Questions and scoring
Write job-related questions, guide AI scoring, and interpret scores fairly.
A good screening question asks for evidence tied to the job. Your scoring guidance tells reviewers and AI what a strong answer looks like. InsightHire's scores help you decide who to look at first. They never make the hiring decision.
Who can do what
- Org admins, recruiters and hiring managers can write questions and scoring guidance, and report AI mistakes.
- Org admins control the AI settings under Settings → Responsible AI.
Write a useful question
- Start with one requirement from the position.
- Ask for a specific example, action or work sample.
- Use plain language and ask one thing at a time.
- Leave out personal questions unrelated to the role.
- Ask a colleague to check it for clarity and bias.
Example: "Tell me about a time you handled an unhappy customer. What did you do, and what happened?" is better than "Are you good with customers?"
Reusable questions live in Templates → Question bank. Video questions are added to an interview template under Templates → Interview templates. Assessment questions are added in the assessment wizard. See Async video and Assessments.
Write scoring guidance
For each video question, fill in:
- What are you looking for in this response? The evidence a strong answer contains.
- What are the red flags? Signs of a weak or concerning answer.
AI Generate Guidance drafts both from the question. Edit the draft so it matches what your team actually values. For multiple-choice assessment questions, mark the correct options instead.
How AI scoring works
After a candidate submits a video answer, it is transcribed and scored from 0 to 100 against your guidance and the job. The question scores are combined into an interview score. Written assessment answers are scored by AI too. Multiple-choice answers are marked against the correct options.
For the full explanation, see How AI interview scoring works.
Score gates and human review
A Score Gate in a journey sends candidates down a Pass or Fail path based on their score. See Journey builder.
- Failing a gate never emails the candidate a rejection. They're flagged for your team to review.
- A candidate who chose Prefer a human review isn't scored by AI, so the gate can't fail them automatically.
Prefer a human review
Every journey starts with a short notice explaining how AI is used. Candidates can also tick Prefer a human review. If they do, AI doesn't score their answers, and a person on your team reviews them instead. Their application shows that they requested human-only review. AI résumé scoring is skipped for them, and AI avatar steps are paused until a recruiter arranges an alternative. They are never called for an AI phone screen; on an AI Phone Screen journey step they record video answers instead.
Org admins choose when the option is offered under Settings → Responsible AI, in "Prefer a human review" for candidates:
- Automatic: only where the law requires it (based on the job or candidate location, and also when the location is missing or remote)
- Always offer
- Never offer (candidates still see the AI notice)
The same page shows how many candidates saw the notice, continued with AI, or opted out in the last 90 days.
Correct the AI
- On a score, select Report AI mistake, say what was wrong, and optionally give the score you'd have given.
- On a candidate, select Correct AI (or Unreject if AI rejected them) to fix their score or put them back in the pipeline. The candidate isn't emailed.
By default a correction changes only that candidate. If an org admin turns on Use reviewer corrections when scoring later candidates in Settings → Responsible AI, recent corrections for the role also guide scoring of later candidates. Each affected score lists the notes that shaped it.
Scoring calibration
Org admins can tune AI scoring under Settings → Responsible AI → Scoring calibration:
- Grading: Harsh, Balanced or Generous
- Score adjustment: a fixed change from −15 to +15 points
Changes apply to new scores and rescoring. Existing scores don't change. The Scoring calibration & drift section compares AI scores with your own reviewers' scores for candidates who have both.
Understand the overall score
A candidate's overall score can combine résumé fit, video answers, assessment results and, where used, culture fit. The mix depends on what the position asks for and what the candidate completed.
Use the score to decide what to look at first. Then open the candidate's report and read the evidence.
Trust and authenticity signals
Trust and authenticity signals point to activity that deserves a closer look. They do not prove fraud or dishonesty. Check the context, look for a reasonable explanation, and follow your organization's review policy. See Trust score.
Calibrate your team
- Have two reviewers score the same small set of answers.
- Discuss where they differ.
- Clarify the scoring guidance.
- Compare the AI score with the evidence.
- Revisit score gate thresholds if they put the wrong people first.
