The AI advises, the human decides
Every consequential decision in a hiring process should have a human name attached to it. The AI produces evidence; a hiring manager interprets it. We do not auto-reject at any stage.
AI hiring · done right
This page is the playbook we follow at VoxxHire. What to automate, what to keep human, how to stay compliant with EU AI Act and GCC labour laws, and what our AI interviewer actually does under the hood.
We built VoxxHire's AI interviewer against these rules. They are the difference between an AI that helps and an AI that quietly creates a legal and ethical problem for your company.
Every consequential decision in a hiring process should have a human name attached to it. The AI produces evidence; a hiring manager interprets it. We do not auto-reject at any stage.
Candidates get told, before they start, that an AI will conduct the interview and that a human will review the output. They see what competencies are being assessed. They can request a human-only review on reasonable grounds.
The AI asks every candidate for the same role the same core questions, with calibrated follow-ups. Not because it is dogmatic — because consistency is what makes evaluation defensible.
Gender, nationality, age, religion, marital status, disability — none are inputs to the AI's recommendation. The scorecard rates competency and evidence from the candidate's own answers.
Every interview recorded (with consent), every scorecard versioned, every human override logged. Six-year retention by default, exportable on request.
Scheduling takes seconds. Interview takes 20-30 minutes. Scorecard arrives in under 90 seconds. Fast is a feature of candidate experience, not just operational metric.
The useful question is not \"AI vs human\" — it is \"which step of the process is AI better at, and which is a human better at\". Our answer, after 2 million interviews on the platform:
| Automate | Keep human |
|---|---|
| First-round competency screens | Hiring-manager technical deep-dive |
| Scheduling, reminders, follow-ups | Compensation and offer negotiation |
| Scorecard evidence capture | Final hire/no-hire decision |
| Candidate FAQ responses | Culture-fit and manager-fit conversation |
| CV parsing and shortlist suggestions | Reasonable-adjustment calls and accommodations |
| Follow-up emails with structured content | Rejection calls for finalists and senior candidates |
EU AI Act (in force since 2026). AI used in hiring is classified high-risk. You must register the system, document its purpose and training data provenance, maintain human oversight, and show that the system does not disparately impact protected classes. VoxxHire tenants in the EU get the regulatory file pre-populated — you supply role-level rubrics, we supply the system documentation.
UAE Labour Law and DIFC / ADGM employment regulations. No standalone AI-hiring statute yet, but the broader principles apply: candidates cannot be rejected on protected grounds, employers must be able to justify their decision on request, consent must be meaningfully given. We align by default with these principles across all GCC tenants. ADGM published 2026 guidance that treats AI hiring tools substantively similar to EU — expect DIFC to follow in 2027.
KSA Personal Data Protection Law. Candidates have the right to know what personal data is processed, by whom, for how long, and the right to object. Our candidate consent flow discloses all four, and candidates can request export or deletion of their data at any time.
India DPDP Act. Same consent-and-purpose-limitation principles. We default to India-region data residency for Indian candidates, and we do not retain interview recordings beyond the retention period you configure (default: 6 months after decision).
Equal employment and reasonable accommodation. Any candidate can request a human-only alternative on reasonable grounds — disability, specific-language need, or stated discomfort with AI interviews. The alternative is a scheduled recruiter call using the same structured questions. We log the request, the accommodation, and the outcome.
The candidate joins a voice interview from any device. No headset required, no app install. We detect language — English, Arabic, Hindi, Urdu, Tagalog, Bahasa — and conduct the conversation in it.
The questions come from the role-level rubric you built once. For a Senior Software Engineer, that might be: system design (3 questions), practical coding scenarios (2), collaboration and conflict (2), ownership and ambiguity (2). Each question has an acceptable-answer rubric that the AI uses to grade — not a free-text GPT-pleasing output.
The AI asks a follow-up when the first answer is thin, moves on when it has enough signal. The whole interview takes 20-30 minutes. The candidate hears the questions clearly, has 90 seconds to think before answering any question, and can repeat or skip at any point.
Within 90 seconds of the interview ending, the scorecard is in your dashboard. Each competency has a 1-5 score, each score cites the exact quote from the candidate that justifies it. If an evaluator disagrees, they can override — the override is logged, the AI's original rating remains visible, and the comparison feeds our calibration.
We never produce a hire-or-no-hire recommendation. The scorecard tells you where the candidate is strong and weak, with evidence. The decision belongs to your hiring manager, and your audit trail reflects that.
You pay per completed AI interview, not per recruiter seat. One completed interview = one credit. We charge nothing for scheduling, no-shows, or interviews that fail for technical reasons on our side.
Credits do not expire. Buy a pack when the price is right; use it when hiring picks up. We don't believe in seat-based pricing for a tool you don't use every day.
Send us a JD and a rubric. We'll build the AI interview and let you take it yourself before we show it to a candidate.
Book a walkthrough