Structured interview questions and scorecard with ChatGPT

A structured interview gives every candidate the same questions and scores their answers against criteria written in advance. ChatGPT prepares the whole kit in about 20 minutes: eight behavioral questions, descriptions of strong and weak answers, a rating scale, a timed script and a scorecard. The value comes from the competencies you feed it. Vague competencies give vague questions, and the interview drifts back into gut feeling.

Competencies with a situation attached

Pick three or four competencies that actually separate good hires from bad ones in this role, and attach a real work situation to each. "Communication" alone is too broad. "Explains a delayed shipment to an angry client without promising what logistics cannot deliver" is a competency the model can write questions for. For an accountant it might be "finds the cause of a reconciliation difference under deadline pressure".

The prompt

Role: you are an HR partner who designs structured interviews. Context: role {{title}}, key competencies: {{competency 1 with a situation example}}, {{competency 2}}, {{competency 3}}. Interview length {{minutes}}, interviewers: {{roles}}. Task: prepare eight behavioral questions about past experience (two per competency) with a description of a strong and a weak answer for each, a rating scale from 1 to 4, an interview script with timing and a scorecard. Leave out questions about age, marital status, health, religion and plans for children. Format: a table competency | question | strong answer | weak answer, then the script by minute, then the scorecard. No long dashes.

The model returns a 45-minute script if you give it that length. {{roles}} tells it how to split the questions between, say, the hiring manager and HR.

Clean the question list

Read every question and ask what it measures. The model tends to slip in openers such as "tell me about yourself" or hypotheticals such as "what would you do if". Neither measures past behavior. Replace them with "tell me about a time when" questions tied to your situations. Then delete anything that touches marital status, children, age, health or religion, even indirectly: "how do you balance work with family?" is still a question about family.

The strong and weak answer columns deserve a second look. A good pair describes observable behavior: "names the specific step they took and its result" against "describes what the team did in general". If both descriptions sound like adjectives, rewrite them.

What to do

1. Choose three or four key competencies for the role and attach one work situation to each.
2. Run the prompt in a new ChatGPT chat with the role, competencies, length and interviewers.
3. Delete hook questions and hypotheticals and keep behavioral questions about past experience.
4. Remove anything touching marital status, children, age, health or religion.
5. Turn the output into one scorecard used for every candidate.
6. Fill in the scorecard right after each interview, before talking to colleagues.

Why the order of scoring matters

The structure only helps if ratings stay independent. When interviewers discuss a candidate first and score afterwards, the loudest opinion sets everyone's numbers. Score immediately, alone, then compare. The process works when all candidates got the same questions, every question had a criterion, and all scorecards were completed before the discussion.

Plans and data

The free plan handles one interview kit easily within its message limits. Plus at 20 USD per month adds Projects, handy if you keep a library of kits by role; Business at 25 USD per seat with annual billing keeps team data out of training by default. See ChatGPT pricing. Turn off Improve the model for everyone in Data controls on personal plans, and keep candidate names and ratings in the hiring system. HR managers can find the saved version on the structured interview page, and if you are choosing an assistant, see ChatGPT vs Claude.