent Trends report has AI adoption inside HR climbing from 26% to 43% in a single year, with 64% of those teams pointing it at recruiting. Cost-per-hire and time-to-hire have both gone up over the same three-year window. The tool is faster. The funnel is slower. The reason, in most shops, is that AI is being used to write more job ads, not better ones, and the extra volume is filling pipelines with applicants who shouldn’t be there.
A good job ad does the opposite. It deflects the wrong applicants and pulls in the right ones. That is what this workflow is for.
Start by writing what the role actually does on a Wednesday
Most ads die in the first paragraph because they were written from a job spec, not from the job. Before you open ChatGPT or Microsoft Copilot, spend ten minutes with the hiring manager and answer four questions in plain language: what does this person do in week one, what do they do in month three, what is the messiest part of the role, and what does someone need to have done before to be useful here. Voice-record the conversation if you can. Those answers are your raw material.
Now drop them in. Use a structured prompt rather than “write me a job description”:
You are helping me draft a job ad for a [ROLE TITLE] at [COMPANY], a [SIZE + INDUSTRY] team. Here are my notes from the hiring manager: [PASTE NOTES]. Write a first draft with these sections: a two-sentence hook describing the work itself (no company puffery), a “what you’ll actually do in your first 90 days” list of 5 concrete deliverables, a “what we need you to have done before” section listing 3 to 5 capabilities phrased as past actions (“you have led a payroll migration”, not “experience with payroll”), a short paragraph on how the team works, and a one-line salary band. Keep it under 400 words. No corporate adjectives. No “rockstar” or “ninja”. No “fast-paced environment”.
You will get back something with shape, in your tone, with the hiring manager’s actual context inside it. It will still need editing. The point is that you are now editing a draft that knows what the role is, not a template stuffed with adjectives.
Run two checks before anyone sees it
Job ads have two failure modes that AI is unusually good at catching, and unusually bad at producing on its own.
The first is biased or coded language. Words like “competitive”, “dominant”, “ninja”, “young energy”, and even adjective-heavy openers shrink your applicant pool in ways that show up in the data later, not when you publish. Tools like Textio and Ongig are built to flag this in real time and score the ad. If you don’t have a paid tool, ask a general model to do the same job:
Audit this job ad for language that is likely to discourage qualified applicants from underrepresented groups. Flag every phrase that is gender-coded (masculine or feminine), age-coded, ability-coded, or culturally narrow. For each flag, suggest a neutral rewrite. Then list any “must have” requirement that is more likely a “nice to have” in disguise. Output as a numbered table.
The second failure mode is the requirements list that filters out the people you actually want. Research from LinkedIn and others has shown women apply when they meet most of the listed requirements while men apply when they meet a smaller share. If your “essential” list has 12 items, you have already narrowed your pipeline before publishing. Cut it to 5. The rest go in “useful, not required”.
The counterintuitive bit: write the rejection before you write the ad
The single change that lifts the quality of inbound applicants is usually not in the ad. It is in the screening question. Ask AI to write the screen first, then back into the ad:
Based on the role notes pasted above, generate three short screening questions for the application form that would distinguish a strong applicant from a generic one. Each question should require a specific past example, not a yes/no, and should take an applicant under 3 minutes to answer. Then suggest the wording in the job ad itself that signals to applicants that we read these answers carefully.
A good screening question does what a paragraph of marketing copy can’t. It sets the bar in the applicant’s head before they hit submit. The lazy applicants self-select out. The serious ones take the brief seriously. This is the lever that actually shifts cost-per-hire, because the cost is downstream of pipeline quality, and pipeline quality is downstream of who chose to apply.
The failure mode to watch for
The AI-written ad that reads like every other AI-written ad. Tells include: openers like “Are you a passionate [ROLE]?”, filler about “fast-paced collaborative environments”, parallel-structured bullet lists where every line starts with the same verb, and a “what you’ll bring” section that lists soft skills with no behavioural anchor. Candidates are starting to recognise these. So are the bots that summarise jobs back to them.
A simple sanity pass before you publish:
Read this job ad as a sceptical senior candidate currently employed elsewhere. Flag any sentence that sounds like marketing copy, any line that could appear in any other company’s ad with two words swapped, and any claim about culture or values that is not backed by something concrete. Suggest one sharper alternative for each.
If three out of four sentences come back flagged, the draft is generic. Rewrite from the hiring manager’s words again, not from the AI’s first attempt.
What this gets you
A job ad in roughly 25 minutes that reads like a human wrote it about a real role, with screening questions that do the heaviest filtering, language that doesn’t quietly shrink the pool, and a description specific enough that the wrong applicants don’t bother. That is what good looks like. It is also what shows up in lower cost-per-hire and shorter time-to-hire, eventually. The teams seeing those numbers improve from AI are the ones using it to be more selective at the top of the funnel, not to publish more ads at the bottom.
The technology will keep getting better. The discipline of writing for the role you actually have, with the language that doesn’t repel the people you want, is the part that stays your job.