The Line That Matters
The clearest sentence anyone has written about AI in hiring is this: AI is a tool. The interviewer is the decision-maker. The line is simple to state and easy to lose track of when you are looking at a dashboard with a 92/100 next to a candidate's name. The dashboard is not your hiring decision. It is one of several inputs into a decision that a person, not a model, has to make.
This piece is about why the line matters and how to keep it visible in day-to-day practice.
What AI Does Well
Three things, mostly.
Transcription. Modern speech-to-text in 40+ languages, fast, high accuracy. Saves the reviewer the cost of replaying a sentence to catch what was said. Lets you search across responses for keywords. Keeps a written record that can be audited later.
Structured scoring. Given a clear rubric and a candidate's answer, an AI can produce a score and a written rationale that is consistent across candidates. The benefit is not that the AI is smarter than the recruiter, it is that it is more consistent. Every candidate gets the same evaluation against the same criteria. Schmidt & Hunter (1998) documented the case for structured interviewing being 2x more predictive of job performance than unstructured. Consistency, not cleverness, is where AI helps.
Summarisation. Three sentences capturing the essence of a 12-minute interview. Saves the reviewer time scanning what to focus on, lets the hiring manager get up to speed before a live round, and helps multiple reviewers calibrate.
That is the useful list. It is genuinely useful. interviewstream (2025) reports video review is 6x faster than phone, and AI is part of why.
What AI Does Not Do
Several things, and they are the things that actually decide a hire.
It does not judge fit. Whether someone will work well with your specific team, in your specific culture, on your specific problem set, is not a thing a model can score. It is a judgement call that requires knowing the team, the culture, and the problem set. Recruiters and hiring managers know those. Models do not.
It does not weigh trade-offs. A candidate may be weaker on technical depth but stronger on communication and ownership. Whether that combination fits a particular role is a strategic call. The AI score is one number; the strategic call is a decision about what your team needs right now.
It does not predict five years out. What we can score from an interview is current capability and signal at this moment. Whether that signal becomes seniority, leadership, and impact over five years is a developmental question that no interview, AI-assisted or not, can fully answer.
It does not replace empathy. When a candidate stumbles, gets nervous, or has an off day, a human reviewer can read the moment and decide whether to give weight to the stumble or to the recovery. An AI sees the words and the score. The candidate's lived reality is not in the transcript.
Why The Law Already Made This Decision
The EU AI Act, in force from August 2026, classifies AI used in employment decisions as high-risk. The practical consequence is that any AI scoring system used in hiring needs human oversight, transparency about what the AI does, documented decision-making, and the ability for candidates to contest decisions. NYC Local Law 144 requires bias audits of automated hiring tools. Colorado's AI Act and Illinois's AIVIA add similar requirements at state level in the US.
The legal language is technical, but the spirit is simple. Regulators figured out before most teams did that automated decision-making in hiring produces bias at scale, and bias at scale is illegal in ways that bias by an individual recruiter is not. The fix that the law mandates is keeping humans in the loop.
StormInterview is built around this from day one. The AI scores are advisory. The recruiter and hiring manager see them, weigh them, and make the call. The decision is logged, attributable to a person, and contestable. There is no automated rejection. There is no algorithmic offer. There is structure, supported by AI, and a human at the desk.
How To Tell If You Are On The Right Side Of The Line
Three diagnostic questions.
1. If the AI score and your judgement disagree, who wins? If the answer is "we go with the AI", you have crossed the line. The AI is now the decision-maker. The right answer is "the human looks at the disagreement, understands why, and decides".
2. Can you explain the decision in plain language? Hiring decisions need to be explainable. "The candidate scored 87 on AI" is not an explanation. "The candidate showed strong examples of customer ownership and a clear approach to escalation, but did not articulate the trade-off in question 3" is. Explanations come from humans engaging with the content, not from a number alone.
3. Could you defend this decision in an audit? Imagine a candidate asks why they were not advanced. Could you point to specific reasons grounded in their actual answers? If the only answer is "the algorithm said so", you have a compliance problem and a hiring quality problem at the same time.
The Pattern That Works
Teams that use AI well in interviews share a working pattern. They use the AI score as a screening filter, not a decision. They never advance or reject solely on the AI score. They have at least two human reviewers per candidate at the decision stage. They use the AI summary as a starting point for the live conversation, not a substitute for it. They calibrate the rubric and the AI together, periodically checking that the AI's scoring matches what the team actually values.
The result is faster screening with no loss of decision quality, and an audit trail that holds up. That is the value AI brings when it stays in its lane. iCIMS (2025) data on candidate dropout suggests speed in early stages prevents 60% candidate abandonment, and AI-assisted review delivers that speed without removing the human.
What This Looks Like In StormInterview
The interface puts the AI score next to the recording, with the written reasoning visible. The reviewer can agree, disagree, or override. Notes go on the record. Disagreements between AI and human reviewers are flagged for a second look, not silently resolved. The decision page logs who said what and when, so any future review of the hire can trace the path.
None of this is fancy. It is the result of a design decision: AI is a tool, the interviewer is the decision-maker, and the platform makes both visible. Start a free trial of StormInterview and run a single role through it to see the line in practice. The AI will save your team hours. The decision will still be yours, in the way it should be.