Honest AI presence
Candidates are told they are speaking with AI. The interviewer is voice-first and does not use a human face, clone, or photorealistic avatar.
Trust Center
AI interviews that can be explained after the decision.
Recruiter AI is built around recruiter control, candidate notice, evidence-linked scoring, and human review. This page explains what the platform evaluates, what it refuses to evaluate, and how interview records are handled.
Operating principles
Human decisions
The platform can recommend, summarize, and score evidence. It does not automatically reject or advance candidates.
Evidence over black boxes
Scores are designed to point back to transcript segments, screenshots, integrity signals, and the locked interview plan.
Dignified verification
Integrity checks are visible context for reviewers, not hidden surveillance and not automatic decision logic.
How AI Interviews Work
The recruiter briefs the interview before a candidate ever joins.
01
Plan
A recruiter creates a role, writes or speaks a brief, reviews the generated interview plan, edits the rubric, and locks the plan version.
02
Invite
The candidate receives a secure link. They do not create an account. Before starting, they see consent, device checks, and clear interview expectations.
03
Review
After the interview, the recruiter reviews transcript evidence, score dimensions, integrity context, and the AI recommendation before making their own decision.
How Scoring Works
Scores are tied to the locked plan and supporting evidence.
Each interview session references one locked interview plan version. The scorer evaluates the conversation against that plan's rubric, target competencies, questions, and red flags.
The scorecard is designed to cite evidence rather than produce an unexplained number. Evidence can include transcript segments, screenshots, completion context, and integrity signals.
Each dimension carries its own confidence, not just a number: a confidence band, what the evidence supports, and where it falls short. Thin, indirect, or degraded evidence lowers confidence. It does not quietly change the score.
A recommendation is separate from the final reviewer decision. The system records both so a team can see what AI suggested and what humans actually decided.
What We Never Evaluate
The system is not allowed to score protected or irrelevant traits.
- Race, ethnicity, religion, gender, age, disability, or other protected attributes.
- Physical attractiveness, room background, lighting quality, or camera image quality beyond basic session usability.
- Accent, voice style, or whether English is the candidate first language.
- Schools, names, photos, or other proxies when they are not role-relevant evidence.
- Integrity signals as standalone rejection rules.
Integrity Signals
Verification is context, not punishment.
Recruiters need to know whether an interview was completed under reasonable conditions. Candidates also deserve to know what is being observed. Integrity signals are surfaced as review context and are never automatic rejection rules.
We do not use face recognition, voice recognition, gaze tracking, mouth-movement analysis, face or voice matching, or any biometric identity check. No face-detection or voice-fingerprinting model runs on the candidate. Integrity context is limited to non-biometric signals like the ones below.
Tab focus
Fullscreen transitions
Response timing
Specificity probes
Constraint pivots
Fictional-premise probes
Multiple-display checks
Session completion
Device checks
Some role-specific interviews may include a deliberately fictional premise to check whether an answer is confidently invented or honestly uncertain. Any resulting signal is review context for a human, not a verdict.
System Reliability
Our glitches are our problem, not the candidate's.
Provider slowdowns, dropped connections, reconnects, and audio failures are recorded as system faults, kept separate from anything about the candidate. Those windows are excluded from scoring, and evidence captured during them is marked low-fidelity.
A fault can only ever remove weight from an evaluation. It never adds a penalty. The candidate's report shows how much time was affected and says plainly that it was not counted against them.
Identity
Attestation and evidence, not biometric verification.
Candidates reach the interview through a signed private link and attest that they are the invited person and will complete it themselves. The platform may surface identity and integrity confidence signals for a human reviewer to weigh.
It does not verify legal identity, request ID documents, perform face or voice matching, or store biometric templates or embeddings, and it never claims a candidate's identity is “verified.” Identity stays a human judgment backed by evidence, never an automated gate.
Candidate Data Rights
Candidates should understand what is captured and how to ask for it.
- Candidates receive notice before interview capture begins.
- Candidates give explicit consent before the interview starts.
- Session records are tied to the employer role and candidate invitation.
- Candidates can request access, export, correction, or deletion by contacting the employer.
- Hard deletion is reserved for explicit candidate purge or legal deletion flows.
Human Oversight
The AI interviewer does not own the hiring decision.
AI recommendation
The platform can summarize evidence, score against a rubric, surface concerns, and suggest a hiring recommendation.
Reviewer decision
A recruiter or hiring team member reviews the scorecard and records advance, reject, hold, follow-up, or other workflow decisions.
AI-In-Hiring Compliance Posture
The product is designed for auditability from the start.
AI systems used to evaluate candidates can trigger employment, notice, audit, and data governance obligations. Recruiter AI is structured so employers can document the interview plan, candidate consent, scoring evidence, integrity context, and human decision path.
This trust center is not legal advice. It is the product posture: transparent candidate notice, human oversight, audit logs, evidence-linked scorecards, and no autonomous hiring decisions.
Security And Privacy Overview
Data isolation and least exposure are part of the architecture.
- Tenant data is company-scoped in the application and backed by Postgres row-level security.
- Candidate interview links use signed invitation tokens, not candidate accounts.
- Realtime session access uses a separate short-lived session token.
- Critical events are written through a durable Postgres queue instead of ephemeral pub/sub.
- Provider integrations are isolated behind backend adapters so browsers never call AI providers directly.
For candidates