The job-fit questionnaire

What you need

• The job's real duties, written plainly: what a normal day or week looks like
• The two or three things that matter most to you in this role. Mine are availability and reliability.
• A Claude account (Free works). If staff or candidate details will go in, a Team account is safer, because it doesn't train on your content by default.
• A spreadsheet for the score sheet and the records

Start with the duties. When people take a job and don't stay, you most likely pitched it incorrectly. Describe the workday as it actually runs: the quiet hour, the busy morning, the phone ringing while a client waits at the desk.

Ask for the score sheet as a spreadsheet, not an artifact, which is a small page Claude builds inside the chat. Free can't store data in an artifact between sessions.

On Free, Pro or Max, open Settings, then Privacy, and turn off "Help Improve our AI models" before any answers go in. That switch controls whether your chats can be used to train Claude. On any plan, put numbers on the sheet instead of names, and keep the name list in a file that never goes into Claude.

The prompt

Replace the words in [brackets] with your own.

Some states and cities have their own rules on AI in hiring, such as telling candidates before AI is used. New York City and Illinois are two examples. Check your state labor department's site before the first questionnaire goes out.

The first line has Claude work like an industrial-organizational psychologist, the specialist larger companies hire to build hiring tests. It's a method, not a diagnosis. The last line tells Claude where its role ends. It drafts. You decide.

Why 15 to 25 questions? Ten leaves about three questions per section. With more, no single lucky answer decides the outcome.

Indirect doesn't mean hidden. Candidates know it's a hiring questionnaire about the job. Each question simply doesn't announce what it scores. Ask "Are you reliable?" and everyone says yes. Ask what they'd do when a coworker doesn't show on a busy Monday, and you learn something useful.

Then use this second prompt in the same chat.

That sheet only calculates from your scores. Claude never scores or ranks a candidate on it. You do. Set the weights before any answers come in, and keep them the same for everyone.

Steps

1. List the duties and what matters most, the way you'd explain a workday to a new hire.
2. Paste the first prompt. Edit the questions until you'd ask every one in person.
3. Cut any question you couldn't explain to the candidate or to a lawyer.
4. Build the score sheet with the second prompt and set your weights.
5. Add these lines to the top of the questionnaire: "This job-fit questionnaire asks how you'd handle the work in this job. AI helped us write these questions. A person reads every answer and makes every decision."
6. Send it to every candidate for the role. Score each answer with your guide, make the call and keep the records.

What it looks like when it's done

Here's a sample for a made-up job: a front-desk coordinator at a 12-person insurance office. The coordinator greets clients, runs the appointment calendar, answers phones, handles upset clients at the desk and covers when the schedule changes. Say you asked for 20 questions: eight on reliability and six each on skills and people. Here are three.

Sample question 1 (reliability and commitment): the schedule change

On Tuesday, a coworker asks you to cover their Thursday evening desk hours, when the office stays open late for client appointments. You're already scheduled to open early Friday and can only cover one of the two. Which is closest to what you'd do?

• A. Say yes and figure out Friday later.
• B. Say no, since those aren't your hours.
• C. Say you can't do both, and point them to the office manager or the shared schedule.
• D. Say yes, then tell the office manager Friday morning that you can't open.

Scoring guide: C scores 4. It keeps the promised schedule and arranges cover properly. B scores 3, A scores 2 and D, the last-minute gap, scores 1.

Sample question 2 (job skills): the double-booked appointment

It's 9 a.m. on a Monday. Two clients walk in for the same 9 a.m. appointment with the same agent, and the phone is ringing. What do you say, and what do you do next?

• 4: Greets both right away, answers or returns the call, offers an honest wait or a new time, tells the agent and checks back.
• 3: Honest wait or a new time and checks back, but doesn't tell the agent.
• 2: Fair wait, but no plan to check back or tell anyone.
• 1: Promises a short wait to keep them happy.

Sample question 3 (clients and teammates): the upset client

Tell me about a time a client or customer was upset about something you didn't cause. What happened, what did you do, and how did it end?

• 4: A real moment. Listened first, fixed what they could and brought in the agent or a manager when it was beyond their authority.
• 3: Stayed calm and fixed it, but skips what they did or how it ended.
• 2: Talks in general, such as "I always stay calm," with no real moment.
• 1: Argued, blamed a coworker or walked away.

None of the three asks about the candidate's life or personality. Question 1 never says "reliability," and the candidate still knows it's about the schedule.

Here's one row of the score sheet, question columns hidden. Every number comes from scores you typed.

Read across the rows, start with the reliability reading and make the call yourself.

Plan the reading time, too. Twelve candidates and 20 questions make 240 answers, or two to four hours of scoring. Put it on the calendar like any other appointment.

Major corporations have paid tens of thousands of dollars for hiring assessments over the years. Some of the well-known assessment platforms start at $10,000 a year and rise with headcount. This version costs you a Free account, about 90 minutes and a spreadsheet.

Check the result

• Every question ties to a real duty without announcing what it scores.
• Reliability and commitment carries the most weight.
• Every candidate got the same questions.
• Nothing on the do-not-ask list appears.
• The scoring guide is yours, and you could defend each score.
• A person, not Claude, decided who moves on.
• Candidates were told what the questionnaire is and that AI helped.

If it goes wrong

• Claude adds a personal question, such as "How long do you see yourself staying?" Cut it, then tell Claude: "Rewrite question [number] so it's only about the job."
• The questions drift into personality, like "Are you a people person?" Tell Claude: "Replace any question about what kind of person they are with a real job situation."
• A question gives the answer away, such as "Would you ever skip a workday?" Tell Claude: "Rewrite question [number] as a job situation where every option sounds reasonable."
• The sheet ranks, sorts or fills in scores. Tell Claude: "Remove every rank, sort, label and score you added."
• Claude returns a block of code instead of a spreadsheet. Turn on "Cloud code execution and file creation" under Settings, then Capabilities. On Team, an admin turns it on under Organization settings.
• You forgot the disclosure lines. Add them now, and send them in a short note to anyone who already answered. Some places require notice before a candidate answers, so check your state and city rules.

Common questions

Can I use AI in hiring without breaking the law or being unfair?

This setup gives AI one narrow job: it helps write the questions and builds a blank score sheet. Every candidate gets the same questions, tied to real duties. A person reads every answer, scores it and makes every decision. Candidates are told AI helped. Claude never scores or ranks anyone, and you keep the records.

This post can't tell you whether that meets the rules where you hire. Hiring rules vary by state and city, and some places now have rules just for AI in hiring. Check with your state labor department or an employment attorney before the first questionnaire goes out.

Why not simply ask who'll stay?

Everyone says they'll stay. Guesses about who stays tend to track age, pregnancy or family plans, and you can't screen on any of those. Reliability on the job is what you can measure, and it's who you want on the schedule.

How long should I keep candidate answers?

A federal rule, 29 CFR 1602.14, requires the employers it covers to keep hiring records, including application forms, for one year. The year runs from the date the record was made or the hiring decision was made, whichever is later. Your state may require longer, so check your state labor department for its number. Keep the questionnaire, the answers, your score sheet and your notes together.

Can AI screen resumes?

Not in this setup. Claude drafts the questions and builds a blank score sheet, and you do the reading and the scoring. Having AI screen or rank people is a different job with its own rules.

Anthropic's usage policy, effective November 12, 2026, treats "screening, ranking, advancing, or rejecting candidates" as a high-risk use. For those uses, it says "a qualified person must meaningfully review" the AI's recommendation. The person a decision is about "must be clearly told that AI was used to produce it." For the full picture, read Can AI screen resumes? What's legal and what's fair.

That's a prime example of how a founder or business owner can use AI to get ahead.

Hiring rules vary by state and city. This post isn't legal advice; for anything specific to your business, check with your state labor department or an employment attorney.