How to build an evidence-based internship screening rubric

An internship screening rubric is a short list of job-related capabilities, the evidence that would demonstrate each one, and a consistent way to score that evidence. Build it before applications arrive. This prevents a polished résumé, prestigious school, or one memorable answer from quietly becoming the standard for everyone else.

The purpose is to decide who deserves a closer look, not to claim that a number measures a person's worth. A useful rubric helps reviewers explain why an application did or did not advance and identify what remains unknown. Keep the actual responsibilities in view: an intern who will write customer emails and track campaign results needs different evidence from one who will debug a backend service.

Start with the work, then choose three to six criteria

Write down the first three projects this person would plausibly own in their first month. For each, ask what a successful junior contributor would actually do. Translate those activities into observable criteria. For a content and operations internship, a starting set might be:

These numbers are illustrative, not universal passing rules. A minimum is a prompt for review: what does the role truly require on day one, and what can be taught? Do not set every minimum to the maximum. For an internship, growth potential and evidence of learning may matter more than years of experience. If a requirement is legally or operationally essential, document it separately and confirm it with a human.

Avoid criteria such as “culture fit,” “top college,” or “sounds confident.” They are hard to score consistently and often stand in for unrelated signals. In the United States, the EEOC advises employers to use selection procedures that are job related and appropriate for the work; the EEOC's employment test guidance is a useful starting point. Apply local law and your own counsel where relevant.

Define the anchors before anyone is scored

For each criterion, describe what scores mean. You need not write six elaborate paragraphs for every item, but reviewers must know the difference between no evidence and weak evidence. For writing clarity, for example:

• 0: No writing sample or answer that addresses the prompt.
• 1: A sample exists, but its audience or intended action is unclear.
• 2: The sample communicates a coherent idea with limited audience adaptation.
• 3: The candidate chooses an appropriate message and explains a relevant choice.
• 4: The message is concise, audience-specific, and supported by feedback or iteration.
• 5: The candidate shows strong craft, tests or feedback, and a thoughtful revision tied to an outcome.

A score of zero because a PDF could not be read should be labeled unavailable evidence, not “unable to write.” Ask for a usable file or have a person read the original. A portfolio may be private, inaccessible, or produced under a nondisclosure agreement. Offer another way to demonstrate the skill.

Ask questions that produce evidence

“Are you passionate about marketing?” tells you little. “Describe one message you wrote for a particular audience. What did you want them to do, and what changed after you saw the response?” is closer to the criterion. Ask what the applicant personally did, what constraints they faced, what they learned, and how they know. Keep the form reasonably short. Our application question guide has a candidate-facing template.

If you use a practical exercise, score it with its own criteria. Do not retrofit the application rubric to a different kind of work. For example, an application can establish that someone has tried paid campaigns; a 45-minute exercise can show how they would structure a new test. The assignment design guide explains the handoff.

Calibrate with a small sample

Before publishing, give two reviewers three fictional or anonymized sample responses. Let them score independently, then compare. A disagreement of two points usually means the evidence description is ambiguous. Resolve the language, not the person. Record one or two examples of each score band. Repeat the calibration if you change the role materially.

During real reviews, write a one-sentence evidence note for every score that affects progression. “Built the landing page and measured signups for two variants” is auditable; “smart candidate” is not. If information is missing, record the question you would ask next. Treat any automated score as a first pass that needs human review, especially at a threshold or when files are unreadable. See how to interpret AI screening scores.

Put the rubric into Badal

In Badal's workspace, create a job post with organization and recruiter names, a clear role description, and each criterion's name, evidence description, relative weight, and 0–5 minimum. The public form can collect a résumé or work material and context questions. Badal evaluates submissions against the role's criteria and shows criterion scores and evidence in a candidate report. The service does not source candidates, and the recruiter remains responsible for final keep or reject decisions.

You can edit criteria while a role is live. New evaluations use the current version; older reports retain the rubric version used at the time. Do not compare raw scores from two materially different rubric versions as though they were one contest. The live criteria guide covers a fair change process. Badal's documentation describes the current controls and the OpenAPI schema is the exact contract for agents.

Reusable agent brief

Help me design a screening rubric for [role] at [organization]. Ask one question at a time about the work, must-have constraints, teachable skills, and evidence I can reasonably request. Propose three to six job-related criteria, a plain-language evidence description, relative weight, and 0–5 minimum for each. Show score anchors and two likely ambiguous cases. Ask me to approve the rubric before creating or changing the role in Badal.

The finished rubric should let a second reviewer reach a similar conclusion from the same application. If it cannot, revise the criteria before increasing application volume.

Frequently asked questions

How many criteria should an internship rubric contain?

Start with three to six. If you need ten, you may be combining multiple jobs or scoring nice-to-have skills as if they were essential. Group related evidence and retain only distinctions that would change a real advancement decision.

Should missing evidence receive a zero?

Record why it is missing first. An unanswered required question differs from an unreadable upload or a skill the form never asked about. A numeric score without that distinction can make an operational problem look like a capability judgment.