A hiring rubric is a written set of weighted criteria that defines what a great hire looks like for one specific role, and how much each criterion matters. It's the difference between "I liked her" and "she scored 9/10 on the criterion that matters most, and here's the evidence."
Rubrics are the highest-leverage artifact in hiring because everything downstream inherits their quality. Interviews become comparable. CV review becomes consistent. Debriefs argue about evidence instead of impressions. And if you use an AI ATS, the rubric is literally the program the AI runs: the machine scores exactly what you told it to value.
This is the method we built into Rubrily, and it works just as well on paper. There's a free Excel template that implements every step below.
Step 1: Write criteria before you meet candidates
The single most common rubric failure is writing it after interviews start, which quietly turns it into a justification engine for a favorite candidate. Write the rubric when the role opens, from three inputs: what the role actually does day-to-day (not the job ad), what separated your best past hires from regretted ones, and what failure would look like in the first six months.
Step 2: Keep it to 5–8 criteria
More than eight criteria and every score matters less, because a candidate's real signal drowns in half-point noise on things nobody truly cares about. Fewer than five and the rubric can't distinguish candidates. If two criteria always score together ("communication" and "stakeholder management"), merge them. A useful shape for most roles: two or three hard-skill criteria, two execution criteria (ownership, problem-solving), one or two collaboration criteria.
Step 3: Tier the weights
Not all criteria are equal, and pretending they are is how a charming candidate with a fatal gap gets hired. Use four tiers:
| Tier | Meaning | Example (backend engineer) |
|---|---|---|
| Must Have | Weak here = weak hire, regardless of the rest | Core technical proficiency |
| Very Important | Major driver of success | Problem-solving under ambiguity |
| Important | Real contributor, compensable | Communication |
| Good to Have | Tiebreaker | Domain familiarity |
The tier answers a precise question: if this were the candidate's weakest area, would we still hire them? If no, it's a Must Have. If probably, it's Important or below.
Step 4: Define what evidence looks like
For each criterion, write one line describing observable evidence: "walks through a real system they built and handles probing follow-ups" beats "strong technical skills." This line is what keeps two interviewers (or a human and an AI) scoring the same answer the same way. If you can't describe the evidence, you can't score the criterion; rewrite it until you can.
Step 5: Score 0–10, with a written justification, every time
Two rules make scoring trustworthy. Justify every score in a sentence: what did you actually observe? A score you can't justify is a feeling wearing a number. And never score what you didn't observe: if a criterion never came up, mark it unscored and exclude it from the average rather than defaulting to a 5 (which silently rewards vagueness) or a 0 (which punishes the candidate for your interview plan). This is the same "no fabricated scores" principle we hold the AI's scoring to.
How do I use a hiring rubric in interviews?
Map each criterion to at least one planned question before the interview, score immediately after (not at the end of a five-interview day), write the justification while the evidence is fresh, and compare candidates by weighted totals, then read the justifications behind any close call before deciding. For the first-round call specifically, this is packaged ready to copy in how to run a phone screen: a 20-minute script, a question bank, and a four-criterion rubric.
Step 6: Reuse and calibrate
A rubric is an asset, not a one-off. Reuse it across every candidate in the role (that's the entire point), and calibrate it between roles: if everyone scores 8+ on a criterion, it isn't discriminating and needs a harder evidence bar; if a regretted hire scored well, find which criterion failed to catch the problem and sharpen it.
Doing this at scale
Everything above is manual-friendly, and the template computes the weighted math for you. The scaling problem is applying it to two hundred applicants: nobody has interview hours for that. That's the gap AI screening closes: in Rubrily the same rubric drives an async AI interview and an AI CV evaluation for every applicant, each criterion scored 0–10 with a written justification. The method stays yours; the labor becomes the machine's.
FAQ
What's the difference between a rubric and a scorecard? The rubric is the definition (criteria, weights, evidence descriptions), created once per role. A scorecard is one candidate's scores against that rubric. One rubric, many scorecards; comparing scorecards only works because the rubric behind them is fixed.
Should candidates see the rubric? Share the criteria themes, not the scoring mechanics: candidates prepare better and self-select more honestly when they know what the role values. Keep the exact weights internal, since publishing them invites answer-shaping rather than honest signal.
How is a rubric different for AI screening vs human screening? The rubric itself is identical. That's the point. What changes is enforcement: humans drift under fatigue and charm; the AI applies the same criteria, weights, and evidence bar to candidate two hundred as to candidate one, and writes down its justification for every score.
Get the free template: a working Excel rubric with weighted tiers, 0–10 scorecards with required justifications, and automatic candidate ranking. Download it here, or let Rubrily run your rubric automatically — join the waitlist →