AI Hiring Audit Checklist: 20 Checks Before and After You Deploy a Tool

Audit an AI hiring tool in four parts: know what you use, test how it behaves, check the legal and notice duties and keep monitoring after launch. The checklist below has twenty checks, grouped by stage. It supports compliance work but does not replace legal advice or an independent audit where a law requires one. Status as of October 2026; rules in this area are changing, so confirm current requirements for the places where you hire. Keep a record of each answer, the date and who reviewed it. A tool you cannot explain or test is a tool you should not use for decisions about people.

Use the list for a new purchase and again each year.

A. Inventory and purpose (4 checks)

1. List every tool that scores, ranks, filters or recommends candidates: resume screeners, assessments, chat screeners, video analysis, scheduling tools that filter.
2. For each, write its purpose, the decision it informs and who uses the output.
3. Record which roles and locations it covers. Rules depend on where candidates live and where you hire.
4. Note whether any tool analyses faces, voices or emotions. Some places restrict or ban that. See AI hiring compliance in 2026.

B. Criteria and data (4 checks)

1. Confirm the criteria are tied to the job and written down. See screening criteria scorecard.
2. Identify the inputs. Does the tool use anything that can proxy for a protected characteristic, such as name, address, school, gaps or graduation year? See AI bias in hiring.
3. Ask where any training or reference data comes from, and how the vendor handles it.
4. Check how resumes are read. Parsing errors can disadvantage candidates. See resume parsing explained.

C. Human oversight (3 checks)

1. Define what a person reviews and when. Is the output advisory or does it filter automatically?
2. Make sure the reviewer can see the reasoning or evidence, as well as a number.
3. Provide a route for a candidate to ask for human review or an alternative. Check which laws require it.

D. Outcomes and fairness testing (4 checks)

1. Calculate selection rates and impact ratios by group at each stage where lawful data exists. See the four-fifths rule.
2. Use enough data. Small samples need care and may need significance tests.
3. If a ratio is low, investigate causes and consider less discriminatory alternatives.
4. Check for different treatment: test pairs of resumes that differ only in a protected characteristic or a proxy, and compare outputs.

E. Notices and legal duties (3 checks)

1. Identify the laws that apply by location. Examples include New York City's Local Law 144, Illinois's amendment to its Human Rights Act on AI in employment effective 1 January 2026 and Colorado's amended AI law, reported to take effect on 1 January 2027, plus data protection and the EU AI Act where relevant. Confirm current text before relying on a summary.
2. Prepare candidate notices and disclosures. See candidate AI disclosure.
3. Check data protection duties: lawful basis, retention, access and deletion rights, and rules on automated decisions. See GDPR and recruitment data and candidate data retention.

F. Records and monitoring (2 checks)

1. Keep records: audit results, test data, notices, decisions and who approved them. Some laws specify retention periods.
2. Set a schedule for re-audit and for reviewing outcomes, and a trigger for earlier review, such as a model update or a change in role.

Questions for a vendor

Be wary of any vendor that claims a tool is "bias-free" or "compliant" in general terms. Compliance depends on how an employer uses a tool.

Documenting the audit

Record for each tool:

• The date, the reviewer and the version.
• The purpose and the decision it informs.
• The findings and the actions.
• The date of the next review.

Keep it short and consistent. A one-page record per tool is more likely to be kept up to date than a long report.

Responsibilities

Name an owner for each tool in HR or recruiting, and involve legal and data protection colleagues. Hiring managers should know what the tool does and does not do. See structured interviews for how to keep the human part of the process consistent.

How Resume World approaches this

Resume World scores applicants against criteria you define, shows the evidence behind each score, keeps the decision with a person and retains decision records, which supports checks 5, 9, 10 and 19. Auditing outcomes, publishing results where required and giving notices are employer tasks. See compliant AI hiring for the product description.