Types of Video Interviews: One-Way, Two-Way and AI Formats Compared

Asynchronous vs live, recorded vs conversational, talking head vs screenshare: how the common video interview formats differ, and what the research says about each.

Almost every hiring process now has a camera in it somewhere. Greenhouse's May 2026 survey found 63% of US job seekers had faced an AI interview in the previous year, up 13 points in six months. But "video interview" covers four formats with very different architectures, and the architecture decides what the recording can tell you.

Key points

• The formats differ on five things: which way the conversation runs, who writes the questions, whether anyone can react to an answer, when it happens, and what the conversation is about.
• Two of the four are one-way. The candidate performs to silence, and nobody can ask "why?"
• Follow-up questions are where interviews get their information. Research on one-way video shows that even scripted probes change the result.
• Unlimited preparation and retakes raise ratings without raising honesty. Candidates and recruiters both know it.
• Candidates are already voting: pre-recorded, AI-scored video is the single biggest reason they walk away from a hiring process.

The four formats

1. Self-recorded pitch. The candidate records a talking-head video with a tool like Loom, answering a prompt from the hirer ("tell us why you're a good fit for this role, in under three minutes"), and attaches it to the application. The prompt is usually specific enough to keep answers in the same lane; what the candidate controls is the emphasis, the length and the number of takes.

2. One-way AI video interview. A platform shows fixed questions, records an answer to each within a time limit, and either a human reviews later or an algorithm scores the answers. Same questions for everyone, no one listening as they're answered. This is what most candidates mean by "AI interview."

3. Conversational AI interviewer. A voice or video agent asks questions and reacts to answers in real time. It can probe, clarify and redirect. The loop is closed, and the candidate still picks the moment. This category has two branches that differ in what the conversation is about:

• 3a. Conversation about experience. The agent asks about past work, the way a phone screen would. Products like this are built to replace the recruiter's first call.
• 3b. Conversation over work. The candidate first does a short, scoped task on their own time, in their own tools, with nobody watching. Then they open a screenshare, camera on in the corner, and walk back through it while the agent asks about the decisions behind it. The subject of the interview is an artifact the hirer can see. This is the branch Pudding is in.

4. Live human video call. A person on Zoom. Fully two-way, but it needs both calendars, it happens once, and unless the interviewer follows a script, every candidate gets a different interview.

What the missing follow-up costs

The first two formats share one property: the candidate speaks, and nothing responds. That turns out to be the most expensive line in the table.

Researchers at the University of Guelph and Saint Mary's University tested what happens when a one-way video interview adds follow-up prompts. Across two experiments with about 675 participants, the prompts were generic and written in advance ("tell us more about why you chose that course of action"). Even so, 77% of candidates who answered them scored higher than they had on their first answer alone, and the gain held against a control group given the same extra speaking time with no prompt. 86% of participants who went without prompts said they would have preferred them.

If scripted follow-ups move results that much, a live one that responds to what the candidate just said is doing a different job entirely. HR professionals interviewed for a 2025 PLOS One study described the one-way format as "the candidate just answers pre-set questions and moves on," with no way to probe what they actually know. One admitted that in a technical interview run this way, "we must take their word for it."

Candidates feel the same gap from the other side. A 2026 study of asynchronous AI interviewers, drawing on 18,000 Reddit posts and 17 interviews, found candidates expected something that could converse and got "a glorified pre-recorded form." One described it as "performing to silence. No response, no presence, just me and my doubts."

What unlimited rehearsal costs

Both one-way formats let the candidate prepare, and the pitch lets them re-record as many times as they like. A 2025 study by Lukacik and Bourdage found that more preparation time and re-recording raised interview ratings without increasing deceptive behavior. The polish is genuine. It just isn't a sample of how the person thinks when something unexpected comes up.

Recruiters see the gap when the same candidate shows up live. In the PLOS One study, HR professionals said candidates "appear very high" in communication on recorded video, then "do not exhibit the same energy" face to face. One said that after a few one-way interviews as a candidate, they had simply memorized their answers. Others described candidates reading ChatGPT output on camera while appearing to think. In the Reddit study, 41 users and 9 interview participants described tuning answers to keywords and using outside AI tools, because a system that can't react can't notice.

What the other two formats cost

Two-way formats pay in time. A live human call requires two calendars and happens once, so the structured-interview discipline that makes it comparable across candidates is often dropped under pressure. Sackett and colleagues' 2022 reanalysis of decades of selection research put structured interviews at a predictive validity of .42 and unstructured ones at .19. A friendly, improvised Zoom is the .19 case.

Conversational AI closes the loop without the calendar problem, which is why it is spreading. Retakes work differently here, too. Re-recording a pitch means polishing the same answer to the same prompt. Redoing a conversation means facing a new set of follow-ups, so a second attempt is another sample rather than a better-rehearsed one.

The split between 3a and 3b is about subject matter. In 3a the agent can only probe what the candidate says about their work. In 3b it is probing work it can see, so "why did you do it that way?" has something concrete to point at, and every candidate is explaining the same brief. The cost is that 3b asks more of the candidate (which is why Pudding pays for every completed session) and more of the reviewer, so it belongs late in a process, with a few people who have already cleared the early rounds.

Where candidates have landed

Candidates are not neutral about these architectures. In Greenhouse's survey, 38% had already walked away from a hiring process because of an AI interview, and the top reason, named by 33%, was "pre-recorded video interviews scored by AI with no human present." Gartner's 2025 survey of 2,918 candidates found only 26% trust AI to evaluate them fairly, and 25% trust an employer less once they learn AI is doing the evaluating.

Notice what the complaint is. It isn't video, and it isn't AI as such. It is being recorded with nothing on the other end.

Frequently asked questions

What are the main types of video interviews?

Four, by architecture: the self-recorded pitch (a Loom-style talking head attached to an application), the one-way AI video interview (fixed questions, recorded answers, often algorithm-scored), the conversational AI interviewer (a live agent that reacts to answers, about either past experience or a piece of work the candidate shows), and the live human video call. The first two are asynchronous and one-way; the last two are two-way.

What is the difference between a one-way and a two-way video interview?

In a one-way interview the candidate records answers and nobody responds; the recording is reviewed later by a person or an algorithm. In a two-way interview something reacts in real time, whether a human or an AI agent, so follow-up questions depend on what the candidate just said. The research above suggests that the follow-up, not the camera or the AI, is what separates the formats' results.

Is a one-way AI interview the same as a self-recorded pitch?

Structurally, almost. Both are one-way recordings nobody reacts to. The one-way interview breaks the ask into several fixed questions with time limits, and often adds algorithmic scoring. The pitch is a single prompt with unlimited retakes and more room for the candidate to decide what to emphasize.

Does adding scripted follow-ups fix a one-way interview?

It helps, measurably, but the prompts are written before anyone hears the answer. They can ask for more; they can't ask about what was said.

Why do candidates withdraw from AI interviews but accept live human ones?

The research points at the loop, not the technology. Formats where something responds, whether a person or an agent, don't draw the same reaction as formats where the candidate performs into silence.

Where Pudding fits

Pudding is a 3b tool. Your team writes the brief. The candidate does the work on their own schedule, then opens a screenshare whenever they're ready and explains it while a Pudding agent asks about the choices behind it. The agent asks and records; your team reviews the work and the recording and makes the call. Every completed session is paid automatically through Stripe, with no invoices or payout admin on your side.

Because it asks real work of the candidate and real attention of the reviewer, it is not a top-of-funnel screen. It earns its cost in two places: small pools of hard-to-find specialists, where every candidate matters and the good ones have other offers, and late rounds, where you're separating a few finalists who have already cleared your phone screen and look equally strong on paper.

Sources

1. Greenhouse (2026). 2026 Candidate AI Interview Report. https://www.greenhouse.com/newsroom/63-of-job-seekers-have-faced-an-ai-interview-most-havent-had-a-good-one-yet
2. Patel, R. D., Powell, D. M., Roulin, N., & Spence, J. R. (2025). Tell Me More! Examining the Benefits of Adding Structured Probing in Asynchronous Video Interviews. International Journal of Selection and Assessment, 33(1). https://doi.org/10.1111/ijsa.12514
3. İlhan, Ü. D., Güler, B. K., Turgut, D., & Duran, C. (2025). Opportunities and challenges of asynchronous video interviews: Perceptions of human resources professionals from Türkiye. PLOS One, 20(6). https://doi.org/10.1371/journal.pone.0325932
4. Expecting Too Much, Getting Too Little: Exploring the Challenges and Design Opportunities of Asynchronous AI Interviewers (2026). arXiv. https://arxiv.org/abs/2601.02775
5. Lukacik, E.-R., & Bourdage, J. S. (2025). Does Design Matter? The (Limited) Effects of Six Asynchronous Video Interview Design Features. International Journal of Selection and Assessment, 33(1). https://onlinelibrary.wiley.com/doi/10.1111/ijsa.12511
6. Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection. Journal of Applied Psychology, 107(11). https://pubmed.ncbi.nlm.nih.gov/34968080/
7. Gartner (2025). Survey of 2,918 job candidates on AI in hiring. https://www.gartner.com/en/newsroom/press-releases/2025-07-31-gartner-survey-shows-just-26-percent-of-job-applicants-trust-ai-will-fairly-evaluate-them