An AI agent that automates the reference checking process. It calls the provided references, asks a structured set of questions about the candidate, transcribes the conversation, and provides a summarized, unbiased report in the applicant tracking system (ATS).
Reference checks are one of the slowest steps in recruitment. Recruiters play phone tag with busy referees, take notes by hand while trying to listen, and write up summaries afterwards. The questions asked vary between recruiters, making it hard to compare candidates fairly, and a strong candidate can accept another offer while the checks drag on.
An AI reference check agent built with Vapi, speech-to-text and an LLM can call referees at a convenient time, ask a consistent set of structured questions and draft a summary for the recruiter. Results are stored in the ATS, such as Greenhouse or Lever. The summary is only a starting point: it cannot be truly unbiased, because referees' views are subjective, so the recruiter reviews the transcript and makes the judgement.
The candidate consents to reference checks and provides referee details, and the recruiter chooses a structured question set for the role.
Referees receive a message explaining the check and choose a time, then the agent calls, introduces itself as an AI assistant and asks for consent to record.
The agent asks structured questions about the working relationship, strengths, development areas and suitability, with follow-up prompts for vague answers.
The call is transcribed, and an LLM drafts a summary quoting the referee's own words, which is attached to the candidate's ATS record with the transcript.
The recruiter reads the summary and transcript, follows up personally on concerns and decides how to use the reference.
A recruiting agency uses an AI agent to conduct reference checks. The agent calls a candidate's previous manager, records and transcribes the call, and generates a summary highlighting strengths and areas for development, cutting the time-to-hire by three days.
Vapi, Speech-to-Text, LLM for summarization (GPT), ATS Integration (Greenhouse, Lever)
Reference information is personal information under the Privacy Act 1988, and candidates should know how referees will be contacted and how their information is used. Recording calls engages state and territory listening devices laws, so obtain referee consent before recording. Questions must avoid protected attributes under federal and state anti-discrimination law, such as health, pregnancy, age or family responsibilities, and should focus on job-related performance. Referees may make defamatory or inaccurate comments, so store transcripts securely and use them only for the hiring decision. Candidates may request access to information held about them, depending on your circumstances.
This solution requires careful evaluation of cost-benefit ratio. Consider implementation only after thorough analysis and risk assessment.
Many will, especially if contacted in advance and offered a convenient time. Some prefer a human, and they should be able to request one. Explaining up front that the call is structured, short and recorded with their consent helps. Response quality can vary, so recruiters should watch for guarded answers.
Not entirely. Structured questions make checks more consistent, but referees' opinions are subjective and the summary depends on how the model interprets answers. Presenting summaries that quote the referee, reviewing transcripts and avoiding questions about protected attributes reduce bias risk, but a recruiter must still apply judgement.
One to two months is typical. The work includes integrating with your ATS, designing question sets for different role types, setting up consent and recording notices, testing voice quality and summaries with internal staff, and running a pilot with real candidates before broader use.
Get expert guidance on implementing AI Reference Check Agent for your business
Contact Our AI Experts