New — The State of AI Hiring 2026. Every number sourced→
AI & Technology

Advantages and Disadvantages of AI in Recruitment: What the 2025-26 Evidence Says

The advantages and disadvantages of AI in recruitment with numbers on both sides: field-experiment gains, bias audits, candidate trust and the law.

Card contrasting one advantage, +12% job offers in an AI-led interview field experiment, with one disadvantage, 38% of job seekers withdrawing over an AI interview.

TL;DR: The advantages and disadvantages of AI in recruitment depend on one control: whether a person reviews the evidence before any adverse decision. With humans deciding, AI-led voice interviews raised job offers 12% in one field experiment of 70,884 applications (Jabarian & Henkel, 2025). Yet 38% of job seekers say they have withdrawn from a process because of an AI interview (Greenhouse, n=2,950, 1 May 2026; vendor survey).

General information, not legal advice. Have counsel confirm how this applies to you.

AdvantageDisadvantage
AI-led interviews with recruiters deciding: offers +12% (Jabarian & Henkel)38% of job seekers have withdrawn from a process because of an AI interview (Greenhouse; vendor survey)
Structured interviews have the highest mean operational validity of the common methods, .42; AI can enforce structure (Sackett et al., 2022)Three LLMs ranking resumes favored white-associated names 85% of the time versus 9% for Black-associated names (University of Washington, 2024)
78% of applicants chose the AI recruiter when offered the choice (Jabarian & Henkel)26% of candidates trust AI to evaluate them fairly (Gartner, 31 Jul 2025)
Consistency: the same structure applied to every candidate (a benefit of structure, not proof AI beats a trained interviewer)21% of companies using AI in hiring auto-reject at all stages with no human review (Resume Builder, n=948, Oct 2024; vendor-commissioned)
Recruiters, not the AI, made the hiring decisions in the field experiment (Jabarian & Henkel)70% of job seekers were not clearly told AI would evaluate them (Greenhouse)

Updated 28 October 2026. Evidence current as of 21 October 2026.

Key takeaways

  • In the one large field experiment, AI-led interviews with human decision-making raised job offers 12%. It covered one recruitment firm (RPO) hiring for 43 client accounts, one country and an entry-level customer-service role (Jabarian & Henkel, 2025).
  • Structured interviews have the highest mean operational validity of the common methods (.42, Sackett et al., 2022). That is a benefit of structure, which AI can enforce. It is not proof that AI beats a trained human interviewer.
  • Bias is documented in AI resume ranking: white-associated names favored 85% of the time versus 9% for Black-associated names (University of Washington, 31 Oct 2024). The study is scoped to resume screening, not interviews.
  • 21% of companies using AI in hiring auto-reject at all stages with no human review (Resume Builder, n=948, Oct 2024; vendor-commissioned), and 26% of candidates trust AI to evaluate them fairly (Gartner, 31 Jul 2025).
  • We found no independent time-saving figure. The last section shows how to measure your own.

The pros and cons of AI in recruitment depend on when each item applies, and the evidence is sharper than a tidy list. In one large field experiment, AI-led interviews with humans deciding raised job offers by 12% (Jabarian & Henkel). In one large survey, 38% of job seekers said they had withdrawn from a process because of an AI interview (Greenhouse, 2026).

Both numbers are sourced, and both are narrow. The drawbacks are design-dependent, and the controls that address them are what define evidence-based AI interviews. This page gives you each number with its sample and scope, and the one control that decides which column you end up in.

Is AI in recruitment good or bad? It depends on one control

It depends on whether a person reviews the evidence before any adverse decision. Where that control exists, the strongest data we have is favourable. Where it doesn't, the drawbacks below stop being hypothetical. Where that review sits at each step of a hiring process is laid out in a stage-by-stage approval-gate map.

Decision diagram: if a person reviews evidence before any adverse decision, results are favourable +12% offers; if not, 21% auto-reject with no human review. Below, the four conditions under which the

The table below puts every benefit and drawback on one footing: the number, the sample, and the conditions under which it holds or fails. The last two columns are my reading of each study's scope.

ClaimNumberSample and sourceHolds whenFails when
AI-led interviews improve outcomesOffers +12% vs human-led; higher starts and retention, no decline in hired-worker productivity (per RePEc listing)70,884 applications, 48 entry-level customer-service postings, one RPO firm hiring for 43 client accounts, Philippines (Jabarian & Henkel, 2025)High-volume entry-level screening with human decisionsNot shown for senior roles; not shown beyond one RPO firm and one country
Applicants accept AI interviewers78% chose the AI recruiterSame experiment, same caveats (Jabarian & Henkel)Candidates are offered a choiceCandidates aren't told or can't opt out (see the Greenhouse rows below)
Structure predicts performanceMean operational validity .42Sackett, Zhang, Berry & Lievens, J. Applied Psychology 2022 (paper)The interview is structuredThe interview isn't structured. Also says nothing about AI vs a trained human
Adoption pays off39% have implemented AI in HR; 56% don't formally measure success; 88% of HR leaders report no significant business value yetSHRM State of AI in HR 2026, n=1,900+, 3 Apr 2026 (SHRM); Gartner, 28 Oct 2025 (release)Outcomes are measured against a baselineNobody measures. Gartner's figure covers HR AI tools in general, not recruiting alone
AI resume ranking carries biasWhite-associated names favored 85% vs 9%; male-associated 52% vs 11%3 LLMs, 550+ resumes, 500+ listings, 9 occupations, 3M+ comparisons, 120 names (University of Washington, 2024)An LLM ranks resumes without auditNot tested on interviews
Rejection happens without a person21% auto-reject at all stages; 50% at resume stage only; 29% keep human oversight of all rejections948 business leaders, 22 Oct 2024 (Resume Builder; vendor-commissioned, 2024 data)The system rejects on its ownA person reviews evidence before each adverse decision
Candidates withdraw or distrust38% withdrew; 70% not clearly told; 51% of completers never heard back; 26% trust AI to evaluate them fairlyGreenhouse, n=2,950, 1 May 2026 (report; vendor's own survey); Gartner, 31 Jul 2025 (release)Candidates aren't told and no human is presentDisclosure and a human-review route are in place
Compliance gap57% of HR professionals in regulated states unaware of applicable AI laws; 12% of aware professionals took compliance stepsSHRM 2026, n=1,900+ (SHRM)The team doesn't know which laws applyThe team has mapped applicable law and acted on it

Print the scope next to every figure: job role, country, what was measured. Strip those and a number starts saying things the study never did.

The decision rule fits in one sentence: the benefits hold when a person reviews evidence before any adverse decision. The final section turns that into a pilot plan.

Advantage: can AI improve hiring outcomes?

The strongest evidence for AI in recruitment is a natural field experiment, not a vendor case study. Jabarian and Henkel's paper, "Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews" (SSRN 5395709, August 2025), covers 70,884 applications to 48 entry-level customer-service postings handled by PSG Global Solutions, a recruitment process outsourcing firm, for 43 client accounts in the Philippines.

Relative to human-led interviews, AI-led interviews produced 12% more job offers. According to the paper's RePEc listing, they also came with higher job starts and retention, and no decline in the productivity of hired workers. We cite those further outcomes without figures, because secondary summaries give different numbers for retention. Recruiters, not the AI, made the hiring decisions.

An AI-led interview here means a voice interview conducted by AI rather than a human recruiter, followed by a human decision. The mechanics are covered in how AI interviews work.

The caveats belong in the same breath as the result. This is one recruitment firm (RPO) hiring for 43 client accounts, in one country, hiring for a high-volume entry-level role. The study doesn't show the same gains for senior roles, and it shouldn't be quoted as if it did.

Candidates also took to it in this setting. When applicants were offered a choice, 78% chose the AI recruiter (Jabarian & Henkel, same experiment, same caveats).

A good result in one setting says nothing about bias in another, which is the next question.

Disadvantage: can AI make bias worse?

It can, and the evidence is specific. In October 2024, University of Washington researchers reported a study of three large language models ranking more than 550 resumes against more than 500 job listings across nine occupations, over three million comparisons in all (University of Washington, 31 October 2024; AAAI/ACM AIES).

The models favored white-associated names 85% of the time against 9% for Black-associated names. They favored male-associated names 52% of the time against 11% for female-associated names. Black male names were never ranked above white male names.

The scope matters as much as the result. The study tested LLM resume ranking with 120 names. It did not test interviews, and this page doesn't stretch it to cover them.

Two phrases stay off this page: "AI eliminates bias" and "bias-free." A structured process is consistent, which is a property it can demonstrate. Unbiased is a claim it would have to prove. Where bias claims reach a court, the tracker of AI hiring cases records the filings and rulings.

Advantage: is structure the real benefit?

Mostly, yes. Sackett, Zhang, Berry and Lievens (J. Applied Psychology, 2022) put the mean operational validity of structured interviews at .42, the highest among the common selection methods (paper).

Read that carefully. It is a benefit of structure, and an AI interviewer can enforce structure on every candidate. It is not evidence that AI performs better than a trained human using a structured format. The research favors the format. The Society for Industrial and Organizational Psychology's write-up of the paper says "structured interviews may in fact be the strongest predictor of job performance—not cognitive ability" (SIOP).

A fair comparison also looks at the human baseline. In our reading, human-only processes carry their own bias and their own inconsistency, so the right question is whether a tool beats a well-run structured human process, not whether it beats zero. On the machine side, Reuters reported in October 2018 that Amazon had scrapped an experimental recruiting tool that showed bias against women (Reuters).

Time savings are often promised, so we looked for an independent, primary-source figure for hours saved per hire. We did not find one. Treat any such number you meet as unsourced until it names its study. If you want a number you can trust, measure your own, and the final section shows how.

Disadvantage: what happens when no person reviews the decision?

This is where design decides the outcome. A Resume Builder and Pollfish survey of 948 business leaders, published 22 October 2024 and fielded 9 October 2024, found 21% of companies using AI in hiring auto-reject at all stages without human review, 50% at the resume stage only, and 29% keep human oversight of all rejections. 33% said AI "always" or "often" produces biased recommendations (Resume Builder). The survey was vendor-commissioned and it is 2024 data, so weigh it accordingly.

Whether this is a disadvantage of AI or of a particular configuration depends on which controls the system enforces. The practical test is a short scorecard: can a person see the evidence behind each recommendation, and must a person approve before a candidate is rejected?

There is also an explainability gap. Rejected applicants want substantive feedback, not generic replies (Olckers, Vidler & Walsh, 2022; survey of recent job applicants, sample not stated in the abstract). Showing the reasoning behind a decision is what explainable AI in hiring is about.

Advantage and disadvantage: what do candidates think?

Both sides show up in the data, and they come from different studies with different people, so they can't be netted against each other.

For: 78% of applicants chose the AI recruiter when offered the choice (Jabarian & Henkel).

Against: the Greenhouse 2026 Candidate AI Interview Report (n=2,950, 1 May 2026; the vendor's own survey) found that 38% of job seekers have withdrawn from a process because of an AI interview, 70% were not clearly told AI would evaluate them, and 51% of those who completed one never heard back. The top walk-away triggers were AI-scored recorded video with no human present (33%), no disclosure (27%) and AI monitoring (26%) (Greenhouse). Separately, 26% of candidates trust AI to evaluate them fairly (Gartner, 31 July 2025).

Read walk-away as a funnel and a trust signal, not as candidate behavior to police. People leave processes that don't tell them what's happening. The full set of candidate-trust figures, each with sample and date, sits in the AI recruitment statistics.

The compliance gap is the first problem. In SHRM's State of AI in HR 2026 (1,900+ respondents), 57% of HR professionals in regulated states were unaware of the AI laws that apply to them, and only 12% of those who were aware had taken compliance steps (SHRM).

In the EU, Annex III of the AI Act lists "Employment, workers' management and access to self-employment" among its high-risk areas (point 4, Regulation (EU) 2024/1689), which is the category recruitment tools fall under. The AI Act entered into force on 1 August 2024. Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI, entered into force on 27 July 2026 and moved the application of the high-risk rules for Annex III systems from 2 August 2026 to 2 December 2027 (EUR-Lex). Article 5(1)(f) of the AI Act prohibits using AI systems to infer emotions of a natural person in the areas of workplace and education institutions, except for medical or safety reasons. The text says "workplace" and does not mention recruitment explicitly, so how far it reaches interview analysis is a matter of interpretation. The full set of duties is in the guide to EU AI Act obligations for recruitment.

In the United States, the court-record fact most often cited is Mobley v. Workday (N.D. Cal.). On 16 May 2025 the court granted preliminary certification of an age-discrimination (ADEA) collective action. The case concerns the claim that a software vendor can be treated as the employer's agent. That claim is an allegation, not a finding.

As of 19 October 2026, the CourtListener docket showed the case still moving. On 12 July 2024 the court granted in part and denied in part Workday's motion to dismiss (Dkt 80). A Fourth Amended Complaint was filed on 10 July 2026 (Dkt 390), and the plaintiffs' motion to certify a class was filed on 14 September 2026 (Dkt 418), with a hearing set for 9 March 2027. None of that is a finding on the merits.

Outside the courts, enforcement is also an open question. New York City regulates automated employment decision tools under Local Law 144. In an audit issued on 2 December 2025, the New York State Comptroller wrote that the city's enforcement agency "surveyed websites and bias audits of 32 companies and identified just a single issue of non-compliance," while the auditors "reviewed the same companies and identified at least 17 instances of potential non-compliance" (NY State Comptroller).

Legal exposure is a drawback that depends on jurisdiction and on what you can show a regulator or court about how a decision was made.

When do the benefits hold? A decision rule and a pilot plan

The benefits hold when four things are true:

  1. A person reviews the evidence before any adverse decision.
  2. Scoring is structured and traceable to the candidate's answer.
  3. Candidates are told that AI is part of the process.
  4. Selection rates are monitored.

Adoption without payoff is the base rate. 39% of organizations have implemented AI in HR, with recruiting the top use at 27%, and 56% do not formally measure success (SHRM State of AI in HR 2026, 1,900+, 3 April 2026). In a separate Gartner survey, 88% of HR leaders said their organizations had not realized significant business value from AI tools (Gartner, 28 October 2025). That one covers HR AI tools generally, not recruiting only.

The fix is boring and effective: measure before you scale. Run a pilot on one role family and log the following.

MeasureWhat to logWhy it matters
Time-to-shortlistDays from application close to a reviewed shortlist, pilot vs current processThe only time-saving figure you can trust is your own
Completion rateCandidates who finish the AI step / candidates invitedThe first sign of friction
Withdrawal rateCandidates who drop out after the AI step is announced or startedThe funnel cost behind the Greenhouse figures
Adverse-impact ratiosSelection rate by group at each stageWhere bias would show up in your own data
Human override rateHow often a reviewer changes the AI's recommendationShows whether the review is real or a rubber stamp
Rejection-without-review rateShare of rejections no person looked atShould be zero in a design that follows the decision rule

Then fill in the net-effect worksheet. Inputs: current screening hours per role, applicants per role, rejection-without-review rate, withdrawal rate. Outputs, after the pilot: hours moved from screening to review, and a plain two-column result of where the AI helped (time-to-shortlist, consistency) and where it added risk (withdrawals, overrides, impact ratios). If the pilot shows no gain on your own numbers, you have your answer, and it cost you one role family.

Frequently asked questions

Does AI reduce bias in hiring?

Not by default. Bias is documented in AI resume ranking (University of Washington, 2024; scoped to LLM resume ranking). What structure offers is consistency, and consistency is what lets you measure outcomes. For how measurement works, see what a bias audit measures.

It depends on the jurisdiction. EU law treats recruitment AI as high-risk, with obligations from 2 December 2027 (EUR-Lex). The Digital Omnibus on AI (Regulation (EU) 2026/1744) moved the Annex III application date from 2 August 2026 (EUR-Lex). Rules differ by jurisdiction within the US as well, and New York City's Local Law 144 is the best-known local rule. A state-by-state view is in where AI hiring is regulated.

What are the risks of AI in recruitment?

The documented ones are bias in resume ranking, rejection with no human review, candidate walk-away, legal exposure and the explainability gap. The conditions table above shows each one with its number, sample and the conditions under which it applies.

Does AI save time in hiring?

We found no independent, primary-source figure for hours saved per hire. The honest answer is to measure your own: log time-to-shortlist in a pilot, as in the plan above.

Does AI hurt candidate experience?

It depends on how the process is run. In one field experiment 78% of applicants chose the AI recruiter (Jabarian & Henkel), while in a separate survey 38% of job seekers had withdrawn from a process because of an AI interview (Greenhouse). Those are different studies with different populations. Disclosure and a human-review route are the two controls the candidate data points to.

Should small companies use AI in recruitment?

Only if a person can review the evidence before any adverse decision and the team can measure the result. Company size isn't the test. The decision rule is.

Does AI replace recruiters?

In the strongest field evidence, recruiters made the final hiring decisions (Jabarian & Henkel). We have no source showing replacement.

Who is liable if an AI tool discriminates?

It depends on the facts and the jurisdiction. Mobley v. Workday (N.D. Cal.) is the court-record case on whether a vendor can be treated as the employer's agent; the claims there are allegations, not findings. As of 19 October 2026 the docket showed a Fourth Amended Complaint (10 July 2026) and a class-certification motion filed on 14 September 2026. The filings are summarized in the Workday AI hiring lawsuit guide.

See the evidence behind a score

See an evidence pack in a demo and read the evidence behind the score. In ZeaHire, candidates wait in Pending Review until a person approves or rejects them.

AI interviewsHiring technology
Was this useful?

Comments

No comments yet — start the conversation.

Leave a comment

Your email is never published. Comments appear once a moderator approves them.

Replying to
Plain text only. Be kind and stay on topic.

By commenting you agree to our privacy policy.