You click the link, allow the camera, and a voice asks you to walk through a system you built. There's no face nodding back at you. That silence is what throws most engineers off in an AI video interview, and it's the easiest part to prepare for.
A growing number of US companies put an asynchronous AI-run interview early in the process. We do it too: after the coding assessment, candidates for our open roles do a recorded interview in English, on their own schedule, before they meet a human team. So here is what I'd tell a friend who has one tomorrow.
Key Takeaways
- Set up your room, camera, and audio the day before. Bad audio hurts more than a bad answer.
- Answer with a structure: the situation, what you did, the result, what you'd change.
- Speak in the English you use at work. Clear and a bit slow beats fast and fancy.
- Don't read a script and don't bluff. Specific, honest answers are what this format rewards.
Get the setup right first
An AI interview can't ask you to repeat yourself the way a person would. If the transcription misses half your words, your answer is half gone. So the technical setup matters more here than in a normal video call.
| Item | Do this | Avoid this |
|---|---|---|
| Audio | Wired earbuds or a headset mic, quiet room | Laptop mic next to a fan or a street window |
| Camera | Eye level, face lit from the front | Window behind you, camera looking up your nose |
| Connection | Cable or close to the router, other devices off the network | Shared Wi-Fi while someone streams video |
| Browser | The browser the platform recommends, updated, extensions off | An old browser with ad blockers that break the recorder |
| Time | A slot where nobody will interrupt you for 45 minutes | Squeezing it into a lunch break |
Do a test recording on your phone or laptop and listen back. If you sound muffled to yourself, you'll sound worse to a transcription model.
How to structure an answer when nobody reacts
In a live interview, the interviewer's face tells you when to stop. Here nobody does, so engineers either stop after two sentences or talk for eight minutes. Both hurt.
Use a simple four-part shape for behavioral and experience questions:
1. Context. One or two sentences. "At a logistics startup, our order API was timing out during peak hours."
2. What you did. The biggest part. Your decisions, not the team's. "I profiled the slow endpoint, found an N+1 query in the pricing lookup, and moved it to a single batched query with a cache."
3. Result. A number if you have one. "p95 latency went from about 2.1 seconds to 300 milliseconds."
4. What you'd change. This is the part most people skip, and it's the part that shows seniority. "I'd add a load test to CI so we'd catch it before customers did."
Aim for 90 seconds to 3 minutes per answer unless the question asks for more. If the platform shows a timer, glance at it. If it allows a retry, use it only when something actually went wrong, like a noise or a lost train of thought.
For technical questions ("How would you design a rate limiter?"), think out loud. Say the tradeoffs. "Token bucket is simple and handles bursts. If we have several instances, I'd keep the counters in Redis." The system can only score what you say, so a silent 30 seconds of thinking followed by a one-line answer loses information.
What these interviews usually look at
Every company tunes its own scoring, and we don't publish ours. But in general, AI interview tools are set up to look at a few broad things:
- Whether you actually answered the question asked
- Technical depth and accuracy for the role
- How clearly you explain a decision, including tradeoffs
- Spoken English you can work in every day
Notice what's missing: accent. A clear Colombian or Brazilian accent is fine. What causes trouble is speaking so fast that words blur together, or switching to Spanish when you get stuck. If you blank on a word, describe it in simple English and keep going. That's exactly what you'd do in a real standup. If English is your weak spot, this plan for going from B2 to C1 is where I'd start.
Don't read, don't bluff
Two things backfire.
Reading from a script is easy to spot. Your eyes move across the screen and your voice goes flat. Keep notes to a few bullet points per project: the problem, your role, one number.
Bluffing is worse. If you don't know Kafka, say you haven't used it in production and explain how you'd approach learning it, or what you used instead. A made-up answer gets checked later, by a human, in the team interview. At that point it costs you the job.
A Concrete Version
A composite example, not a real candidate.
Mariana is a mid-to-senior React Native engineer in Medellin. Her English is solid in writing, shakier out loud. The night before, she records herself answering "Tell me about a hard bug you fixed." First take: 55 seconds, rushed, no result. She listens back and rewrites her notes as four bullets: offline sync, duplicate orders, idempotency keys, crash rate from 1.8% to 0.3%.
Second take: 2 minutes 10 seconds, slower, with a closing line about adding a Detox test for the offline case. She moves her desk so the window is in front of her instead of behind, and switches from laptop mic to wired earbuds.
The next day, one question is about a library she barely knows. She says so, then explains how she evaluated a similar one. That honest answer holds up when the client team asks about it a week later.
The Honest Counterpoint
AI interviews are not a perfect filter, and I won't pretend otherwise. They can struggle with strong accents on bad audio. They reward people who are comfortable talking to a camera, and some excellent engineers aren't. That's why at Ruzora it's one step among several, next to a coding assessment and interviews with real people, and never the only thing we look at.
If a company uses an AI interview as the entire process, with no human ever involved, be careful. That's a sign they may not invest much in you after the hire either. And if an "AI interview" link comes from an unknown sender and asks you to install software, stop. Check the company first (here's how).
Frequently Asked Questions
Can I use notes during an AI video interview?
Short bullet notes are usually fine unless the instructions forbid them. Reading full answers is not, and it's obvious on camera.
What if my connection drops in the middle?
Most platforms save progress or let you resume. Contact the recruiter right away with the time it happened. Don't just start over silently.
Does my accent count against me?
Clarity counts. Accent, on its own, shouldn't. Slow down, finish your words, and use a decent mic.
The Bottom Line
Prepare the room the day before, practice two or three answers out loud with the context, action, result, and lesson shape, and be honest when you don't know something. That covers most of what separates a good AI interview from a bad one.
If you're ready to try it, browse our open roles. Roles like AI engineer and React Native engineer include this step after the coding assessment.
Roberto Espinoza is CEO of Ruzora, which places senior LATAM engineers with US startups. Browse open roles.
