Relevant to TestGorilla users with the Owner, Admin, and Recruiter roles.
Approx. reading time: 4 minutes
In this article
When you create a custom interview, you define what a "good" answer looks like for each question. The AI uses your criteria to provide accurate, consistent evaluations across all candidates.
Why scoring criteria matter
The AI acts as a digital grader. Better criteria leads to more accurate and useful scores. The AI uses your inputs to:
- Score candidate responses on a consistent 0-5 scale.
- Provide reasoning explaining exactly why a candidate received their score.
- Ensure fairness by applying the same rubric to every applicant.
Anatomy of good scoring criteria
Effective criteria move beyond "gut feeling" and focus on observable evidence. Think of it as a checklist for the AI.
| Attribute | Definition |
|---|---|
| Specific | Name exactly what you want to hear (e.g., "ROI"). |
| Observable | Focus on behaviors and outcomes, not hidden traits like "confidence." |
| Measurable | Include concrete elements the AI can check for. |
Example:
❌ Vague: "Shows good communication skills"
✅ Specific: "Explains complex concepts clearly, uses concrete examples, and structures the response logically."
Writing criteria by question type
Use the following templates as a baseline to guide the AI's evaluation.
Behavioral Questions ("Tell me about a time...")
Strong answer includes: A specific example from their experience, a clear explanation of their personal actions (not just the team's), and quantified results where possible (numbers, percentages, outcomes).
Situational Questions ("How would you handle...")
Strong answer includes: A structured approach to the problem, consideration of multiple stakeholders, practical and specific steps, and acknowledgment of potential challenges.
Role-Specific & Technical Questions
Strong answer includes: Familiarity with [specific tools/methods], references to relevant experience, and an understanding of [industry context] with appropriate detail.
Common mistakes to avoid
To keep the AI accurate, avoid these common pitfalls:
- Too vague: If a human can't define it, the AI can't score it. Be specific about what to listen for.
- Too long: Stick to 2-4 key elements. Overloading the criteria can dilute the score quality.
- Trait-based: Avoid descriptors like "is confident" or "seems nice." Instead, look for "speaks clearly" or "uses professional language."
Tip: Review the AI scores for your first few candidates. If the scores don't align with your judgment, refine your criteria to be more specific about the elements you value most.