Reference checks are the part of hiring that everyone says is important and nobody does well. The pattern is familiar. The hire is about to be made. Someone realizes references have not been called. A recruiter scrambles, calls three references the candidate provided, gets glowing reviews from people the candidate handpicked, and the check is done.
This adds zero signal. The check exists to provide legal cover, not to inform the decision.
AI can improve this on the legitimate side and degrade it on the illegitimate side. Both deserve attention.
What works. AI can analyze the interview transcript and identify two or three areas where the panel had genuine uncertainty. Then it can generate reference questions targeted at exactly those areas. Instead of asking "is the candidate a good performer," the reference is asked "the candidate mentioned leading a migration project, can you describe their actual technical contribution versus their coordination role." The reference is forced to be specific.
What is creepy. Some tools scrape former colleagues from LinkedIn and call them without the candidate's knowledge. This is legal in some jurisdictions and a trust disaster everywhere. Do not do this.
What is illegal. Some jurisdictions prohibit asking about salary history, medical conditions, family status, and various protected categories during reference checks. AI systems that generate reference questions need guardrails for this. Most do not.
Platforms like Mazle generate reference questions from interview evidence and run them through legal filters before they reach the recruiter. The work is targeted. The compliance is automatic.
References done well are a real signal. References done as theater are worse than nothing because they create false confidence.
