Responsible AI in hiring: what we use, what we don't
Our operating rules for AI assistance across sourcing, screening and coordination.
The AI-in-hiring conversation has become polarised in a way that is not helpful to anyone. On one side, vendors selling autonomous screening agents. On the other, blanket policies banning any AI use in the recruitment workflow. Neither is a serious position. We have spent two years developing an operating model that treats AI as a tool with specific jobs, boundaries, and audit trails.
Where we use AI
- Sourcing signal enrichment — surfacing public information about candidates we would have found manually, faster. Never as the sole basis for outreach.
- Draft screening notes — AI-generated summaries of interview transcripts that a human recruiter reviews and rewrites before the candidate ever appears in a shortlist.
- Coordination — scheduling, reminders, and status updates. The parts of the job humans do worst and machines do best.
- Market benchmarking — pattern detection across compensation and movement data, always audited by a researcher before it reaches a client.
Where we do not use AI
- Never for shortlisting. The decision to include or exclude a candidate is made by a human recruiter who can be named in the shortlist document.
- Never for reference calls. The signal in a reference is in the pauses and the specifics; both are invisible to an AI.
- Never for candidate rejection communication. If we cannot spare the two minutes to write a real message, we do not deserve the trust for the next mandate.
- Never for scoring candidates against protected characteristics, directly or indirectly. This is a bright line that requires actual audit, not just intent.
The audit trail
Every AI-assisted step in our workflow is logged with the model used, the prompt, the human reviewer, and the outcome. A client can request the audit trail for any mandate at any time. We have been asked for it exactly three times in two years, but the existence of the audit is what makes the rest of the policy real.
"Responsible AI in hiring is not a philosophical position. It is an operating manual, and if you cannot write it down in one page, you are probably not being responsible."
