The short answer
Human in the loop (HITL) is a way of designing an AI system so that a person reviews, approves or corrects its output at set points before the result takes effect. In hiring, human in the loop means software may find, rank, message or interview candidates, but a named person checks the work and owns any decision that changes a candidate's outcome.
Human in the loop explained
Oversight comes in levels, and most hiring teams mix them by task. The question for each action is how much harm a mistake could do and how easily it can be undone.
An example set of rules for an AI recruiting agent: suggested sourcing lists can be generated freely; first outreach messages need approval until the template has been reviewed, then send automatically; rejections after an interview always need a recruiter's sign-off; offers and compensation always go to a person. The riskier and less reversible the action, the closer the human sits to it.
NTRVSTA's agent, Carolina, is built around this kind of setting: each action can be set to suggest, ask first or auto, and offers and compensation changes always wait for a person. Our guide to AI recruiting guardrails covers how to set those levels for each step.
Common mistakes
- Rubber-stamping. People tend to accept what a machine suggests, especially when busy. If a reviewer approves hundreds of items an hour, the loop exists only on paper.
- Reviewing only rejections. Check who the AI advances too. A skewed shortlist is as much of a problem as a skewed rejection pile.
- No record of who approved what. If you cannot show which person signed off on a decision, accountability is unclear when a candidate asks.
- A reviewer without the facts. The person needs the reasons behind the AI's output, such as the score breakdown or why a resume ranked high, not just the result.
How to measure it
Override rate = AI outputs a reviewer changed or rejected / AI outputs reviewed. A rate near zero for months can mean the AI is excellent or that nobody is really looking, so pair it with a periodic spot check against an independent second review. Bias checks such as a bias audit are a separate, more formal layer of oversight.
Related terms
- Agentic AI: software that takes steps toward a goal, which is why oversight levels matter.
- AI recruiter: an agent doing recruiting work under a team's rules.
- Bias audit: an independent test of whether a tool's outcomes differ by group.
- Automated employment decision tool: NYC's term for software that scores or ranks candidates.
- Interview calibration: aligning human reviewers so their judgments are consistent.
Questions
What is the difference between human in the loop and human on the loop?
With human in the loop, a person approves each output before it takes effect. With human on the loop, the AI acts by default and a person monitors the results and can intervene or change the rules.
Does human in the loop remove bias from AI hiring tools?
Not on its own. A reviewer can catch individual errors, but people also carry bias and tend to accept machine suggestions. Pair human review with structured criteria, spot checks and a formal bias audit where one applies.
Which hiring steps should always have a human in the loop?
Offers, compensation and any final decision to reject or hire should always involve a person. Many teams also require approval for first outreach templates and for rejections after an interview.
Want to see per-action approval settings on a live hiring workflow? Book a 20-minute demo.
