Most small teams lose more hours to hiring than to any other job nobody trained them for. You write the ad in a hurry, get two hundred applications you cannot possibly read, and end up picking whoever looked least risky on paper. Knowing how to use AI for hiring fixes the boring half of that: the writing, the sorting, the sourcing, the note taking. It will not choose the person for you, and you would not want it to. Here are the seven steps, with copy-paste prompts you can run in whatever tool you already pay for.
What AI Can (and Can’t) Do When You’re Hiring
Before the steps, one honest line about where the tools stop. AI is very good at reading and producing text, and hiring is buried in text. It is not good at knowing what a good hire looks like inside your particular team. It also carries whatever bias sat in the data it learned from.
| What it does well | What it cannot do |
|---|---|
| Turn messy notes into a clear job ad | Know what “good” looks like on your team |
| Rank 200 applications against your must-haves | Judge whether someone will still be there in two years |
| Surface candidates who never applied | Persuade a happily employed person to move |
| Build interview questions and scoring guides | Carry the legal responsibility for a rejection |
Read that right-hand column as a job description for you. Everything on the left is admin you can hand over today.
How to Use AI for Hiring, Step by Step
Seven steps take you from an empty job ad to a signed offer. Each card has the prompt I would actually run, so you can copy it, drop your own details into the brackets and go.
Write a job ad that filters for you
Most ads are a list of demands with no picture of the work. Give the model your rough notes and ask it to lead with what the person will do in the first 90 days. You will get fewer applications and better ones, because people who are not a fit can see it for themselves.
You are an experienced hiring manager. Turn these rough notes into a job ad of about 250 words: [paste your notes]. The role is [role] at a [size] company in [industry]. Lead with what this person will actually do in their first 90 days. List five must-have requirements and mark everything else as optional. Cut the jargon and cut anything that would put off a strong candidate who is not actively job hunting.Post it where the matching is done for you
A job board shows your ad to whoever happens to look that week. A matching platform works the other way round: it holds candidate profiles, reads your role, and brings people to you. This is the one part of hiring where a specialist tool genuinely beats a general model, because the model has no candidates to match you with.
Screen applications without reading all of them
Screening is where good people get lost. Do not ask which candidate is best. Ask for a score against your must-haves with a line of evidence for each one, so you can check the reasoning instead of trusting it.
Here are [number] applications for [role]: [paste or attach]. Score each one from 1 to 5 against these must-haves: [list them]. Give one line of evidence from the application for every score you assign. Flag any candidate you scored low only because of a career gap, a non-traditional background or a foreign qualification, and say what they would need to show in an interview. Return a table sorted by score.Go and find the people who never applied
The person you want is usually employed, not browsing job ads. Use the model as a research assistant that builds your search strings, then run them yourself on the platforms and communities where that work actually happens.
Build me a sourcing plan for [role] in [location or remote]. Give me 10 search strings I can run on LinkedIn, GitHub and industry communities to find people already doing this work at [type of company]. For each string, tell me who it will surface and, just as important, who it will miss.Write the interview before you meet anyone
Unstructured interviews mostly measure how comfortable someone is talking to strangers. Have the questions and the scoring written down first, so two interviewers rate the same answer the same way.
Design a structured interview for [role], 45 minutes long. Give me six questions that test these must-haves: [list them]. For each question, describe what a weak, an average and a strong answer sounds like, so different interviewers score it consistently. Add one work sample task that takes a candidate under an hour to complete.Let it take the notes, not the decision
Writing while listening is why you remember the first and last candidate and nobody in between. Record with consent, then turn your notes into a scorecard tied to evidence. The second prompt is the one that saves a hiring panel from talking in circles.
Here are my notes from an interview with [candidate] for [role]: [paste them]. Turn them into a scorecard against these criteria: [list them]. Quote what the candidate actually said as evidence for each score, and mark anything I wrote down as an impression rather than evidence.Compare these interview scorecards: [paste three to five]. Show me where the panel disagrees most and what extra evidence would settle each disagreement. Do not pick a winner.Check the decision against your own bias
This is the step everyone skips, and it takes two minutes. Ask the model to argue with your shortlist. If the only thing separating your top two is that one of them reminds you of your best hire from four years ago, better to know that now.
Review my shortlist for [role]. Here are the scorecards: [paste them]. List the ways this ranking could be biased: similarity to the current team, prestige of previous employers, communication style, gaps in employment, where someone studied. For each one, tell me what evidence in the file supports or contradicts my ranking.Common Mistakes to Avoid
The teams that get least out of these tools all make the same handful of mistakes. They let the software reject people automatically instead of ranking them. That is how you lose the career-changer who would have been your best hire.
CVs, salary details and personal data get pasted into whatever free tool a colleague recommended, with nobody checking where that data ends up. Another common one is asking for the best candidate instead of asking for evidence. You get a confident answer with nothing behind it.
Outreach written by AI has the same problem. Every message lands in the same polite, weightless voice that strong candidates learned to ignore months ago. And interviews run with no written scoring guide leave the notes with nothing to be scored against.
AI Hiring Bias, Lawsuits and the Laws You Cannot Ignore
This is the part the vendors skip. Hiring is one of the few uses of AI that is already regulated. The downside lands on a person who never agreed to be assessed by software.
Amazon scrapped an internal screening tool after it taught itself to downgrade CVs that mentioned women’s clubs and colleges. That was a company with more machine learning talent than almost anyone.
Your obligations depend on where your candidates live, not where your office is. New York City requires an annual independent bias audit and advance notice for automated employment decision tools. Illinois requires consent before AI analyses a video interview. The EU AI Act puts hiring in its high risk category, with duties for the employer as well as the vendor. Colorado passed a broader AI act covering employment decisions. Its start date has moved more than once, so check where it stands before you rely on it.
The practical rule is simple. Use the tools to shortlist and to gather evidence. Keep a human on every rejection, and write down why you hired the person you hired.
AI Hiring Tools Worth Knowing
Honestly, the general models overlap far more than their marketing suggests. ChatGPT, Claude and Gemini will all write the ad, score the applications and turn your interview notes into a usable scorecard. The free tiers are enough to run a small hire from start to finish.
Where a specialist tool earns its money is matching, because a general model has no candidates to offer you. That is a different kind of product: a platform that already holds profiles and brings the right ones to your role.
Where Jack & Jill AI fits
Jack & Jill AI is built around exactly that gap, and it runs in both directions. Jack is the side candidates use to find work. Jill is the side you use as an employer: you describe the role, and the platform handles the matching and the first round of screening. What reaches you is a shortlist rather than an inbox.
It is the one tool in this guide that does something a chatbot cannot. That is why it turns up in our Jack and Jill AI review and on the candidate side of this same workflow. Trying it costs nothing, so it is a cheap thing to test against your current process.
Want More Than How to Use AI for Hiring?
Hiring is one half of a story we cover from both ends. Our guide to How to Use AI for Job Search shows exactly what your candidates are doing with the same tools. That is useful information when you are reading their applications. For the bigger picture, we looked at the evidence on AI replacing human jobs. We also looked at why Gen Z AI job training is changing what a junior hire can do on day one.
Frequently Asked Questions
Do companies really use AI to hire?
Yes, and most of them have for years. Applicant tracking systems were ranking CVs long before ChatGPT existed. What changed is that the same tools are now cheap enough for a company hiring two people a year.
How do companies use AI for hiring?
Mostly for writing and sorting: drafting the ad, ranking applications against requirements, finding candidates who never applied, and turning interview notes into scorecards. The decision itself stays with a person, and in several places the law now requires that.
What is AI in recruitment?
It is a catch-all term for software that reads text and ranks it. That means matching a job description against CVs, scoring answers, and summarising long applications.
Is AI hiring biased?
It can be, and it inherits the bias of whatever it learned on. Amazon scrapped an internal tool that had taught itself to downgrade CVs from women. The realistic defence is to use it for shortlisting rather than rejection, and to check who your filters screen out.
Are there laws about using AI in hiring?
There are, and they vary by where your candidates are. New York City requires an annual bias audit and advance notice for automated hiring tools. Illinois requires consent before AI analyses a video interview. The EU AI Act treats hiring as high risk.
Can AI replace a recruiter?
Not the part that matters. It can produce a shortlist far faster than a person can. But it cannot persuade a good candidate who already has a job to take yours. It cannot tell you whether someone will still be happy in two years.
What is the best free AI tool for hiring?
The free tier of whichever general model you already use will handle the ads, the screening and the notes. For the matching itself you need a platform that holds candidate profiles, and those usually charge the employer once you hire.