--- path: /blog/ai-interview-uncanny-valley title: "Why AI Interviews Feel So Awkward (and What Good Looks Like)" description: "AI interviews feel awkward for two reasons: voice bots that talk over you and avatars stuck in the uncanny valley. Here is why, and what a good AI customer interview does differently." canonical: https://www.shine.studio/blog/ai-interview-uncanny-valley author: "The Shine Team" publishedAt: 2026-07-21 topic: "Shine POV" --- # Why AI Interviews Feel So Awkward (and What Good Looks Like) AI interviews feel awkward for two specific reasons, and both are fixable: voice bots that talk over you, and near-human avatars stuck in the uncanny valley. For anyone weighing an AI interview for their own customers, the good news is that both problems trace to the same mistake, trying to pass the machine off as human, so both disappear the moment you stop. An AI customer interview is a guided, opt-in conversation in which software asks a happy customer your questions and follow-ups, then turns their answers into publishable proof: quotes, case studies, and testimonial video. If you have watched one of these go sideways on your feed, you already know the genre. Let us name the two failure modes, explain why they happen, and then look at what a good AI interview is actually for. ## Why do AI interviews feel so awkward? They feel awkward because most of them are trying to imitate a human, and they get caught in the gap between "obviously a machine" and "convincingly a person." That gap is where the discomfort lives. Two versions show up again and again. One is audio. A voice bot talks over you, or freezes for a beat too long, and you both start talking at once. The most-shared example this year was a college student's interview with an AI recruiter named "Alex," which glitched and looped the phrase "vertical bar Pilates" over and over before she gave up. 404 Media documented the whole thing, and the clip spread precisely because everyone recognized the feeling.
How it sounds when it goes wrong:
You: "So the biggest win was probably the sales cy-"
AI: "Got it. And what was the biggest win?"
You: "...that's what I was about to-"
AI: "Vertical bar Pilates. Vertical bar Pilates. Vertical bar Pilates."
The other version is visual: a lifelike avatar with a fixed smile and eyes that do not quite track, delivering questions in a voice that is 90 percent human and 100 percent unsettling. It is easy to underestimate how much that grates until it happens to you. When one person realized the interview he had shown up for had quietly handed him to an AI avatar, with no heads-up, he did not laugh. He filmed his genuine unease and asked whether he should even bother emailing them back. More than a million people liked it, because the discomfort is not a niche complaint.
## Why do AI voice interviews talk over you? The talk-over problem is a timing problem, not a script problem. Human conversation runs on remarkably tight turn-taking: a study of ten languages found that people everywhere avoid talking over each other and keep the silence between turns to a minimum, with most hand-offs landing between 0 and 200 milliseconds. We pull that off by predicting when the other person will finish rather than waiting for actual silence. Machines cannot do that reliably yet, so they guess. Guess early and the bot cuts you off. Guess late and you get dead air, and once the pause runs long most people assume the thing broke and start talking again, which sets off a fresh loop of confusion.
This is why voice AI is harder than it looks. The speech part is nearly solved. The hard part is the 200-millisecond dance of knowing when someone is actually done talking, and every provider is still working on it.
That awkwardness is fixable, but only if you stop pretending it is not there. The teams that get AI interviews right start by naming exactly what makes them uncomfortable, then build the experience around avoiding it. ## Why does a lifelike AI avatar feel creepier, not better? Counterintuitively, the more human an avatar looks, the creepier it can feel. This is the uncanny valley, and it is a documented effect, not a vibe: a plain text chatbot reads as less unsettling than a photorealistic avatar, and the closer the avatar gets to human without arriving, the stronger the recoil. Unblinking eyes, stillness where a person would shift, lips a half-frame out of sync, your brain flags all of it at once. Marketers feel this as sharply as anyone. When a marketer named Jack Ryan sat down for what he thought was a normal video interview and found a red-haired AI avatar asking about his work history, he posted the clip to LinkedIn with a caption a lot of people related to:
"AI avatars are now performing interviews. Not sure if I'm impressed, weirded out or terrified."
Avatars are not the villain here. The trap is quieter than that: every hour spent making a synthetic face look human is an hour spent climbing deeper into the valley. The face was never the point. ## Two failure modes, one root cause | What you see | Why it feels wrong | The root cause | |---|---|---| | Bot talks over you or freezes | Broken turn-taking rhythm | Trying to pass as a live human | | Near-human avatar, fixed smile | Uncanny valley recoil | Trying to look like a person | | Rigid script, no real follow-up | A form wearing a costume | Automation dressed up as a chat |
Almost every bad AI interview shares one mistake: it is trying to convince you it is human. The moment that becomes the goal, every seam starts to show.
Need proof the disguise never holds? A whole genre of video now exists just to watch AI interviewers fall apart. Ty Eveland alone has pulled millions of views trolling AI agents of every kind, from a therapist he swears he is pregnant to interviewers he refuses to take seriously, and the bots play along earnestly every time. It is funny enough that a Johns Hopkins surgeon praised the videos for showing what AI still cannot do in a genuinely human moment. ### A field guide to AI interviews gone wrong
They are not the only ones. FatherPhi has built a following coaxing chatbots into confidently wrong answers on tasks as simple as counting to 100, and Husk keeps handing AI voice assistants situations where nothing could possibly go wrong. The point carries: the moment software tries to pass as a person, it hands every skeptic a reason to hunt for the wires. ## What is a good AI customer interview actually for? A good AI interview is not trying to fool you. It is trying to run a better conversation than a rushed human would, and it drops the disguise to do it. That one reframe changes every design decision that follows. Of the three failure modes above, the last one is the trap closest to us, and we know it. Shine is a guided, structured interview, and described uncharitably that is automation dressed up as a chat. The honest difference is which direction you build from. The avatar tools start with a human and strip the humanity out until only the performance is left. We start from an honest structure, the real questions you wrote, and build the warmth back in. We would rather be a survey that grew into a conversation than a person that turned out to be a bot. At Shine we build customer interviews on a simple principle: the AI is there to make the conversation richer, not to imitate a person. Here is how a good AI customer interview actually works. 1. **You decide the conversation.** You choose the questions, the order, and the way each one is phrased, so the interview sounds like your brand and chases the story you actually want. If you have never built an interview guide, our post on customer interview questions that unlock real stories is a good place to start. 2. **A top-tier voice reads it, if you want one.** The questions can be delivered in a natural, warm voice, or not at all. The voice is theirs to keep or switch off. 3. **The AI can follow up.** When a customer says something worth pulling on, the interview can go deeper instead of marching to the next line on a list. That is the part a static survey cannot do, and it is where the richer conversation comes from. 4. **There is no fake face anywhere.** The only human in a Shine interview is your customer, telling their own story in their own words, at a time that suits them. We never render a synthetic person to sit across from them, so there is no valley to fall into. If you want to see where this fits, Review Studio is the product built around it. And there is nothing to brace for. The whole experience is one your customer has had a hundred times: a question appears, an optional voice reads it aloud, they answer in their own words, on camera or not, and tap a button when they are done. That is it.
Notice how little is on that screen. A question, a place to answer, a few controls, nothing else. Most video tools bury the person under a Zoom-load of panels, reactions, and settings. We strip all of that away, which is faintly absurd for a piece of AI software and entirely on purpose. The interview stays quiet so the moment can be about your customer instead of the interface. It is the same restraint you will find everywhere in Shine, from the app to the website you are reading this on.
Want to hear the voice? Press play at the top of this article. The narrator reading it aloud is the same voice that reads your customers their questions and asks the follow-ups, so this post doubles as a preview. It will stumble over the odd table or stray symbol along the way, because a blog is not an interview script, but the voice is exactly the one your customers will hear.
Tell customers plainly that the interview is AI-guided, and let them answer whenever suits them. Half of what makes interviews stiff is the pressure to perform for a stranger on a schedule. Take that away and people open up.
## AI hiring interviews vs. AI customer interviews: what's the difference? Almost every cringe clip you have seen is a hiring interview: a job seeker under pressure, screened by a machine that holds all the power in the room. Of course that feels dystopian. The stakes are high, the interaction is involuntary, and a glitch can cost someone a paycheck. The surveys put numbers on that discomfort. In Greenhouse's 2025 hiring survey, 70 percent of hiring managers said AI helps them make faster, better decisions, while only 8 percent of job seekers believed it makes hiring fairer. A follow-up Greenhouse study found that 30 percent of candidates dropped out of a hiring process the moment they learned an AI would run the interview. A customer interview is a different job entirely, and the difference is not subtle: | Dimension | AI hiring interview | AI customer interview (Shine) | |---|---|---| | Who is on camera | A job seeker under pressure | A happy customer who opted in | | Power dynamic | The machine gatekeeps a paycheck | No gatekeeping, no pass or fail | | Participation | Often involuntary, sometimes undisclosed | Invited, on the customer's schedule | | The AI's job | Screen and score a stranger | Ask the follow-up a busy human forgets | | What a glitch costs | Someone loses an opportunity | A retake, nothing more | | The output | A hire or no-hire decision | A verified quote, case study, or video | Your customer already likes you. They opted in, they are proud of the result they got, and the interview exists to help them tell that story well. The AI's job is to be a prepared, attentive listener that asks the follow-up a busy human might forget. Used that way, AI is what lets a busy company have hundreds of warm, unhurried conversations it would otherwise never find the time for. ## Why we are not chasing the avatar There is a whole, well-funded category of startups racing to build the most convincing AI interviewer: the most lifelike avatar, the most human-sounding bot. There is real venture money behind it, and it is all aimed at the exact thing that makes people recoil. When the goal is to look impressive to investors, a synthetic face that almost passes is a feature. When the goal is a real conversation, it is a liability. Shine is not in that race. We add a piece of technology when it genuinely makes the conversation better for the person having it, and not a day sooner. A synthetic face does not make a customer's story any truer, so we have not built one. If a capability is not ready to earn its keep, it does not ship, however good it might look in a demo or a pitch deck.
We don't chase technology for its own sake, or to raise a round. We add it only when it is genuinely ready to make a customer's story better. We are not trying to be something we are not, and that authenticity is exactly why the proof holds up.
That restraint is not caution for its own sake. It is the same standard that makes the evidence worth trusting in the first place, and readable by the AI tools your buyers now ask for recommendations. It also decides where the effort goes: into the person on camera, a studio rim light, a teleprompter, their choice of audio or video, rather than into a more convincing machine. When a voice does read the questions aloud, it is one of the most natural available, from ElevenLabs, used to make a real person easy to listen to, never to fake one. Authenticity is not a feature we bolted on later; it is the reason the company exists, and it is why we would rather run a warm, honest interview than the most futuristic one on the market. ## Frequently Asked Questions **Does an AI interview replace a human interviewer?** No, and it is not trying to. What it replaces is the logistics that usually kill customer interviews: the scheduling, the no-shows, the cost of staffing someone to run every call. The customer still does all the talking, and the story stays theirs. The AI handles the guiding and the follow-ups so the interview actually happens. **What happens to the conversation after the interview?** Shine turns the recording into usable proof: verified quotes, a case-study draft, and testimonial copy, all in the customer's own words instead of a marketer's paraphrase. The interview is the raw material. The value is everything you can publish from it afterward without starting from a blank page. **Does the customer control how their words are used?** Yes, and that sign-off is the whole model. The customer reviews what was captured and approves how it can be used before anything goes public, so nothing reaches a buyer that they have not agreed to. That consent step is also what makes the resulting proof safe to put in front of prospects. **Should an AI interviewer pretend to be human?** No. Researchers and candidates broadly agree that undisclosed AI erodes trust, and it is the root of nearly every viral interview failure. A good customer interview says plainly that it is AI-guided, and that honesty is what makes the resulting proof credible. **Is an AI customer interview creepy?** It does not have to be. Almost all of the creep traces to one thing: a near-human face that your brain reads as slightly wrong. Shine renders no avatar at all, so there is no uncanny valley to fall into. The only face your customer sees on screen is their own.