Ask candidates about AI interviews and you'll hear everything from "I could finally interview after my kids were asleep, in my own language" to "I talked at a camera for a company that never replied." Both are real. The variable isn't the AI — it's the implementation, and employers choose the implementation.
This piece is written for the employers doing the choosing. (We build an AI interview product, so we have a stake; we've also written the candidate-facing version of this page, which is itself one of the standards below.)
Why candidates dislike bad AI interviews — legitimately
The complaints cluster into four, and none of them are irrational. The void: recording answers into a system that produces silence — no outcome, no reasons, sometimes not even a rejection. The black box: knowing a machine scored you but not on what; the fear isn't automation, it's unexplainable judgment. The ambush: discovering mid-application that you're being recorded and analyzed, with consent buried in a checkbox. The gauntlet: hostile framing — surveillance language, no practice, trick timing — that treats applicants as fraud suspects rather than people considering your company.
Notice what's absent from that list: candidates don't primarily object to asynchrony. Answering on your own schedule, without five rounds of calendar tennis, is the part candidates consistently like — especially employed candidates, parents, and anyone in a different timezone. The resentment attaches to disrespect, not to format.
What a respectful AI interview looks like
The implementation choices that separate the two experiences are concrete, and worth demanding from any vendor (including us):
Consent before recording, always. Conditions shown up front, explicit agreement captured, no recording until it's given. In Rubrily's AI interview the interview literally cannot start otherwise — the consent screen is a gate, not a footnote.
Transparency about the format. The invitation shows the role, the duration, and what will happen before the candidate commits anything. Surprise is the cheapest thing to remove from a hiring process.
Practice before grading. A free walkthrough of the flow and interface, so the tool itself is never what's being tested. First-time-with-the-format nerves are noise, not signal, and a fair process filters noise out.
Their language, their schedule. Candidates pick the interview language from the options the employer enabled and take it when they're ready. Both choices move the evaluation toward substance: someone reasoning in their strongest language at their best hour is showing you their actual ability.
Evaluation that actually happens — with reasons. Every answer scored against defined criteria with a written justification, so the human reviewing the report can see why — and so the process behind an outcome is defensible rather than vibes. This is where AI interviews, done properly, beat the status quo they replace: at volume, the alternative to an AI interview was never a warm human conversation. It was silence after a CV skim. An interview where every applicant is heard and scored on the same rubric is more respect, not less.
A human making the call. The AI screens and explains; people decide. Candidates should be told this, because it's true and because it matters to them.
Do candidates actually mind being interviewed by AI?
Research and practitioner surveys consistently find the objection is conditional: candidates broadly accept AI screening when the process is transparent, consistent, and explained — and resent it when it's opaque or one-sided. The strongest predictors of a positive experience are knowing what will happen, getting a fair chance to show ability, and receiving an outcome. All three are implementation choices.
The parts employers control on the integrity side
Monitored interviews (tab-switch detection, camera presence, AI-answer detection) are legitimate at volume — candidates also benefit when cheating doesn't pay. The respectful version states the conditions plainly before consent, frames them as fairness rather than suspicion, and treats flags as context for a human reviewer, never as auto-rejection. The hostile version hides the monitoring and lets a tripwire reject silently. Same technology; opposite experiences.
Why this is self-interested advice for employers
Candidate experience at volume is brand at volume. Every applicant is a potential customer, referrer, reviewer, and future re-applicant; a role with 400 applicants generates 395+ non-hires whose lasting impression of your company is your process. The respectful implementation costs nothing extra — consent screens, practice modes, and explained scores are product features, not staffing — and it compounds: candidates who felt fairly treated reapply, refer, and say so publicly. The gauntlet does the opposite, in public, on review sites. Choose accordingly — and if you're choosing a vendor, put the candidate page of each one side by side and ask which process you'd want your own name in.
FAQ
Should we tell candidates an AI will interview them? Yes — legally required in a growing number of places (Illinois notably), and strategically wise everywhere: surprise is the single biggest driver of AI-interview resentment. Say what the format is, how long it takes, and that a human makes the decision.
Do candidates get their results? At minimum, candidates deserve a timely outcome — silence is the worst-reviewed behavior in all of hiring. Whether to share score detail is the employer's call by policy and jurisdiction; what makes detail shareable at all is explainable scoring, since you can't communicate reasons you don't have.
What about candidates uncomfortable on camera? Practice mode, clear instructions, and self-paced re-recording of the setup (not the answers) remove most of the discomfort; language choice removes more. For accommodation needs, the process should offer a human alternative on request — and the 2025 ACLU complaint over a denied captioning accommodation is the cautionary tale for handling this badly.
An interview process you'd put your own name into: consent first, practice included, every answer scored with reasons, humans deciding. Start free →