49% of Job Seekers Think Your AI Screening Is Biased. Here’s the Documentation That Answers Them.

Key takeaways:

• Nearly half of employed job seekers (49%) believe AI recruiting tools are more biased than human recruiters, a number the American Staffing Association and The Harris Poll first published in 2023. Fresh 2026 data shows the distrust hasn’t moved: only 21% of candidates think most employers use AI responsibly.
• And it’s not just about candidate perception. NYC’s Local Law 144 names employment agencies directly, and the Mobley v. Workday case is testing whether an AI vendor can be treated as an employer’s agent, which means the liability doesn’t stop at the software company that built the tool.
• Fairness now has to be provable, not just true. That means an independent bias audit, a documented candidate notice process, and a vendor contract that doesn’t let a supplier’s word stand in for evidence.

Almost half of employed job seekers, 49%, believe the AI tools screening them are more biased than a human recruiter would be. The American Staffing Association and The Harris Poll first published that number in September 2023, based on a poll of over 1,200 employed U.S. adults. It’s still the number ASA leads with today, because nobody has re-run the question and found something better. It’s a three-year-old complaint the industry hasn’t yet addressed.

2026 data echoes the sentiment

Greenhouse’s May 2026 survey of 1,200 U.S. workers found that 63% of workers have now been interviewed by AI, up 13 points in six months. Adoption is accelerating. But trust is flat:

• Only 21% of candidates believe most employers are using AI responsibly.
• 70% say they weren’t clearly told upfront that AI would evaluate them.
• 38% have walked away from a hiring process because it included an AI interview. The single biggest reason, cited by 33% of that group, was a pre-recorded video interview scored by AI with no human present.

Only 19% of candidates want less AI in hiring overall. What they’re asking for instead is proof it isn’t cutting corners: a way to request a human review (46%), clear disclosure before the interview starts (44%), and evidence the tool has been audited for bias (29%).

That last one, the audit evidence, is the piece most agencies can’t currently produce.

Why “biased” is now a legal question

The EEOC has been clear since 2023 that an AI tool producing a disparate outcome can violate Title VII even when the tool never references race, age, or gender directly. What’s changed by 2026 is where the liability falls when a candidate acts on that.

Mobley v. Workday has moved past the early motions on the theory that an AI vendor can function as an employer’s agent, not just a tool the employer chose. For a staffing firm, that theory cuts both ways. Your firm is the agent using someone else’s screening tool to fill someone else’s requisition, and several of the newer state statutes were written with exactly that structure in mind.

New York City’s Local Law 144 names “employment agencies” specifically, not just the employers who hire them. If your firm screens or ranks a candidate for a New York City role with an automated tool, the audit and notice obligations rest with you, not just the client. California’s FEHA rules extend liability to anyone “acting on behalf of an employer” for recruitment or screening. Illinois’s amended Human Rights Act (effective January 1, 2026) reaches “agents, including recruiters and other third parties acting on an employer’s behalf.” See our full state-by-state breakdown, including where Colorado pulled back its requirements this year for the jurisdiction-by-jurisdiction detail.

Across every version of this law, a vendor’s assurance that its tool is fair doesn’t hold up as your defense. You need your own paper trail.

What “proof” looks like

Building an independent audit program from scratch takes real legal budget and time, which is exactly why most firms have leaned on a vendor’s word instead of their own paper trail. Strip away the jurisdiction differences, though, and a defensible file comes down to five things:

1. An independent bias audit, done by a third party that isn’t your firm or the vendor, within the past year. This is the one piece 29% of candidates said they specifically want to see evidence of, and it’s the anchor requirement under NYC’s law regardless of where else you place candidates.
2. A public or on-file audit summary, dated, with the data sources and metrics it covered. Not a vendor marketing page. A record you can produce.
3. A documented candidate-notice process, showing what candidates were told about AI use, when, and how they could ask for a human alternative. Disclosure timing and format is where several of the newer state laws differ, but “we never told them” isn’t a defensible position under any of them.
4. A vendor contract that assigns responsibility, spelling out who conducts audits, who cooperates with them, and who owns the data if a claim gets filed. If your contract is silent on this, you’ve agreed to carry risk your vendor created.
5. A human-review log, showing a person actually looked at contested or borderline decisions, not just that a human theoretically could have.

A candidate, a regulator, or a plaintiff’s attorney can all ask the same question: how do you know your screening tool isn’t doing what 49% of job seekers already assume it is? The five items above are the only real answer. A confident LinkedIn post about “responsible AI” is not.

The document trail is becoming the product

AI screening is moving from something agencies quietly use to something candidates now expect to be told about and given a say in. As more of hiring runs through automated tools by default, the firms that can hand a candidate or a client a real audit trail, not a vendor’s word, are the ones proving the “we do this responsibly” claim instead of just making it. That’s a harder thing to build than a faster screening funnel. It’s also a harder thing for a competitor to fake.