Home

Writing Job Posts With AI: What to Check and How to Set a Pay Range

Sources checked: November 9, 2026. AI-assisted draft, pending editor review.

It's Sunday night, your shift supervisor just gave two weeks' notice, and the job post has to go up before the holiday rush. Pasting your notes into an AI chat and asking for a job description takes about a minute, and the draft will read well. Whether it describes your job, asks only for what the work needs and states pay you can stand behind is a separate question, and you're the one who has to check it.

Say the owner is Dana. She runs a café in Hoboken, New Jersey, with 12 employees, and she's had at least 10 people on payroll all year. Dana is made up, and so is the draft she gets back. The laws, cases and wage figures below are real, with sources at the end.

Dana types what she knows about the job, in her own words:

The draft comes back polished, with a summary, duties and requirements. Four of its lines never appeared in her notes. This draft was written for this article to show the pattern. It isn't output from a named product.

Every one of those lines sounds like a job post, which is why they're easy to miss. A language model writes what job posts usually say. It has no way to know what this job is.

None of the evidence below measured AI-written job posts. It covers resume screening and application software, which are the closest examples found for this article. Read the table for the pattern, then for what a small employer can do about it.

What the record establishes is narrower than "AI is biased," and still useful. Language models pick up patterns tied to sex, race and age from the text they learned from, and the EEOC's suit over software that screened people out by age settled for $365,000. Dana's "recent graduate" line is the small-business version of that risk. A job post is also the one place where she can catch it by reading, because nothing goes out until she posts it.

AI can organize Dana's notes. It can't verify anything about her café, so everything in the right column has to come from her.

Dana's corrected process has five steps, and only one of them belongs to the tool.

Dana starts with public data. The pay explorer on this site shows U.S. Bureau of Labor Statistics (BLS) wage estimates by state and metro area. The closest BLS title to her job is "First-Line Supervisors of Food Preparation and Serving Workers," and these are the New Jersey figures.

The table is a starting point and no more. One BLS title groups many different jobs, from a small café to a hotel kitchen, and statewide figures blend very different local markets. The section at the bottom of the pay explorer, "But is this information enough?", explains what else employers consult.

Next Dana looks at what she pays today. In this example her two current shift supervisors earn $22.00 and $23.50 an hour, and her budget tops out at $25. She posts $22 to $25 an hour. Both ends are numbers she would really pay, and the range makes sense next to the people already doing the job.

At a large employer, I helped build the pay ranges we used in job postings. We set them from four inputs: outside salary surveys, what current employees in the same role were paid, the pay band for the level, and differences by work location. The range usually ran about 10% to 15% above and below the midpoint. A small business won't have most of that machinery, and it doesn't need it. My advice for a posting in New York or New Jersey is simpler: put only amounts you'd actually pay at both ends of the range.

Dana compares the range with current pay before posting because a posted range is public. Economists Cullen and Pakzad-Hurson found in Econometrica that pay transparency reduces individual bargaining power and tends to lower average wages. A Kenan Institute review notes a paper finding that a transparency law reduced the gender pay gap by 1.2% to 2%, which the review says was driven mainly by lower male wages. Ranges change behavior on both sides, and Dana's current supervisors will read the posting too.

Two things stay out of the AI tool at this step. Don't paste a current employee's salary or review into it to "match" a range. And don't let the tool decide the range: it has no access to your pay data or budget and may produce a plausible number with no basis.

Some states require a pay range in job postings. These two examples come from official pages.

On the facts of this example, with 12 employees all year and a job in Hoboken, Dana appears to be covered by New Jersey's rule. Her posting would need the hourly range and a general description of benefits and other compensation, which is why "competitive pay and great perks" had to go.

New York's guidance, which applies to New York postings, adds a standard for the range itself. It must be the salary or range the employer believes, in good faith, to be true at the time of posting, and it should give the applicant a legitimate idea of the expected pay.

After the checks, the post is shorter than the AI draft, and every line is true. This is still the hypothetical example.

Her assistant manager reads it before it goes up. Dana saves the final version and the date she posted it, since postings can be examined later.

Want the next guide when it's published? Join the list.