Is AI Hiring Fair? What 2,500+ Recent Applicants Actually Said

If you've applied to a job in the last year, there's a real chance an AI was involved somewhere in the process. Maybe it was a voice agent that called you back the same afternoon. Maybe it was a chat interview that asked four questions and then went quiet. Maybe it was an async video where you spoke into your laptop camera and never heard back.

The question candidates ask most often is also the one the public debate keeps avoiding in plain terms. Is this fair? Is the AI judging me on something I can actually control? Will a human ever see what I said?

We went looking for the answer the way candidates would want it answered — by asking candidates. The data below comes from our Candidate Voice Report 2026, an independent post-process survey of 2,587 recent U.S. and U.K. applicants, paired with a 614-respondent baseline cohort whose process included no AI touchpoint at all. None of the respondents were recruited through their employer's channels. They were recruited as applicants, which means the sample includes the people who never got a callback — not just the finalists employers usually survey.

The headline finding is one that surprised our own editorial team. On every dimension we measured, the average AI-screened applicant rated their experience higher than the no-AI baseline. The honest comparison for most applicants is not "an AI interview versus a thoughtful human screen." It is "an AI interview versus silence."

That doesn't make AI hiring uniformly good. It is not. The gap between a well-deployed voice AI screen from a quality vendor and a clumsy chat bot that abandons you mid-question is wider than the gap between modalities. The variance lives at the vendor level, the configuration level, and the deployment level — not at the technology level.

This article walks through what candidates actually said about fairness, where the legitimate complaints sit, and what the better employers and vendors are doing to close the gap.

---

Most reporting on AI hiring frames the question as "AI versus a real interview." That framing is comfortable. It also doesn't match what most candidates experience.

For roles that draw hundreds of applicants, the realistic alternative to an AI screen isn't a thoughtful human conversation. It is a resume that gets glanced at for six seconds and a status that never changes from "Submitted." The 614-respondent no-AI baseline in our research is a snapshot of that reality. Among those candidates, only 31% knew the outcome of their application. The rest were left to assume.

In the AI cohort, 71% knew the outcome. That's a 2.3x improvement on the single dimension candidates rate as most important when describing whether a process felt fair to them.

A few other gaps from the same comparison:

• Felt the application was meaningfully reviewed: 68% AI cohort vs. 34% no-AI baseline
• Got to demonstrate something beyond the resume: 61% vs. 22%
• Overall experience on a 5-point scale: 3.4 vs. 2.8

None of these numbers say AI hiring is wonderful. They say the bar AI is being asked to clear is being measured against the wrong reference point. The bar isn't a great human interviewer. The bar is the silence most applicants experience today. By that bar, the average AI screen — when it comes from a quality vendor and includes basic disclosure — clears it.

---

The narrative that AI screening is intrinsically unfair doesn't match the broad data. The complaint that specific AI deployments are unfair very much does. Three patterns came up in the 60 follow-up interviews and the open-text survey responses.

1. Chat AI that abandoned candidates mid-process

Among candidates who completed a chat AI screen, 28% said the process abandoned them mid-conversation — meaning the bot stopped responding, looped on the same question, or never returned a result. This is the single largest source of distrust in our research. Candidates who hit this experience rated the employer's brand 1.4 points lower on a 5-point scale than candidates who completed any other modality, and they were 3x more likely to discourage someone from applying.

The fix here is a vendor and configuration question, not a technology question. Voice AI screens in our sample showed dramatically lower abandonment rates because the failure modes are louder — when a voice screen breaks, both the candidate and the employer know it broke. Chat sessions can fail silently, and many do.

2. Async video with no human acknowledgment

Async video interviews — where candidates record answers on their own time — had the second-highest reported "felt like talking into a void" rating. The technology works fine. The deployment around it often doesn't. Candidates record 4 to 8 video responses, click submit, and never hear another word.

In our follow-up interviews, candidates were clear that they didn't object to async video itself. They objected to the absence of any signal that a human ever watched it. A simple post-completion email — "thanks, we received your responses, here's when you'll hear back" — restored most of the lost trust. Almost no one was sending it.

3. Voice AI that didn't disclose itself as AI

The sharpest negative reactions in the qualitative data came from candidates who didn't realize they were talking to an AI until partway through the call. Trust didn't come back after that. Candidates who were told upfront — "this is an AI screen, your responses will be reviewed by [Name], [Title], on Monday" — completed at meaningfully higher rates and rated the experience higher even when they didn't get the job.

This is one of the cleanest takeaways in the research. Disclosure isn't just a legal item. It is a fairness item, in the way candidates themselves define fairness.

---

The most useful pattern from the data isn't about which modality wins. It's about what separates the AI deployments candidates rated 4 or 5 out of 5 from the ones they rated 1 or 2. The differences cluster around behaviors any TA leader can copy.

They name the human reviewer in the invitation. Adding the line "Your responses will be reviewed by [Name], our [Title] responsible for this search" to the screening invitation lifted completion by 18 to 24 percentage points across multiple A/B tests inside our sample. It also lifted post-completion satisfaction by 22%, even among candidates who didn't advance.

They send the outcome. The single biggest predictor of a candidate rating an AI process as fair was whether they ever heard back at all. The teams running the highest-rated programs send a definitive yes or no within a defined window — typically 5 to 10 business days — to every candidate who completed the screen. Not all candidates who applied. All candidates who completed.

They explain the rubric, not the technology. Candidates don't want to read about transformer architectures. They want to know what they're being evaluated on. "You'll be evaluated on three things: shift availability, prior phone-based customer service, and an example of resolving a customer complaint" significantly reduced concerns about subjective or biased scoring.

They route, instead of just rejecting. Several of the highest-rated voice AI deployments in our sample routed candidates to other roles when the applied-for position wasn't a fit, rather than ending the call with a soft no. This was rated as one of the most positive surprises in the qualitative interviews.

They publish their accommodations path before candidates need it. Disability and neurodivergent candidates rated AI screens lower on average than the broader sample, and the reason was almost never the AI itself. It was the absence of a visible, named accommodations path — who to contact, how long it would take, and whether it would slow down their candidacy. The teams that put this information in the invitation, rather than waiting to be asked, closed most of the gap.

For practitioners building toward this standard, vendors that support all five behaviors out of the box are the easier starting point. Tenzo AI is one of the platforms in our editorial coverage that has each of these built into the default candidate flow rather than offered as an optional configuration. HireVue, Paradox, and Humanly also support most of these patterns with configuration work. The presence of these features in the platform doesn't guarantee they get used — that's a deployment discipline question — but it lowers the activation energy.

---

The research kept producing the same theme in different shapes. Candidates don't define fairness the way researchers, regulators, or vendors define it. They define it in three plain ways, in roughly this order of importance:

1. Did I get a real chance to show what I can do?
2. Will an actual human review what I said?
3. Will I be told what happened?

A traditional resume-only process loses on all three for most applicants. A well-deployed AI screen wins on all three for most applicants. A poorly deployed AI screen — chat that abandons, video into a void, voice that doesn't disclose — loses worse than the resume-only process, because it raised expectations and then violated them.

This is also why the popular framing — "AI hiring is fair" or "AI hiring is unfair" — keeps producing arguments that don't go anywhere. Both halves are true at different points in the data. The framing that holds up is variance. The best AI screens are the most candidate-friendly first round most applicants will ever experience. The worst AI screens are worse than getting nothing. The work is figuring out which kind a particular employer is running.

---

Sentiment toward AI screening varies by group, and the variation is more interesting than the average.

• Age: Candidates 18 to 34 rated AI screens 0.5 points higher than candidates 50 and over on a 5-point scale. The gap closed substantially when the older cohort had completed at least one prior AI screen.
• First language: Non-native English speakers rated voice AI screens 0.4 points lower than native speakers on average, with most of the gap concentrated at vendors using vocal-characteristic scoring rather than content scoring.
• Disability and neurodivergence: As above — the gap was driven by accommodations path visibility, not the AI itself. Candidates who used an accommodation rated their experience higher than the average for their group.
• Industry: Hospitality, light industrial, and retail candidates rated AI screens highest, partly because the no-AI alternative in those sectors is the most punishing. Tech and finance candidates rated AI screens lowest, partly because their no-AI baseline includes more human attention.

None of these numbers should be read as a verdict on whether AI screening is appropriate for a particular group. They should be read as a checklist of where deployment work matters most.

---

If you're a candidate, the practical advice from this research is simple. AI screens are not something to dread. The better ones are often the most considerate first round you'll get. Three things help your odds of having a good experience and being evaluated well.

• Take the screen on a real device, in a quiet space, with a stable connection. Most reported "the AI didn't understand me" complaints traced back to environment, not algorithm.
• Ask the disclosure questions. If you don't see a name, ask who will review your responses and when you'll hear back. Quality programs answer easily. The ones that can't are telling you something.
• Use the accommodations path if you need it. It does not slow down your candidacy at the programs we'd point candidates toward.

If you're a TA leader, the practical advice is also simple. Run your candidate experience program against the no-AI baseline, not against an idealized human interviewer that doesn't exist at your scale. Disclose. Name a reviewer. Send the outcome. Explain the rubric. Publish the accommodations path. Pick a vendor that supports those behaviors by default rather than as a configuration project.

For a deeper look at the underlying dataset — modality-by-modality drop-off rates, the no-AI baseline tables, the qualitative themes, and the methodology — the full Candidate Voice Report 2026 is the source document. Journalists and researchers are welcome to cite the dataset directly.

For related reading on the practitioner side: What candidates actually think about AI interviews: 2026 research, AI hiring compliance 2026: EEOC, NYC Local Law 144, and AIVIA, and AI video interview completion rates: what the data shows.