What candidates really think about AI interviews, and how to earn their trust

What the surveys say

Four sources, three years apart, tell a consistent story. AI interviews are now common. Most candidates meet them without warning. And very few believe they make hiring fairer.

What sits behind the numbers

Read together, the surveys point to four specific complaints, not a blanket rejection of AI.

• Surprise. In the 2026 Greenhouse survey, 21% learned AI was involved only after the interview had started. Undisclosed AI use was the second most common reason for walking away (27%).
• No one on the other side. The top reason for leaving was a pre-recorded, AI-scored video interview with no human present (33%).
• Silence afterwards. Of candidates who completed an AI interview, 51% never heard back. A screen that takes someone's time and returns nothing feels extractive.
• Doubt about who decides. Pew found Americans oppose AI making the final hiring decision by 71% to 7%.

There is a counterweight. In the same Pew survey, 47% said AI would be better than humans at treating all applicants similarly, against 15% who said it would be worse. In the Greenhouse data, perceived bias was nearly identical for AI and human interviewers. Candidates see the consistency argument. What they reject is being judged by software without knowing it, and without recourse.

An eight-step trust playbook

None of this needs new technology. It needs decisions made before the first candidate is invited.

1. Disclose before it starts, beginning with the job post

Say in the job post that the first screen may be AI-led, and repeat it in the invitation. Disclosure inside the interview is too late; the candidate has already given their time. The law is moving the same way. The EU AI Act's duty to tell people they are interacting with an AI system applies from 2 August 2026, and New York City has said notice must come 10 business days before an automated employment decision tool is used.

2. Explain what is evaluated, and by whom

Name the competencies the screen covers and say who reviews the result. In the 2026 Greenhouse survey, 39% wanted a clear explanation of what the AI measures. In Illinois, before AI analyzes a recorded video interview, employers must explain how the AI works and what general types of characteristics it evaluates, and get consent.

3. Offer a human alternative without penalty

Give a real choice: a call with a recruiter at a time the candidate can book, scored on the same rubric. Say plainly that choosing a person will not affect the outcome. Then make it true, by scoring both lanes against the same competencies and comparing them in the same debrief.

4. Let candidates choose the time

The thing AI interviews do better than people is availability. Let candidates take the screen at any hour, on a phone, within a window of several days rather than a same-day deadline. A candidate with a job and a family should not have to take leave for a first screen.

5. Keep it short

An early screen needs three or four scored questions, not an hour. State the length upfront and keep to it. Anything you want to learn at length belongs in a later round with a person.

6. Do not analyze emotion, tone or faces

Score what the candidate said, against the rubric. Do not score expressions, voice tone, accent or "confidence". They are weak evidence about the job and they alarm candidates. In the EU, using AI to infer emotions in the workplace has been banned since 2 February 2025.

7. Give an answer, and feedback where you can

Tell every candidate the outcome within a stated number of days. Where you can, share a line of feedback tied to the rubric; a transcript and a scorecard make that cheap. Silence is what candidates remember. In a 2024 Greenhouse survey, 61% of job seekers had been ghosted after an interview.

8. Say who decides

Name the role that makes the decision and keep a person accountable for it. In the 2026 Greenhouse survey, 38% wanted a human to review the AI's evaluation before any decision. Make that review the default, and do not let a score reject anyone on its own.

A sample disclosure you can adapt

Put a version of this in the job post and the invitation. Replace the brackets, and delete anything that is not true of your process.

How to tell whether it is working

• Completion rate of the AI screen, compared with the human lane.
• Share choosing a person. A high share means the AI lane needs work, not that candidates are difficult.
• Drop-off between invitation and start: candidates who open the invite and never begin.
• Time from screen to answer. If it runs past a week, the speed advantage is gone.
• Candidate comments after the screen, read by a person every week.

How Itya approaches it

Itya is built on the same rules. Candidates give consent before an interview is recorded, every interview for a job is scored against one rubric, and each scorecard cites the transcript moment behind its ratings. By default, Itya's AI recommends and a named person confirms; it never rejects a candidate automatically. What we tell candidates is on our page for candidates, and our safeguards are on the trust page.