Are AI interviews fair? A plain-English guide to the rules and what to ask a vendor

Every hiring team that considers an AI interviewer asks the same two questions, usually in this order: is it fair? and are we allowed? Both are good questions. This is a plain-English guide to the second, written by a vendor, so read it as a map of the territory and the questions to ask — not as legal advice. Your counsel knows your jurisdictions; we know how the tools work.

Start with what the tool actually does

"AI in hiring" covers very different things, and the rules treat them differently. A tool that ranks applicants from résumés without a human looking at them is one thing. A tool that conducts a structured interview, records it, and produces an assessment that a human reads before deciding is another. The distinction that matters in most of the rules below is whether the tool makes or substantially replaces the decision, or informs a decision a person makes.

SealHire is built on the second model, deliberately. The interview is structured — every candidate for a role is asked about the same skills, at the same level, from the same brief. The output is a report: a verdict per skill with the transcript quotes behind it, linked to the recording, plus a recommendation with its reasons. A human reads it and decides. Nothing in the report is hidden, and nothing in it is final.

That design is not only about fairness; it is about being able to answer the questions below.

The rules employers ask about most

New York City's Local Law 144. In force since 2023, it covers "automated employment decision tools" used to substantially assist or replace discretionary decisions for candidates in the city. Employers using such a tool need an independent bias audit, a public summary of the results, and notice to candidates in advance. Whether a given tool is in scope depends on how it is used in the decision — which is why "does the AI decide, or does a person?" is the first question to settle with your counsel.

The EEOC. The federal position is that existing anti-discrimination law applies to AI tools exactly as it applies to any other selection procedure: an employer is responsible for the outcomes of the tools it uses, including tools built by vendors. The practical consequence is that you should be able to explain what a tool measures, show that it measures job-related things, and monitor its outcomes.

Illinois and Maryland. Illinois's Artificial Intelligence Video Interview Act requires notice, an explanation of how the AI works, and consent before AI analyses a video interview, plus limits on sharing and a deletion right. Maryland restricts facial-recognition use in interviews without consent. Both come down to telling candidates what is happening and getting their agreement.

Colorado, and the wider trend. Colorado's AI law treats employment decisions as a high-risk use and puts duties on both developers and deployers. Other states are drafting along similar lines, and the European Union's AI Act classifies employment AI as high-risk. The direction is consistent: transparency, human oversight, documentation, and the ability to explain a decision.

What SealHire does, on its side. Candidates see a consent step before the interview that says it is recorded and that AI produces the assessment. Face detection is used to observe presence and attention during the interview; there is no facial recognition and nobody is identified from their face. Integrity observations are notes with timestamps for a human, never automatic decisions. The full detail is in our privacy policy and the business impact and trust page.

Eight questions to ask any AI interview vendor

1. Does the tool decide, or inform? If it screens people out with no human involved, you are in the most regulated category. If a person reads its output and decides, you are in a different one — but only if that is actually how you use it.
2. What does it measure, and where did the criteria come from? The answer should be the job. On SealHire the criteria are the role's brief — the skills the job needs — not personality proxies or facial expressions.
3. Can we see the evidence behind any score? A score you cannot inspect is a score you cannot defend. Every verdict in a SealHire report links to the quotes and the moments in the recording that produced it.
4. Is every candidate asked the same things? Structure is what makes an interview comparable. Ad-hoc conversations, human or AI, are where inconsistency creeps in.
5. What is the candidate told, and when? Notice and consent should be in the candidate's path before anything is recorded, in language they can understand.
6. Does it analyse faces, voices or emotions? Attention and presence observations are one thing; inferring emotion or identity from a face is another, and several rules treat it as a bright line.
7. What happens to the data, for how long, and who can see it? Retention, access and deletion should be written down and match what the candidate was told.
8. Can we monitor outcomes? You need to be able to look at who is progressing and who is not, across the tool's use, and act on what you find. A vendor that cannot give you that view is asking you to take fairness on faith.

How this looks in practice — one team's setup

A mid-sized company hiring across New York, Illinois and Colorado wanted to use SealHire for first-round interviews. Their counsel's questions were the eight above, in a different order, and the setup that came out of it was simple enough to write on one page:

• The decision stays human. The report's recommendation is an input; a named recruiter reads the evidence and decides who progresses, and that is written into the hiring procedure.
• Notice before the link. The interview invitation says that the interview is recorded, that an AI interviewer conducts it and produces an assessment, and that a person makes the decision. Candidates consent on the page before anything records.
• The same brief for everyone. Every candidate for a role is interviewed against the same brief at the same level, and the brief is saved with the role so it can be produced later.
• Outcomes are reviewed quarterly. The team looks at who progressed from the AI interview stage and who did not, across roles, and keeps the notes.
• Retention matches the policy. Recordings are kept for the period the candidate was told, and the deletion path is documented.

None of that required anything unusual from the tool. It required knowing what the tool measures, being able to show the evidence, and keeping a person in the decision — which is what the rules are asking for.

The rules set a floor. The reason to run structured, evidence-backed interviews is above it: they are fairer in practice than the thing they replace.

Fairness beyond compliance

The rules set a floor. The reason structured, evidence-backed interviews are worth doing is above it: they are fairer in practice than the thing they replace. An unstructured phone screen rewards rapport, shared background, and the interviewer's mood at four o'clock. A structured interview asks everyone about the same skills and writes down what they said. The human who decides still brings judgment — that is the point of keeping a human there — but they bring it to the same evidence for every candidate.

That is also why the report is designed to be argued with. If a hiring manager reads the reasoning and disagrees, the disagreement is visible and specific, and it can be acted on. A black box cannot offer that; it can only ask to be trusted.

What to do next

Settle the "decides or informs" question with your counsel for each jurisdiction you hire in. Write down what candidates will be told and when. Then evaluate tools against the eight questions above — including ours. If you want to see what a report looks like before any of that, the structured interview questions page shows how a brief becomes an interview, and a demo walks through a real one.