How to Ace a Live AI Interviewer — Even When It Glitches

Kendiana Colin was applying for a part-time front desk job at StretchLab, a fitness studio near Ohio State, when an AI named Alex started interviewing her.

Moments in, Alex began glitching — repeating the phrase "Vertical bar Pilates" on loop, unable to stop or move on. Colin, sitting in her dorm room, did what felt natural: she stared back, uncertain whether to interrupt, wait it out, or close the tab. "It sounded like she was laughing," she said afterward.

She posted the clip. It has since cleared 3.2 million views.

The story spread for an obvious reason: everyone interviewing right now felt the same low-grade dread when they watched it. What do you do in that moment? What if the next interview you get is the same — a bot, a glitch, and no one from the company picking up the phone?

These aren't hypotheticals anymore. They're the interview process at thousands of companies in 2026.

How Widespread Live AI Interviewers Actually Are

The numbers are striking even if the experience of sitting across from an AI is not.

According to employer branding analytics published in mid-2026, 87 percent of companies now use AI somewhere in their recruitment process. Nearly two-thirds of active job seekers — 63 percent — say they've been through an AI interview in the past six months, per the Greenhouse 2026 Candidate Experience survey. First-round interviews have gone bot-first at companies across sectors: retail, healthcare, tech, professional services.

The main platform behind the viral StretchLab incident is Apriora, a San Francisco startup that provides "AI recruiter" agents — conversational avatars that conduct live, two-way video interviews autonomously. The candidate talks; the AI listens, responds, probes, and scores. No human is on the call. After the session, a transcript and evaluation get routed to a recruiter for review.

Apriora's CEO Aaron Wang acknowledged the glitch to Bloomberg: "We can't get it right every time." He's right that edge cases are inevitable. What he didn't mention is that candidates often have no idea they're about to interview with a bot when they click the link in their invite email.

The Trust Crisis

There is a specific quality to the discomfort job seekers have with AI interviewers that goes beyond the glitches. The glitches are just the most visible symptom of a deeper problem: no one asked candidates whether they wanted this.

In June 2026, Harvard Business Review published a piece titled "AI Has Broken Hiring. Here's How to Fix It." The thesis: AI has made the hiring process faster for companies and more dehumanizing for candidates, and the asymmetry is becoming a problem for both sides.

Only 26 percent of job seekers say they trust AI to evaluate them fairly, according to the same mid-2026 employer analytics data. Sixty-seven percent say they feel uneasy about AI-led hiring systems. AFL-CIO president Liz Shuler called the practice "unacceptable," stating that "AI systems rob workers of opportunities they're qualified for based on criteria as arbitrary as names, zip codes, or even how often they smile."

The legal environment is moving in parallel. Colorado's AI Act, enacted in 2026, requires developers and users of AI hiring tools to take reasonable care to prevent algorithmic discrimination. HireVue — the largest player in AI-assisted interviewing — is currently under investigation for potential Fair Credit Reporting Act violations tied to how its scoring outputs are used as screening inputs.

None of this is going to stop companies from deploying AI interviewers before your next application cycle. But knowing the landscape going in changes how you approach the session.

What a Live AI Interviewer Actually Does

Live AI interviewers like Apriora are not the same as one-way video platforms. The differences matter for how you prepare.

One-way video (HireVue, Spark Hire) gives you a question, 30–60 seconds to think, and records your answer. You talk at a camera. The AI scores your video afterward. You're not in a conversation; you're delivering a monologue into a recording interface.

Live AI interviewers run a real-time two-way conversation. The AI "listens," processes what you say, and responds — probing, pivoting, asking follow-up questions. The experience looks like a video call with a real person except the face on the other side is rendered, not human.

What these systems evaluate depends on the platform, but common scoring signals include:

• Answer completeness: Did you actually address the question, or did you drift?
• Structure: STAR format (Situation, Task, Action, Result) scores better than stream-of-consciousness narration
• Follow-up responsiveness: Whether you can add depth when the AI asks "can you tell me more about that?"
• Keyword relevance: Terms associated with the role often appear in the question bank and the scoring rubric
• Pacing: Speaking too fast or too slow can register as signals the model was trained to flag

The follow-up question is where unprepared candidates get caught. A question like "Tell me about a time you handled conflict on a team" feels manageable with a rehearsed answer. "What would you have done differently in that situation?" — asked immediately after, based on what you just said — requires you to have actually thought through the experience, or to be able to improvise credibly from it.

When It Glitches — And What to Do

The most psychologically destabilizing part of the Colin video is not the repetition itself. It's watching her try to decide whether to say something, wait, or leave.

That uncertainty is what costs candidates. Not the glitch — the freeze.

If your AI interviewer malfunctions:

Don't wait it out. Five to ten seconds of looping is unlikely to self-correct. Say something.

Interrupt cleanly. "I think there was a technical issue — can you repeat the question?" is enough. You're not being rude. The AI does not have feelings. Speaking up is better than seven seconds of silence on the recording.

Restart if offered. Most platforms provide a way to reconnect. Use it. Session transcripts are usually wiped on restart, so you get a clean start.

Email the recruiter afterward. Companies running AI interviews have a human somewhere in the process. If the session ends without resolution, find the recruiter who sent the invite and write: "I experienced a technical issue during the AI interview today — happy to reschedule or speak with someone at your convenience." You cannot be eliminated for the platform's malfunction, and most recruiters will accommodate the request.

The candidates who handle glitches well have already internalized something important: an AI interview is a different kind of interaction than a human interview. The same composure you'd want during a technical problem in a live coding round — don't panic, keep thinking, keep talking — applies here.

How to Prepare Before You Click "Start"

The practical preparation for a live AI interviewer overlaps with, but is not identical to, preparing for a human first round.

Answer in structured format. STAR answers aren't just helpful for human interviewers who know to look for them — they match the completeness signals AI scoring systems reward. Practice until structure is automatic, not something you consciously impose under pressure.

Prepare for follow-ups on every story. For each experience you plan to describe, think through what you'd have done differently, what was hardest, and what you learned. These probe questions are standard across platforms and they're the hardest to improvise.

Do a tech check. AI interviewer platforms are more sensitive to lighting and audio than a standard Zoom call because the system is processing visual and audio signals alongside your words. Test your setup in advance. A well-lit, quiet room with a stable connection removes variables you don't want the scoring system penalizing you for.

Read the invite email carefully. Many companies now disclose that the first round is AI-conducted. Some don't. If the email is vague, search the company name plus "Apriora," "HireVue," or "AI recruiter" — LinkedIn threads and Glassdoor reviews often surface the format within weeks of a company rolling it out.

Ask questions at the end if the platform allows it. Some AI interviewers — including Apriora — give candidates a question window at the close of the session. Have one ready. Treating the AI interview as a real interview, rather than a checkbox to get through, registers differently in the transcript than disengaged one-word answers.

Staying Composed When Your Brain Is Watching for Exits

The specific anxiety of a live AI interview — the uncanny valley of a rendered face, the uncertainty of what it's scoring, the possibility that it breaks and no one is coming to fix it — is real. It's also manageable if you've named it before you go in.

High-stakes moments tend to empty the mind right when you need it most. A follow-up you didn't anticipate, a glitch that breaks your rhythm, a question you've prepared for but can't locate on demand — any of these can produce the same freeze that makes prepared candidates underperform.

Meeting Copilot's interview assistant is designed specifically for this: a real-time overlay visible only to you, not captured by screen share, that surfaces talking points and structure as you speak. It doesn't answer for you. It keeps you on track when your brain does what brains do under pressure.

The Bigger Picture

The StretchLab video hit 3.2 million views not because an AI glitched, but because it confirmed something millions of people already suspected: the hiring process has changed in ways that weren't announced to candidates.

Daniel Chait, CEO of Greenhouse, described what both sides are experiencing as an "AI doom loop" — a cycle in which companies use AI to filter faster, candidates use AI to apply at higher volume, and the signal-to-noise ratio degrades for everyone. Recruiters report worse applicant quality. Candidates report more opaque and dehumanizing processes. Neither side trusts the other.

You can be frustrated by that. The frustration is legitimate.

But the people who get the offer don't refuse to engage with the new system. They walk in understanding it — what the AI is scoring, what happens when it breaks, and how to stay composed anyway. That composure is not natural for most people.

It's practiced.