500 free credits, no card. Try Smalt AI →

People

How do I write interview questions from a job description

Pull the key competencies from the job description, write one open-ended behavioral or situational question each, add probes, and score them against a rubric.

Smalt AI team ·

Pull the critical competencies out of the job description, then write one open-ended question for each. Add follow-up probes and a scoring rubric with example answers before the first interview, and ask every candidate the same questions. Google’s re:Work guide and US Office of Personnel Management (OPM) guidance describe this structured approach.

How to do it

1. Pick the competencies

Read the job description and list what the person must be able to do. If it is a wish list of twenty bullet points, cut it down to what the role cannot do without. You can ask the hiring manager, or people who already do the job, which items matter most.

What Google says: competencies can include role-related knowledge, problem-solving and leadership, with the leadership expectation scaled to the level of the role (Google re:Work).

What OPM says: questions should be job-related and tied to competencies identified in a job analysis (OPM).

2. Choose the question type

OPM separates two kinds. Situational questions ask what a candidate would do in a hypothetical scenario. Behavioral questions ask for a real past example, such as “tell me about a time when…”. OPM notes that behavioral questions show higher validity for complex professional and managerial work (OPM structured interviews). For a senior role, lean behavioral. For an entry-level role where candidates may have little past experience to draw on, a situational question can be fairer.

3. Write one question per competency

Google’s guidance is one clear, open-ended question per competency, built from a realistic scenario in the job. OPM adds that questions should be clear and concise, and suggests superlatives such as “best” or “most” so the candidate picks a meaningful example, and a STAR structure for the answer (OPM).

Suppose the job description for a customer success manager says “owns renewals for a book of accounts”. A behavioral question for that line could be: “Tell me about the most difficult renewal you have handled. What was at stake, what did you do, and what happened?“

4. Add probes

Decide the follow-up questions before the interview, not during it. Google’s guide calls for predetermined follow-ups (Google re:Work), and OPM refers to standardized probes (OPM structured interviews). For the renewal question, probes might be “What did you try first?” and “What would you do differently?” Every interviewer then asks the same ones.

5. Build the rubric

For each question, write example answers at four levels: poor, borderline, solid and outstanding (Google re:Work). Write them before you meet any candidate. If you cannot describe a solid answer, the question is probably too vague or the competency is not really needed for the job.

Here is our own illustration for the renewal question. It is not from Google’s guide, and your team’s standard may differ.

LevelSample answer
Poor”We offered a discount and they stayed. I don’t remember the details.”
BorderlineDescribes an at-risk renewal but says little about their own actions or the result.
SolidNames what was at stake, the steps they took and the outcome.
OutstandingAs solid, plus finds the root cause, involves the right people and changes a process afterward.

6. Run it consistently

Ask every candidate the same questions and take detailed notes (Google re:Work). On a panel, OPM has each panelist rate individually and then discuss (OPM structured interviews). Google also recommends training and calibrating interviewers and refreshing the questions periodically (same guide).

7. Review the set

Have the hiring manager or current job holders look at the questions, and read each one aloud to check that it is clear (OPM).

Doing it with an AI coworker

If you would rather not draft this from scratch, you can give Smalt the job description as a Google Doc or Word file and ask for the competencies it implies, one behavioral or situational question for each, a few probes per question, and a four-level rubric. It can put the set in a Word document or a Google Doc for you to edit.

What you should expect to receive is a first draft of the structure above. The choices stay with you: which competencies matter, which question type fits the level, and what a solid answer looks like in your team.

If the hiring manager is on Notion, the final set can live there too.

What to check before you rely on it

The job description may be wrong. The method is only as good as the competencies. A job description that is out of date, copied from another company or padded with wishes will produce questions about the wrong things. Compare it with what the hiring manager says the person will actually do.

Structure helps, but it is not a guarantee. A SIOP summary of a 2022 re-analysis by Sackett and colleagues reports a mean operational validity of r = .42 for structured interviews predicting job performance, the highest among the predictors reviewed, with wide variation around that mean (SIOP). Our inference: with that much variation, a structured interview done badly may do worse than the average suggests. We have read the summary, not the original paper.

Draft questions need a person’s eye. Any generated or templated question should be read aloud, checked against the competency it claims to test, and cut if it is leading, double-barreled or tests something the job does not need. Rubric examples written without knowing your team can describe the wrong standard.

Consistency has to hold in practice. The same questions, the same probes and individual scoring only work if the interviewers are trained and calibrated. If one interviewer improvises, the comparison between candidates is no longer like for like.

Local rules differ. What you may ask, and how you must document the process, depends on your jurisdiction and your employer’s policy. The OPM pages describe US federal practice. Check your own legal and HR guidance before you use any question set.

Smalt is an AI coworker that takes jobs like this end to end and hands back the finished file.