Best AI Prompts for Job Descriptions (2026)

Prompts for job descriptions#

Act as a technical recruiter. Write a job description for a [JOB TITLE] at [COMPANY, ONE-LINE DESCRIPTION]. Use only these facts: [PASTE YOUR DUTIES, MUST-HAVE REQUIREMENTS, NICE-TO-HAVES, LOCATION OR REMOTE POLICY, AND PAY RANGE AND BENEFITS]. Do not invent requirements, salary figures, equity or perks. Use sections for summary, responsibilities, required qualifications, preferred qualifications, pay and benefits, and how to apply, and mark anything I did not supply as [MISSING].

Act as an engineering director. Rewrite this draft job description for a [SENIOR FRONTEND OR DESIGN-SYSTEMS ROLE] so that it lists only requirements that are needed on day one: [PASTE YOUR DRAFT]. Split them into must-have and learnable-on-the-job, replace years-of-experience counts with the skill you actually need, and flag any line that is a wish rather than a requirement.

Act as an inclusive-language editor. Review this job post for wording that could discourage applicants or signal a preference for age, sex, race, religion, national origin or disability, such as "digital native", "recent graduate", "young and energetic" or "rockstar": [PASTE YOUR JOB POST]. List each phrase, say why it is a risk, and give a neutral replacement that keeps the meaning. Do not change the duties or requirements.

Act as a hiring manager building a structured interview. For the [JOB TITLE] role with these must-have skills: [PASTE YOUR MUST-HAVES], write six interview questions and a scoring rubric with a 1 to 4 scale and a one-line description of what each score looks like. Every candidate gets the same questions in the same order. Tie each question to a duty in the job description, and do not ask about age, family, health, religion or origin.

Act as an HR compliance reviewer. Check this job post against a plain list: pay range present, essential duties separated from physical or environmental demands, no age or sex preference, no unnecessary degree requirement, accommodation contact included: [PASTE YOUR JOB POST AND THE CITY, STATE OR COUNTRY WHERE IT WILL BE POSTED]. Report each item as pass, fail or "cannot tell", and say which items need a lawyer or HR to confirm for that location.

What to add to any job description prompt#

• The real facts: duties, must-have requirements, location or remote policy, and the pay range you would actually pay. Both Google and Anthropic say to supply the information the model needs instead of assuming it has it, and a model cannot know your pay band.
• The role of each requirement. Mark what is needed on day one and what can be learned, so the model does not turn wishes into hard requirements.
• A short list of what not to invent: salary, equity, benefits, certifications. Ask it to write [MISSING] where a fact is absent.
• The place the job will be posted. Pay-posting and automated-tool rules differ by state and city.
• A fixed output format. Google says you can request a format, and Anthropic recommends putting instructions, context and your draft in separate labelled blocks so they are not confused.
• An example of a past post you liked, if you have one. Google recommends a few varied examples to control phrasing and scope.

A worked example#

Act as a technical recruiter. Write a job description for a Senior Backend Engineer at Northline Freight, a 60-person logistics software company. Use only these facts: Duties: own the shipment-tracking API, on-call one week in six, mentor two mid-level engineers. Must have: 5 years building production services in Go or a similar language; experience with PostgreSQL. Preferred: experience with event queues. Location: hybrid, 3 days a week in Denver, Colorado. Pay: $150,000 to $175,000 base, plus health, dental and 401(k) match. Do not invent requirements, salary figures, equity or perks. Use sections for summary, responsibilities, required qualifications, preferred qualifications, pay and benefits, and how to apply, and mark anything I did not supply as [MISSING].

A good result for this task:

• Uses exactly the pay range and benefits you gave, with no added equity or bonus.
• Keeps "5 years" and the PostgreSQL requirement as written, and lists event queues only as preferred.
• Includes the on-call duty and hybrid location, so candidates are not surprised later.
• Marks application deadline, equal-opportunity statement and accommodation contact as [MISSING] instead of making them up.
• Contains no age, sex or culture-fit phrases such as "young", "digital native" or "work hard, play hard".

Check duties against reality. The EEOC's ADA guidance defines an essential function as "a fundamental duty of the position", and a person who cannot perform one, with or without accommodation, is not a qualified individual. Separate those duties from nice-to-haves, and do not let the model promote a wish into an essential function.

Check the pay line. New York State, for one, requires a job description and a compensation range, plus whether the job is commission-based, for employers with four or more employees. The range must be what the employer believes accurate at posting. Other states and cities have their own rules, so ask HR or counsel which apply to your posting location.

For interview kits, ask every candidate the same questions. The EEOC says employers may ask whether an applicant can perform job functions with or without reasonable accommodation, but before an offer may not ask "Do you need reasonable accommodation to perform this job?" Remove any model-written question about health, age, family or origin.

Keep AI out of the screening decision. New York City's DCWP says employers and agencies cannot use an automated employment decision tool unless it had a bias audit within one year of use, and that candidates must be given notice. Using a model to draft posts and questions is not the same as using it to rank or reject candidates, but if you do the latter, a person should review each decision and your counsel should check local rules first.

The full prompt library is on our prompt library page.

Limits of this guide#

This guide covers drafting posts and interview kits. It is not legal advice, and we are not lawyers. We did not test which prompt wording gives better posts, and we did not confirm the current federal status of EEOC artificial-intelligence guidance: the page we tried to read returned not found, and secondary sources say the May 2023 technical assistance was withdrawn, so we do not rely on it. We read New York State and New York City pages only. Rules in other places, including other pay-posting states, must be confirmed locally. We could not read OpenAI's prompt engineering guide in this pass.