AI-driven reference checking promises a faster process, more objective reports and less manual work. But does it hold up in practice? We walk through what actually gets automated, how much time you save — and the pitfalls you need to know about.
What does "AI reference check" actually mean?
In practice the term covers 3-4 things:
- Automated outreach with structured questionnaires.
- AI-generated reports that summarise answers per competency.
- AI voice agent that calls the referee and transcribes the conversation.
- Sentiment and pattern analysis across multiple referees.
How much time do you save in practice?
A manual reference check typically takes 30-60 minutes per referee. With three referees per candidate and 5 hires per month that's 7-15 hours of HR time. With AI tools that drops to 30-60 minutes total across all 5 hires — your time goes into setup and reading the report, not the data collection itself.
See also our comparison of digital vs manual reference checking.
Does the quality hold up?
When the AI summarises structured answers against defined competencies, the reports are often more consistent than a stressed HR colleague's notes from three different phone calls. You lose the option of live follow-ups — but most tools let you send a follow-up if an answer was unclear.
More objectivity — or new biases?
Structured digital questionnaires reduce classic biases like halo effect and recency bias. But AI models can introduce new biases if trained on biased data. Pick tools that are transparent about which model they use and that don't make automated decisions (decision support only) — that's also a requirement under GDPR Article 22.
AI voice agent: hype or useful?
An AI voice agent (e.g. Bland.ai) calls the referee, runs a structured conversation in Norwegian and delivers a transcript and report. Pros:
- The referee can answer in a busy day without blocking out time.
- Voice often produces more nuanced answers than a written form.
- The transcript provides full traceability.
Cons:
- Some referees prefer humans — always offer an alternative.
- Requires good audio and a quiet room — can be challenging on a poor line.
Pitfalls to watch for
- Black-box reports: Require the AI to show which answers underpin each conclusion.
- Automated decisions: AI should give you decision support, not make the decision for you (GDPR art. 22).
- Data storage outside the EEA: Many AI services send data to the US — check where your tool stores data.
- Forgetting the duty to inform: The referee must know that an AI is processing their answer.
When does it pay off?
Rule of thumb: if you do 2+ reference checks per month, an AI tool saves you more than it costs. If you do 5+, the saving is so large that a manual process is actually a competitive disadvantage — you lose candidates because the process takes too long.
