We had assumptions about how candidates would react to async AI-assisted interviews before we had any real data. Some of those assumptions were right. Several were wrong. A few things we didn't anticipate at all.
This post covers what we heard from the first cohort of applicants who went through the Intervieux flow at early customer sites, the specific feedback that surprised us, and the changes we made in response. We are not claiming this is a representative sample or a formal study. It is what we learned from talking to real candidates after they went through the process.
What We Expected and Got
We expected candidates to appreciate the scheduling flexibility. That expectation was correct. The most consistent positive feedback across the cohort was some version of "I could do it when it worked for me." Candidates with day jobs appreciated not having to take a call during business hours. Candidates with caregiving responsibilities appreciated not having to find a specific window in their schedule for a 20-minute call. The format's time-flexibility was genuinely valued and came up unprompted in a majority of the feedback.
We also expected some candidates to feel uncertain about being evaluated by an automated system rather than a person. That expectation was also correct, but the way candidates expressed it was different from how we anticipated it. Most didn't object to the automation itself. What some candidates said was that they weren't sure how to calibrate their answer length and detail level without the real-time cues that a live conversation provides. "I didn't know if I was giving too much detail or not enough" was a common variant of this concern.
The fix here was partly product and partly communication. We added explicit guidance in the question framing about expected response depth, and we made the progress indicators during the interview clearer so candidates could see they were on track without needing to guess. Both changes reduced the "calibration uncertainty" complaints in subsequent batches.
What We Didn't Expect: The "Thoughtfulness" Feedback
A number of candidates said, without prompting, that the async format gave them more time to think than a live phone screen would have. This was the feedback we found most interesting.
In live phone screens, there is social pressure to respond quickly. Pauses feel awkward. A candidate who thinks carefully before answering may come across as slow or uncertain, even if their eventual answer is better than a candidate who answers quickly with a surface response. In an async format, the time pressure of the live conversation is removed, and some candidates said they felt their answers in the async interview were more accurate representations of how they actually think than their answers in live phone screens typically are.
We are not drawing a strong causal conclusion from this. Candidates who chose to provide feedback may have been disproportionately those who had positive experiences. But the pattern was consistent enough across sites and roles that we take it as a real signal. For candidates who think carefully before speaking, and whose roles do not require rapid improvisation, the async format may genuinely be a better assessment context.
The Negative Feedback We Got Right
Some candidates disliked the format. We want to be honest about what they said.
A recurring theme was that async interviews feel less human than a phone call. Some candidates said they found the experience cold or impersonal. Several said they wanted to know whether a real person was reading their answers, and whether they would receive feedback. The "I feel like I'm shouting into a void" comment, or something like it, came up at multiple sites.
This feedback led to two changes. First, we recommended that customers send a personalized follow-up message within 48 hours of interview completion, regardless of outcome. Not a form rejection, but a short note confirming that responses had been reviewed. The volume concern is real: a team hiring 400 candidates cannot write 400 personalized rejection emails. But even a brief, individually addressed message acknowledging the candidate's specific completion performed substantially better on candidate satisfaction than no message or a generic automated status update.
Second, we added language in the invitation email making clear that hiring team members read the scored summaries and that the AI layer produces a shortlist for human review, not an automated pass-fail decision. This framing change mattered to candidates. The perception that "no human will ever see my answers" was both inaccurate and harmful to the candidate experience. Correcting it upfront reduced the "shouting into a void" feedback significantly.
The Accessibility Questions
A small but notable segment of the feedback raised accessibility-related concerns. Candidates who use screen readers asked whether the interface was compatible with their tools. Candidates with dyslexia noted that the time pressure of a recorded video response format created challenges that were different from what they would face in a text-based or live format.
These questions landed on us as product requirements, not edge cases to dismiss. An interview process that is inaccessible to candidates with disabilities is a legal concern and an ethical one, independent of whether the format is "better" on average. We prioritized screen reader compatibility as a result of this feedback. The text and voice response options in the current format partly address the format-preference concern; candidates can choose the response mode that works best for them rather than being forced into video response.
We are still working on this. Accessibility is not a problem you solve once. The feedback we got was useful because it came from real candidates in real processes, not from theoretical audit. We expect to keep hearing about edge cases we haven't anticipated, and we expect those edge cases to keep improving the product.
What We Have Not Yet Solved
Some candidates, particularly those with significant past experience, felt that the async question format didn't give them sufficient opportunity to demonstrate depth. A five-question structured interview with a word or time limit per response is a constrained format by design. That constraint is what makes it scalable. But it means a candidate with 12 years of relevant experience may answer the same questions as a candidate with 2 years, and the format doesn't automatically surface the difference in depth that a longer, more exploratory conversation might.
This is a real limitation, and we are honest about it. The async structured interview is a first-round filter, not a complete evaluation. Its job is to identify which candidates in a large pool warrant a deeper conversation, not to fully evaluate every candidate's capability. For experienced candidates who are underrepresented in the shortlist relative to their actual qualifications, the answer is a better-designed rubric that rewards depth of behavioral evidence, not a longer first-round format. But we are still learning exactly how to operationalize that recommendation across diverse roles and experience profiles.
See also: Async vs. Live Phone Screens: What Changes and What Stays the Same and Fair Screening at Scale: What Consistency Actually Buys You.