The blank page is the part of writing a job ad that AI is genuinely good at.
A hiring manager has the role in their head (the team they are building, the gap they are filling, the kind of person who would thrive), and what they need is a draft to react to. Not a finished ad. A draft. Something with shape that they can cross out, sharpen, and make their own. That is a 30-second task for a focused human and a 2-minute one without. AI closes the gap.
At Join we ship AI as an assistant, not a decision-maker. That stance shapes which parts of the hiring funnel we automate, and which we deliberately do not. Job-ad drafting sits cleanly in the assistant column: the model clears the blank page; the human keeps the call. Past that boundary the picture gets more mixed. Here is what we have seen work, what to avoid, and where the model should stop.
Prompts that work
- Translate bullets into prose. Feed the model a list of responsibilities, the team’s tools, the seniority band, and the location. Ask it to write the “what you’ll do” and “what we’re looking for” sections in two paragraphs each, in the company’s own voice (described in one line, e.g. “direct, plain English, no jargon”). The output will not be final, but it solves the cold-start problem. Edit, do not accept.
- Generate the screening questions. Once the ad is drafted, ask the model for five screening questions that would surface whether a candidate has the specific experience the role needs. The questions are almost always better than what a hiring manager writes under time pressure, because the manager is anchored on the job description and the model is anchored on what would actually differentiate applicants.
- Rewrite for a different audience. The same ad written for a senior IC reads differently than the same ad written for someone in their first promotion into the role. Ask the model to rewrite the opener for “a senior engineer considering a smaller team” or “a first-time tech lead looking for ownership”. You will get four or five variations to pick from, and the right one is usually obvious.
Pitfalls to avoid
- Letting the model invent the role. If you prompt “write a job ad for a senior backend engineer”, the model will give you a perfectly plausible ad for a generic senior backend engineer. That ad will not describe your role. It will describe the median role. The output is fluent enough that the hiring manager skims it, assumes it is right, and then the applicants who show up are the median applicants. Always start from the specifics. Always.
- Trusting the salary phrasing. Models will happily write “competitive compensation” or “market-rate salary”. Both phrases are noise, both phrases hurt applications, and both phrases will end up in your ad if you do not strip them out. Either put a real range in the ad or leave the line out entirely. Do not let the model paper over the decision.
- Letting it write the rejection email at the same time. Tempting: get the ad and the templates in one session. The problem is that the model defaults to soft, generic, slightly-off-key rejection copy that reads exactly the way auto-rejection reads. The candidates you turn down are also potential customers, future referrals, and future re-applicants. Write that email once, by hand, and reuse it.
Where the model stops being useful
Anywhere the output commits your company to something. Salary bands. Benefits specifics. Equity policy. Visa-sponsorship language. Diversity statements. Anything where being slightly wrong means an applicant calls you on it later.
There is also a regulatory lens. The EU AI Act classifies AI systems used “for the recruitment or selection of natural persons” as high-risk under Annex III, with the full obligations (risk assessment, bias testing, human oversight, transparency) enforceable from 2 August 2026. Drafting an ad is not regulated. Using AI to filter applications or rank candidates is. The dividing line is whether the output makes a decision about a person, and a job ad sits on the safe side of that line only as long as a human is doing the editing.
There is a trust lens too. Gartner’s 2025 survey found only 26% of job applicants trust AI to evaluate them fairly, and 87% want employers to be transparent about how AI is used in hiring. A 2026 job ad lives next to that suspicion. If it sounds like it was generated and never edited, applicants will assume the rest of the pipeline is generated and never edited too.
A reasonable rule: if a sentence in the ad would survive a screenshot on LinkedIn with your logo above it, the model can draft it. If it would not, the model should not write it.
The handoff
The cleanest workflow we see at SMBs that hire well: the hiring manager writes a five-bullet brief, the model drafts the ad, the manager edits for voice and adds the specifics, a peer reads it for cold-applicant clarity, and it ships. Twenty minutes end to end, instead of the two hours it used to take.
That is the part that matters. AI did not replace a writing skill. It collapsed a blank-page tax that nobody enjoys paying.
Use it for the draft. Keep the final pass.
