Hiring with AI: keep a person in charge of every decision
AI can help a hiring team organize information, draft communications, or surface questions. A person still makes the final call on who gets an opportunity.
A human-led process starts with job-related evidence, consistent evaluation, and a clear owner for the final decision. It also makes room for accommodations, transparency, privacy, vendor oversight, and escalation when a tool behaves unexpectedly.
Use this as an operating framework and confirm legal requirements with your counsel. The Equal Employment Opportunity Commission states that federal anti-discrimination laws apply to AI and other technologies used in employment decisions, including recruiting, screening, hiring, monitoring, promotion, and termination. It also notes that seemingly neutral practices can create unlawful disparate impact. [1]
Start by defining the decision
Before choosing a tool, write down the hiring decision it will support:
What evidence is relevant to the role?
Which step is the tool assisting?
Which decisions stay with a person?
Who reviews the output?
What is the escalation path?
For example, an AI tool may help summarize structured interview notes after an interviewer has recorded evidence. Any ranking it suggests should rest on criteria a person can explain.
The more consequential the decision, the stronger the case for human review, documented criteria, and a way to challenge an output.
Use job-related evidence
A practical scorecard might separate:
Dimension | Evidence | Human check |
|---|---|---|
Role capability | Work sample or structured question | Does the evidence match the stated job requirement? |
Communication | Written or live response tied to the work | Is the assessment accessible and consistently scored? |
Judgment | Scenario with tradeoffs | Did the candidate explain assumptions and limits? |
Collaboration | Behavioral example with follow-up | Are interviewers evaluating the behavior itself? |
Learning | Example of adapting to a new workflow | Is adaptability defined for this role? |
AI can help format notes or identify missing evidence. The interviewer remains responsible for whether the evidence is valid, sufficient, and relevant.
Tie every decision to the evidence itself. A polished summary can still omit context, misunderstand an answer, or amplify a biased note.
Make human review a real control
“Human in the loop” works when the reviewer has the time, authority, and information to disagree.
Set a review rule:
The interviewer records observations against the scorecard.
The tool produces a draft summary or comparison, if used.
A named reviewer checks the source notes and corrects errors.
The decision-maker records the reasons for the recommendation.
Ambiguous or adverse outputs go to escalation for a human decision.
The reviewer should be able to see what information the tool used and what it did not use. Keep the original evidence available for the retention period required by the company’s process and applicable rules.
Plan for accommodations and accessibility
Candidates may have disabilities or communication differences that affect how an automated assessment interprets speech, facial movement, timing, or other signals. The EEOC’s factsheet gives examples of technology creating problems when speech patterns or facial recognition are treated as evidence of ability.
Give candidates a clear way to request an accommodation and a human contact who can respond. Build trust in a tool through testing. Check whether the assessment measures the job-related skill it claims to measure, and provide an alternative when the method itself creates a barrier.
Be transparent with candidates and protect vendor details
Candidates should understand when technology is part of an assessment or workflow and how to raise a question. The exact notice depends on the tool, role, location, and company policy, so use approved language.
Internally, document:
The vendor and version or service used
The purpose of the tool
The data it receives
The output it produces
The person who reviews it
How errors and accommodation requests are handled
Ask vendors for documentation about intended use, limitations, data handling, security, testing, and incident response. Your team keeps accountability for the hiring decision, whatever the contract says.
Use a lightweight risk-management loop
NIST describes its AI Risk Management Framework as voluntary guidance for incorporating trustworthiness considerations into the design, development, use, and evaluation of AI systems. Its structure is useful for an internal hiring process: govern the use, map the context, measure performance and risks, and manage issues over time. [2]
For a hiring workflow, that can mean:
Govern: name an owner, set allowed uses, and train reviewers.
Map: identify candidates, decisions, data, affected groups, and failure modes.
Measure: test consistency, accuracy, accessibility, and error patterns on an ongoing basis.
Manage: pause the tool, correct the process, notify stakeholders, or switch methods when risk is too high.
Keep the loop proportional. An automated ranking tool carries far more decision risk than a small drafting assistant.
Build an escalation path
Write down what happens when:
The tool produces an obviously wrong summary
Two tools disagree
An interviewer cannot explain a score
A candidate requests an accommodation
A candidate challenges an assessment
A vendor reports a security or model change
The hiring team suspects a pattern of disparate impact
The first response should be to pause the affected decision, preserve the relevant record, and send the issue to the designated people, legal, compliance, or HR owner. Investigate with people and independent evidence, outside the model that produced the result.
FAQs
Can AI screen resumes?
It can assist with administrative organization when the criteria are job-related, documented, and reviewed. Treat any resume score as a draft that needs testing, oversight, and a way to challenge errors.
Should candidates be told about AI use?
Use transparent, approved language appropriate to the workflow and location. Candidates should know when technology materially affects an assessment and how to ask questions or request an accommodation.
Is NIST AI RMF a legal requirement?
NIST describes the AI RMF as voluntary guidance. It can structure risk management alongside legal review and applicable employment obligations.
Who owns the final hiring decision?
A named human decision-maker who can review the underlying evidence, explain the reasoning, and escalate uncertainty. Accountability stays with that person.
The practical takeaway
Use AI to reduce administrative drag and keep judgment visible. Define the job-related evidence, give humans authority to disagree, support accommodations, oversee vendors, and pause when the process creates unexplained risk.
How Crossbridge keeps people at the center of hiring
Our talent team personally interviews every candidate before you see a profile, so a person who understands your role makes every recommendation. You get 3 to 5 vetted candidates per search and a 60-day replacement guarantee, backed by a 95% lifetime retention rate.
Get candidates for your open role or see how Recruiting as a Service works.