AI for hiring: clear job posts, fair interviews, people deciding

How to use AI for hiring, in short

Use AI for hiring to handle the writing and the structure, while a person makes every decision about people. Your AI is good at turning your notes into a clear job post, a set of interview questions, a scorecard and a fair work sample. It becomes risky when it judges candidates on its own, because it can repeat bias and you may not be able to see why it chose one person over another.

The method in this guide has five steps, and you start with what the job must achieve and write a clear job post from it. Then you ask every candidate the same questions, score them the same way and give a short, fair work sample, before comparing the evidence and deciding yourself.

You also get three prompts, a scorecard you can copy and a plain summary of the legal risks of AI screening. Laws differ by country, state and city, so check the rules wherever you hire.

Step 1: Start with what the job must achieve

It's tempting to start a job post with a list of skills. Start instead with outcomes, meaning what this person should have achieved after six or twelve months. Outcomes tell you which skills really matter, and they show candidates what success looks like.

Answer these questions before you write anything:

• What problem will this hire solve, and what is going wrong or not getting done today?
• What three to five results should they deliver in their first year?
• Which skills are truly needed on the first day, and which could they learn on the job?
• What does a normal working week look like in this role?
• Can the business afford the hire? The free employee or freelancer costs tool compares a hire with a freelancer, and the new hire payback estimate shows when a hire covers their cost.

Write the outcomes as measurable results rather than as everyday tasks. "Reply to every customer email within one working day" is a result you can measure, while "handle customer emails" is only a task.

Step 2: Write a job description with AI

A good job post is clear, honest and reasonably short. It explains what the job is for, what the person will do, what they must already know, what you offer and how to apply. It leaves out wish lists and vague phrases like "rockstar" or "fast-paced".

Give your AI your outcomes and the real facts of the job, and ask it to flag any wording that could put off good candidates for no reason. Typical examples are needless degree requirements or words that hint at a preferred age.

You are an experienced hiring manager at a small business. Write a clear, fair job post from my notes.

The role: [job title], [full-time or part-time], [location or remote], [pay range].
What this hire must achieve in the first year: [three to five outcomes].
A normal week: [main tasks].
Must have on day one: [only true needs].
Nice to have: [things they could learn].
What we offer: [pay, hours, benefits, training, flexibility].
About us: [two or three true facts about the business].
How to apply: [steps and closing date].

Steps:
1. Write the post in this order: what the job is for, the first-year outcomes, what you'll do, what you need, what we offer, how to apply.
2. Keep it under 450 words. Use short sentences and "you".
3. Keep your list of must-haves to five or fewer. Move anything else to nice-to-have.
4. Under the post, list any wording that could put off qualified people for no good reason, such as needless degree demands, years-of-experience counts, or words that suggest age, gender or a type of person. Suggest a fix for each.
5. List anything I left out that candidates will want to know.

Rules: use only facts I've given. Don't invent benefits, values or perks.

Read the draft as a candidate would, and ask whether a strong person would understand the job and whether they fit. Want your AI to know your business before it writes? Our guide on giving your AI context about your business shows how to set that up once.

Step 3: Use structured interview questions and a scorecard

A structured interview means every candidate gets the same questions, in the same order, scored against the same guide. This approach is fairer and makes candidates easier to compare, whereas unstructured chats make it easy to favour people who seem like you.

Write questions that ask for evidence

Ask about real past work rather than hypothetical puzzles or trick questions. "Tell me about a time a customer was angry with you, and what you did" produces evidence, while "Are you good with people?" usually produces a confident yes.

You are helping me run fair, structured interviews for one role.

The role and its first-year outcomes: [paste from your job post].
The four or five things I most need to assess: [for example, customer care, written English, planning, using our booking system].
Interview length: [minutes]. Number of interview stages: [number].

Steps:
1. For each thing I need to assess, write two questions about real past work ("Tell me about a time...").
2. For each question, add one follow-up probe to get specific detail.
3. For each question, describe a strong answer, an average answer and a weak answer, using what the person did, not how they sound.
4. Build a scorecard: one row per thing assessed, a 1 to 4 score, and a box for the evidence behind the score.
5. Say which stage should cover each question, so nothing is asked twice.

Rules:
- Don't include questions about age, family, health, religion, nationality or other personal traits.
- Don't score "culture fit" or confidence on their own. Tie every score to job evidence.

A scorecard you can copy

Each interviewer fills in their own scorecard straight after the interview, before talking to anyone else, so that one strong opinion can't sway the rest of the panel.

On this scale, 1 means no evidence, 2 means some evidence, 3 means clear evidence and 4 means strong evidence beyond what the job requires.

Step 4: Set a fair work sample

A work sample is a small, realistic piece of the job that candidates complete before you decide. For a support role, it might mean replying to three sample customer emails, while for a bookkeeper it might mean finding the errors in a short sample of transactions.

• It matches a real task from the job, using made-up data.
• It takes under two hours, and you tell candidates how long it should take.
• Every candidate gets the same task, the same time and the same instructions.
• You've written the scoring guide before any candidate attempts the task.
• It isn't real unpaid work you'll use, and if it's long, you consider paying for it.
• You offer adjustments for candidates who need them, such as extra time.
• You tell candidates whether they may use AI tools for the task, and score everyone the same way.

Ask your AI to draft the task and the scoring guide together, and then complete the task yourself first. If it takes you 40 minutes, it will probably take a candidate longer.

Should you use AI to screen candidates?

This is the stage where AI for hiring carries the most risk. Screening means using a tool to rank, filter or reject applicants, and a tool that learns from past hires can repeat past bias. A tool that reads CVs can also favour certain schools, words or career gaps without telling you why. You still own the decision, and in many places the law treats it that way.

What the law says in two examples

• United States. The US Equal Employment Opportunity Commission (EEOC) explains when a hiring test or screening tool can be unlawful. That happens if it screens out a protected group and isn't "job-related and consistent with business necessity" (EEOC guidance on employment tests and selection procedures). That applies whether the screen is a quiz or a piece of software.
• New York City. NYC's law on automated employment decision tools says employers can't use such a tool unless it has had a bias audit within the past year. The audit results must be made public, and candidates must receive notice (NYC Department of Consumer and Worker Protection, as of November 2026).

Other countries, states and cities have their own rules, and new ones keep arriving. Before you use any AI screening tool, check the rules where you hire and where your candidates live, and ask an employment lawyer if you're in doubt.

Safer ways to use AI in screening

• Use AI to summarise each application against your written must-haves, then read every application yourself.
• Remove names, photos, ages and addresses before you paste applications into your AI.
• Never let it reject anyone automatically, so that a person always decides who moves forward.
• Write down exactly how you use AI in hiring, and tell candidates about it. The HR AI use policy writer drafts that policy and a staff notice, although it is not legal advice.

Our guide on stopping AI making things up explains why an AI summary can sound certain and still be wrong, which matters even more when someone's job is at stake.

You are helping me compare job candidates fairly. You do not decide who to hire. I do.

The role's must-haves and outcomes: [paste].
The scorecard and scoring guide: [paste].
For each candidate, with names replaced by "Candidate A", "Candidate B" and so on: their completed scorecards, interview notes and work sample result. [paste]

Steps:
1. For each candidate, list the evidence for each scorecard row, quoting the notes.
2. Flag any score that has no evidence behind it, or where interviewers disagree by two points or more.
3. Flag notes that judge personal traits instead of job evidence, such as "not a good fit" with no reason given.
4. Show a side-by-side table: rows are the scorecard items, columns are candidates.
5. List the questions I should still ask, or the references I should check, before deciding.

Rules: don't rank candidates or recommend who to hire. Don't guess at anything the notes don't say.

Worked example: what Spotify's public hiring page shows

Spotify, the music and podcast streaming company, describes its hiring process on its How we hire page. As of November 2026, the page describes the process like this:

• Candidates are asked to show the work they've done and the impact it had, and cover letters, portfolios and personal websites are welcome alongside a CV.
• The first stage is a video or phone interview with a recruiter, followed by an interview with one or two members of the team.
• A final round usually involves several people from across the business, and when there are several interviewers, one leads while the others shadow.
• Beyond job skills, Spotify lists questions it asks itself. Do the candidate's values match the company's, are they adaptable, collaborative and driven, and do they take ownership of their work?
• Candidates who need adjustments are asked to contact their recruiter.

The page doesn't say that Spotify uses AI to write its job posts or screen candidates, and we're not claiming it does.

What a small business can copy

The useful idea is that each stage has a clear job, and the qualities Spotify looks for are written down in advance. A small business can do the same by writing down, for each quality, what evidence would show it.

Here's how prompt 2 might turn one of those qualities into a scorecard row. This is an example draft written for this article, not Spotify's real scorecard:

Quality: Takes ownership of their work.
Question: Tell me about a time something you were responsible for went wrong. What did you do next?
Strong answer: Names their own part in it, describes the fix they led and what they changed after.
Weak answer: Blames others or stays vague about what they did.

Notice that the row scores what the person did rather than how confident they sounded, which keeps the judgement tied to the job.

Common mistakes, and what to do instead

• Letting AI pick the shortlist hides its reasoning from you, and it may repeat bias. Use it to summarise against your must-haves, then read and decide yourself.
• A wish list of must-haves puts off good people, because ten "essential" skills discourage candidates who meet seven. Keep your list of must-haves to five or fewer.
• Different questions for each candidate mean you end up comparing conversations rather than evidence, so use the same questions and scorecard for everyone.
• Scoring "fit" by feel is unreliable, because "I just liked them" often means "they're like me". Tie every score to evidence about the job.
• A work sample that's really free work is unfair to candidates, so keep it short and made up, and pay for longer tasks.
• Pasting full CVs into a chat shares personal details you don't need, so remove names, contact details and photos first.

Where AI Blocks helps

All of these are free: copy the prompt, or download the skill, workflow or specialist, and use it in your own AI. AI Blocks doesn't screen candidates or send any messages on your behalf.

• Outcome-focused job description writer: paste the problem your new hire must solve to get a clear six-month picture and only the skills that are needed.
• Inclusive job post writer: write a job post from a role or an old ad, with clear must-haves, 30, 60 and 90 day goals and biased wording fixed.
• Structured interview guide builder: get set questions, strong and weak answer signs and a scorecard, so you judge every candidate the same way.
• Work sample test designer: design a work sample for any role, with a scoring guide and fairness checks.
• Fair candidate comparer: paste interview transcripts and your scorecard to get scored answers, evidence gaps and bias checks.
• Hiring interview and candidate comparison planner: a workflow covering interview stages, scorecards, a work test, comparisons and reply drafts.
• Hiring document and candidate reviewer specialist: a ready-made specialist for job posts, candidate comparisons and offer clauses to check.
• Hiring volume planner: a free browser tool where you set your hiring goal and pass rates to estimate the applications, interviews and offers you need.

If you'd like these steps to run as one routine, our guide to AI agents for small business explains how to set one up with a person approving each step.

Questions people ask

Can I use ChatGPT to write a job description?

Yes. Give it the job's outcomes, the real must-haves, the pay and what you offer. Ask it to keep the must-haves short and flag wording that could put off good candidates. Check every fact before you post it.

Is it legal to use AI to screen job applicants?

It depends on where you hire, because some places, such as New York City, require a bias audit and notice to candidates for automated tools. Anti-discrimination law applies wherever a screen is used. Check your local rules, and keep a person making the decisions.

What are good interview questions to ask?

The best questions ask about real past work and are tied to what the job needs. "Tell me about a time you..." questions work well with a follow-up asking for detail. Ask every candidate the same ones, and score them against the same guide.

What is an interview scorecard?

It's a short form listing what you're assessing, a score for each item and the evidence behind each score. Each interviewer fills one in alone, straight after the interview, which makes candidates easier to compare fairly.

Can AI help me screen candidates without bias?

Not by itself, because AI can repeat bias from the data it learned from. It can help by summarising applications against your written criteria with names removed, but a person should still read the applications and make the decision.