How to write a screening rubric

Why write one at all

Without a rubric, an interviewer ends up with an overall impression. Schmidt and Hunter describe that as the mark of an unstructured interview: answers aren't scored one by one, and the rating rests on summary judgments. Unstructured interviews were well behind structured ones in both major reviews of the research.

A rubric fixes the question you are answering. You stop asking whether you liked someone and start asking whether they showed what the job needs. It also gives you a written reason for every decision, which helps when you have to explain one.

Step 1: start from the job, not the person

Don't start from your best current employee or the last person who left. Start from the work. Write down what someone in this role has to do well in their first six months.

Then cut. Some things are better checked elsewhere, such as a licence, a right to work, or a technical skill you plan to test. A first-round rubric should only hold what a person can show by talking about their work.

Step 2: choose three criteria

Three is enough for a first round. More than that and you won't have time to hear evidence for each one. For the same reason, combine criteria that overlap.

A good criterion passes these tests:

• It describes something a person does, not who they are.
• It's needed for this job, not just nice to have.
• Two reasonable interviewers would agree on what it means.
• A candidate can show it by describing past work.

Step 3: write the scale

Use a short scale, such as 1 to 5, and describe the points. You don't need a paragraph for each number. Describing the 1, the 3 and the 5 is usually enough, and the points between take care of themselves.

Here is one criterion for a customer support role, Handling hard conversations:

• 1: Gives no real example, or blames the customer.
• 3: Gives a real example and stays calm, but the fix is mostly following a script.
• 5: Gives a specific example, explains how they read the situation, and shows what changed for the customer.

Step 4: tie every question to a criterion

Each question should exist to collect evidence for one criterion. If you can't say which one, cut the question. If a criterion has no question, you have nothing to score it on.

One or two questions per criterion works for a first round. Give the criterion you care about most two questions, and leave yourself room for a person to take a different path to a strong answer.

Step 5: decide what doesn't count

Write down what scorers should ignore. Otherwise it will creep back in. Common things to rule out are accent, grammar, confidence, and how polished someone sounds, unless the role genuinely depends on it.

Avoid vague criteria such as culture fit or energy. They are hard to score and easy to fill with personal preference. The US Equal Employment Opportunity Commission advises against asking for information that would show race, sex, national origin, disability, age, religion, colour or ancestry, and says such inquiries may be used as evidence of intent to discriminate.

Also decide what happens when there isn't enough to go on. A short answer isn't proof of weak skill. It's better to record that you couldn't assess the criterion than to give it a low score.

Step 6: test it before you use it

Ask two colleagues to score the same practice answer on their own. If they land far apart, the wording is too loose, so tighten it and try again. This takes less time than you would expect, and it's the fastest way to find a criterion that only makes sense to the person who wrote it.

Using the rubric well

A rubric only works if it's used the same way for everyone.

• Score each criterion from what the candidate said, and note the evidence.
• Finish one criterion before you look at the next.
• Don't change the rubric halfway through a role.
• Read the evidence before you decide, not just the total.
• Review the rubric after each role and fix what didn't help.

Rubrics in Vettika

In Vettika, you paste a job description and AI drafts questions and a rubric. After each interview, AI scores every criterion from 1 to 5 and quotes the candidate as evidence. Each quote is checked against what the candidate actually said, and any quote that can't be verified is dropped. If someone says too little, the result is "not enough to assess" rather than a low score. People on your team then decide.

Sources

• Schmidt and Hunter (1998), The validity and utility of selection methods in personnel psychology, Psychological Bulletin
• Sackett, Zhang, Berry and Lievens (2022), Revisiting meta-analytic estimates of validity in personnel selection, Journal of Applied Psychology
• US EEOC: Prohibited employment policies and practices

Common questions