Industry Analysis
AI in Hiring
RPO Strategy
Everyone Says Companies Are Abandoning AI in Hiring. They're Not.
What's really happening, and what to demand before you hand your hiring to a provider's AI.
Read the headlines, and you'd think corporate America is quietly unplugging AI from hiring and walking away. It isn't. Adoption is still climbing. What companies are actually retreating from is a much shorter list: fully autonomous AI interviews, screening no one can explain, and candidate scores nobody can account for.
The question in the room has changed. It used to be whether to use AI. Now it's how to use it without getting sued, embarrassing your brand, or hiring the wrong people with great confidence. And if you outsource any part of recruiting, that question lands squarely on your RPO provider's AI, not just your own.
"What companies are actually retreating from is a much shorter list: fully autonomous AI interviews, screening no one can explain, and candidate scores nobody can account for."
More AI in the workflow, less AI in the verdict
Strip out the noise, and the shape is clear. AI is spreading fast through the parts of recruiting that make people faster: sourcing, resume screening, candidate matching, scheduling, chatbots, and skills-based talent rediscovery. What's being pulled back is AI making the actual call on its own: automated rejection, AI-generated interview scoring, autonomous ranking, and interviews with no human anywhere in them.
The engine behind the growth is sheer volume. Greenhouse reports that the average job posting now draws about 242 applications, close to triple the 2017 number. No team clears that by hand, so AI has become the triage layer. Add the pressure to lift recruiter productivity and the shift toward skills-based hiring, and demand only rises. The pattern holds across engagements: more AI in the workflow, less AI in the verdict.
242
applications drawn by the average job posting (Greenhouse)
3x
close to triple the 2017 number
Four forces behind the caution
The pullback isn't timidity. Four hard pressures are pushing it.
The legal ground moved.
In Mobley v. Workday, a court let the technology vendor be treated as an agent of the employer and allowed applicant age-discrimination claims to move forward. Around that case sits a thickening set of rules: New York City Local Law 144, Colorado SB 24-205, California's automated-decision regulations under FEHA, and the EU AI Act. A 2026 class action has gone further, arguing that AI hiring scores function as consumer reports under the Fair Credit Reporting Act. In most of these frameworks, the liability lands on the employer even when the tool belongs to someone else.
The research is unsettling.
A Stanford study of 3.4 million applicants described a pattern it calls algorithmic monoculture: when a handful of vendors supply screening to most employers, one tool's bias quietly spreads across the whole market. Other research suggests AI reshapes recruiting workflows more than it lifts the quality of hire, while raising a quieter worry about recruiters losing their edge to automation.
Candidates are voting with their feet.
Greenhouse's data shows that about 63% of United States job seekers have now been interviewed by AI, and about 38% have dropped out of a process because it included one. Yet only about 19% want less AI overall. The objection isn't the technology. It's being processed instead of considered. Get the experience right and candidates stay.
The integrity arms race is on.
Fabric's analysis of nearly 20,000 AI-conducted interviews found more than a third of candidates showing signs of AI-assisted answering, and close to half in technical roles. Detection keeps losing ground to the assistance tools, which is why some employers are quietly bringing interviews back into the room.
63%
of United States job seekers have now been interviewed by AI
38%
have dropped out of a process because it included AI
1/3+
of candidates showing signs of AI-assisted answering
Where that leaves RPO
Not where the headlines imply. The RPO Association's AI Taskforce reports that fewer than 7% of RPO providers are resisting AI, and that clients are pushing for it in about 93% of engagements. The industry's caution is about pace and governance, not refusal. The Taskforce's sharpest line is aimed at buyers: vendor benchmarks often fall apart under scrutiny, and taking a vendor's word for it is no longer a defensible basis for a purchase. The market is moving from AI that replaces recruiters to AI that works as their co-pilot, and the job of vetting that AI now sits with you.
93%
of engagements have clients pushing for AI
<7%
of RPO providers are resisting AI
"The market is moving from AI that replaces recruiters to AI that works as their co-pilot, and the job of vetting that AI now sits with you."
What to demand before you sign
These are the questions that separate a provider with governed AI from one selling AI features with no controls underneath.
Explainability.
The provider should be able to say why the tool scored or ranked a candidate. If it can't, you can't defend that decision to a regulator, a court, or the person you rejected.
A real bias audit.
Independent, done within the past year, tested against the four-fifths rule and scoped to how you'll use the tool, with the methodology open for review rather than a pass badge.
Human oversight that bites.
A trained person with the authority to overrule the system reviews adverse decisions, and that review is documented. Automated rejection with no human in the loop carries the highest legal exposure.
Liability that doesn't all land on you.
Subramanian's RPOA analysis reports that 88% of AI vendors cap their own liability. Read the contract, see how discriminatory-outcome risk is split, and push for shared responsibility and notification terms.
Data you actually own.
The same analysis reports that 63% of companies treat portability as a top concern. Nail down ownership of your candidate data and the intelligence built from it, and the right to take both with you if the relationship ends.
Compliance across borders.
A tool that clears one jurisdiction can fail in the next. For multi-state or multi-country hiring, confirm it meets the strictest standard that applies, including candidate notice and a human alternative.
An experienced candidate accepts.
Notice that AI is in use, a plain explanation of what's measured, a human option, and accommodations. This isn't only ethics. It's your dropout rate and your employer brand.
Evidence, not marketing.
Ask for a reference client and a pilot in your own roles before you commit and treat headline efficiency numbers as claims to be tested.
Integrity that holds up.
Ask how the assessment stays valid when candidates use AI, and what the false-positive rate is on any cheating flags.
Governance you can name.
Find out who owns AI governance on their side, how often they monitor and re-test, and what happens when something goes wrong.
The bottom line
The smart money isn't walking away from AI in hiring. It's getting deliberate: deploying AI as a co-pilot that makes recruiters faster while keeping people in the decision and getting to demand with the partners who run that AI on their behalf. A provider that answers the questions above with specifics is built for where this is heading. One that can't just tell you something valuable before it costs you.
"A provider that answers the questions above with specifics is built for where this is heading. One that can't just tell you something valuable before it costs you."
If you evaluate AI hiring tools, which of these questions has been hardest for a provider to answer? I'd like to hear what you're seeing.
Sources: Greenhouse candidate and application-volume data; Stanford study of algorithmic hiring (3.4 million applicants); Fabric analysis of AI-conducted interviews; RPO Association AI Impact and Innovation Taskforce; R. Subramanian, RPOA analysis of AI in RPO; and employment-law guidance covering Mobley v. Workday, NYC Local Law 144, Colorado SB 24-205, California FEHA automated-decision regulations, the EU AI Act, and the Fair Credit Reporting Act. This is general information, not legal advice.