Recruiting strategyRead at source, 23 September 2026
ChatGPT for Recruiting: Prompts That Work, and What It Can't Check
ChatGPT is useful for recruiting when you use it to draft, and less useful when you ask it to decide. It can turn your notes into a job description, cut that down into a social job ad, write screening questions and draft follow-up texts in seconds. It can't tell you your local pay rate. It can't confirm that a posting meets your state's pay transparency law; California and Washington require a pay range from employers with 15 or more employees, and New York State from employers with four or more. It can't check its own output for discriminatory wording either. OpenAI's own guidance is to use ChatGPT "as a first draft, not a final source." Give it the facts, and have a person check what comes back.
What ChatGPT for recruiting is good at, and where it stops
ChatGPT is a language model. It's very good at changing the form of text you give it, and unreliable when it has to supply facts you didn't give it. OpenAI's help center says it "can produce incorrect or misleading outputs" and "may express high confidence even in incorrect answers." For recruiting, that splits the work fairly cleanly:
| Task | Good use of ChatGPT? | What you still have to supply or check |
|---|---|---|
| Turning notes into a job description | Yes | Pay, shift, location and the requirements that actually gate the job |
| Shortening a description into a social job ad | Yes | The first line: pay, shift and location |
| Rewriting jargon into plain language | Yes | That the meaning survived the rewrite |
| Drafting knockout screening questions | Yes, with limits | That each question tests a real, job-related requirement |
| Candidate texts and emails | Yes | Names, dates and times, and your company's tone |
| Setting a pay range | No | Your own local wage data. The model will make up a plausible number |
| Confirming legal compliance | No | The agency's own guidance, or your employment counsel |
| Ranking or rejecting candidates | No | Human judgment and, in some places, a bias audit (see below) |
ChatGPT job description prompts that work
A vague prompt gets you a generic description. The fix is to give ChatGPT the facts only you know, and tell it not to invent anything else. Here's a prompt you can paste in and fill out:
Prompt: job description from your facts
"Write a job description for a [job title] at [company], a [type of business] in [city, state]. Pay: [$X–$Y per hour]. Schedule: [days and shift times]. Required, because the job legally or physically can't be done without it: [license, certification, lifting requirement]. Preferred but not required: [list]. What a normal shift involves: [three or four sentences in your own words]. Benefits: [list]. Use plain language a candidate would read on a phone. Put pay, shift and location in the first two sentences. Do not add any requirement, certification, years of experience or pay figure that I have not listed. If something is missing, write [MISSING] instead of guessing."
The last two sentences do the most work. Without them, ChatGPT tends to fill gaps from its training data: a bachelor's degree for a warehouse job, "3–5 years' experience" for an entry-level role, a certification the job doesn't require. Each invented requirement shrinks your applicant pool before anyone has read the ad. If you'd rather not write prompts at all, our free job description generator builds frontline descriptions from the same fields and leaves brackets instead of guessing.
Follow-up prompts worth keeping
- "Rewrite this at an eighth-grade reading level without removing any facts."
- "List every requirement in this description and mark each one as legally required, physically required or preference."
- "Flag any phrase that could suggest a preference for age, sex, race, religion, national origin or disability." This is a first pass for a person to review, not a compliance check.
- "Give me five alternative job titles a [role] candidate would actually search for."
From job description to social job ad
A job description and a job ad have different jobs. The description is the full record. The ad is three to five lines that have to stop someone scrolling who wasn't looking for work. ChatGPT is good at the conversion if you tell it where the ad will run:
Prompt: social job ad
"Using the job description below, write three versions of a Facebook job ad for people who are not currently job hunting. Each version: a first line under 12 words with the pay, the shift and the town; two short lines on what the job is actually like; one line saying they can apply in under a minute from their phone. No emojis, no exclamation marks, no 'rockstar' or 'ninja' language, no claims about company culture I have not given you."
Then check the draft against what actually drives cost. In Boostpoint's 2026 benchmark, what happens after the click explains 70% of the difference in cost per applicant between campaigns. What the ad auction charges explains 18%. A better-written ad helps, but a shorter application helps more. For the copy itself, our job ad copywriting guide has rewrite rules and templates for each role.
Prompts for screening questions, texts and interviews
- Knockout questions: "Write no more than three yes/no screening questions for this role. Each must test a requirement listed as legally or physically required. Do not ask about preferences, availability beyond the stated shift, or anything related to a protected characteristic."
- First-contact text: "Write a text under 300 characters to someone who just applied for [role], confirming we received it and offering two interview times: [time 1] and [time 2]. Sign it from [name] at [company]."
- Structured interview: "Write four behavioral interview questions for a [role] and, for each, what a strong, acceptable and weak answer sounds like, so every candidate is scored on the same scale."
- No-show follow-up: "Write a friendly, short text to a candidate who missed an interview, offering one new time and making it easy to say no."
Don't paste candidate names, phone numbers, résumés or other personal details into the consumer version of ChatGPT. OpenAI says it may use consumer conversations to improve its models unless you turn off "Improve the model for everyone" in Data controls. By default, it doesn't train on inputs from ChatGPT Business, ChatGPT Enterprise or the API. If your team uses ChatGPT for recruiting every day, decide which version it uses before someone pastes in a stack of applications.
ChatGPT interview questions: building a structured set in four prompts
Asking ChatGPT for “interview questions for a warehouse associate” gets you the same generic list everyone else gets: tell me about yourself, where do you see yourself in five years. Those questions are not tied to your job, so the answers cannot be compared. The single prompt above gives you four behavioral questions; this is the longer route to a full question set you can use for every candidate for a role. Run the prompts in order, in one conversation, starting from your own job description.
1. Criteria from the job, not from the internet
“Here is the job description for a [role]: [paste]. List the four or five things a person must be able to do well in the first 90 days to succeed in this job. Tie each one to a specific duty in the description. Do not add anything the description does not support.”
Edit the list before going further. If a criterion is not something you would reject someone over, delete it. This list becomes the rows of your interview scorecard.
2. Two questions per criterion
“For each criterion, write one behavioral question (‘Tell me about a time…’) and one situational question (‘What would you do if…’) set in a real moment from this job, such as [a shift handover / a late load / a resident who refuses care]. Plain language, one sentence each. Allow for candidates with no previous job in this field: the behavioral question should accept an example from school, family or any other work.”
3. Follow-up probes
“For each question, give two neutral follow-up probes an interviewer can use when the first answer is vague, such as asking what the candidate did personally. Do not suggest the answer in the probe.”
4. A check, not a sign-off
“Review the full question list. Flag any question that could lead a candidate to reveal their age, disability or health, pregnancy or family plans, religion, national origin or citizenship, or arrest record, and suggest a rewrite that asks about the job requirement instead.”
The last prompt is a first pass, not a legal review; check the result against our illegal interview questions page, which quotes the federal rules, and against the rules where you hire. Then do the three things ChatGPT cannot. Put the questions into your own words and cut them to what fits in the time you have, usually six to eight for an hourly role. Ask every candidate the same questions in the same order, as set out in how to interview someone. And keep the scoring human: using ChatGPT to write questions is drafting, while pasting candidates’ answers in to score them is the kind of automated decision covered in the screening section below and on our AI screening page. If you would rather start from a tested list, the role pages such as CNA, truck driver and warehouse associate interview questions already have the questions and what a good answer sounds like.
What ChatGPT can't check: bias, compliance and pay transparency
This is where AI-drafted postings most often go wrong, and the model can't catch it for you, because the rules depend on where you're hiring and who you are.
Discriminatory wording
The EEOC says it is illegal to publish a job ad "that shows a preference for or discourages someone from applying" because of race, color, religion, sex, national origin, age (40 or older), disability or genetic information. Its own example is an ad seeking "recent college graduates," which may discourage people over 40. ChatGPT reproduces phrases like "digital native" or "young, energetic team" because they're common in the postings it learned from. You can ask it to flag them, but the decision to publish is yours, and so is the liability.
Pay transparency laws
A growing number of states require pay information in job postings, and the rules differ on who's covered and what has to be included. Four examples from the agencies' own pages:
| State | Who is covered | What the posting must include |
|---|---|---|
| California | Employers with 15 or more employees | The pay scale: a good-faith estimate of the salary or hourly range the employer reasonably expects to pay |
| Washington | Employers with 15 or more employees | Wage scale or salary range, plus a general description of all benefits and other compensation |
| New York State | Employers with four or more employees | The compensation range (minimum and maximum annual salary or hourly pay), and whether the job is commission-based |
| Colorado | Employers covered by the Equal Pay for Equal Work Act | Compensation, benefits and application deadlines |
Other states and several cities have their own versions, and they keep changing. ChatGPT's knowledge has a cutoff date and it doesn't know where your job is filled, so it can't tell you which rules apply. It will also happily write "competitive pay" where a range is legally required. The practical rule: never let the model choose the pay figure, and check every posting against the rules where the job will be filled.
Screening with AI is a different legal question
Using ChatGPT to write an ad is drafting. Using it to score or filter applicants can be regulated. New York City's Local Law 144 requires employers using an automated employment decision tool to have had a bias audit within one year of use, publish information about the audit, and give candidates notice. The city has enforced it since 5 July 2023. If you're thinking about pasting résumés into a chatbot to rank them, stop and check the rules where you hire first.
Where the draft ends and hiring starts
A well-written posting that nobody sees doesn't hire anyone. For frontline roles, most of the people you want are working somewhere else and aren't scrolling job boards. They see a job ad in their Facebook or Instagram feed. What decides whether they apply is mostly what happens after they tap: in Boostpoint's benchmark, 31% of caregivers who clicked finished the form, against 8% of CDL drivers. Much of that difference comes from how long and demanding the form is. That's how Boostpoint Attract is set up: your job description goes in, we write the ad around pay, shift and location, send you a proof to approve, and keep the application inside Facebook or Instagram with two or three knockout questions. ChatGPT can give you the words. You still need the pay, a short application and a quick reply.
Related: AI in Recruiting.
Frequently asked questions
How can I use ChatGPT for recruiting?
Use it to draft: job descriptions from your notes, short social job ads, knockout screening questions, candidate texts and structured interview questions. Give it the pay, shift, location and real requirements, and tell it not to invent anything. Don't use it to set pay, confirm legal compliance or rank candidates, and have a person review everything before it's published.
Can ChatGPT write a job description?
Yes, and it writes a good first draft quickly if you supply the facts: title, pay range, schedule, location, required versus preferred qualifications and what a shift involves. Tell it not to add requirements or pay figures you didn't give it, because it will otherwise fill gaps with plausible but wrong details. OpenAI's own guidance is to use it as a first draft, not a final source.
What is a good ChatGPT prompt for a job description?
State the job title, company, city and state, pay range, schedule, legally or physically required qualifications, preferred qualifications, what a normal shift involves and benefits. Ask for plain language with pay, shift and location in the first two sentences. End with: do not add any requirement, certification, years of experience or pay figure I have not listed, and write [MISSING] instead of guessing.
Is it legal to use ChatGPT to write job postings?
Drafting a posting with AI is generally no different from drafting it yourself: you're responsible for the result. The posting must not show a preference based on protected characteristics under EEOC rules, and in states such as California, Washington, New York and Colorado it must include pay information. Using AI to screen or rank applicants is a separate question. New York City, for example, requires a bias audit and candidate notice for automated employment decision tools.
Does ChatGPT know pay transparency laws?
Not reliably. Its knowledge has a cutoff date, the laws differ by state and city and keep changing, and it doesn't know where your job will be filled. California and Washington require a pay range from employers with 15 or more employees, and New York State from employers with four or more. Check the relevant agency's own guidance rather than the model's answer.
Is it safe to paste candidate information into ChatGPT?
Avoid it on consumer plans. OpenAI says it may use consumer conversations to improve its models unless you turn off Improve the model for everyone in Data controls, and that by default it does not train on inputs from ChatGPT Business, ChatGPT Enterprise or the API. Decide which version your team uses before anyone pastes in résumés or contact details.
Can ChatGPT find candidates for me?
No. It can write the ad, but it can't put the ad in front of people. For frontline roles, most good candidates are employed and not browsing job boards, so the ad has to reach them where they already spend time, such as Facebook and Instagram, with an application short enough to finish on a phone.
Can ChatGPT write interview questions?
Yes, and it does it well if you give it your job description and ask for questions tied to specific duties, with one behavioral and one situational question per requirement. Asked cold, it returns generic questions whose answers you cannot compare. Edit the output into your own words, ask every candidate the same questions in the same order, score them yourself on a written scale, and check the list against the federal rules on illegal interview questions before you use it.
The draft is the easy part
Boostpoint writes, checks and runs Facebook and Instagram job ads for frontline employers, with pay, shift and location up front and short in-app applications.
Book a Demo