AI Interviewer API: Full Guide

An AI interviewer API lets your product run a live, two way interview where an AI asks the questions, follows up when an answer is vague, scores the candidate against a rubric once the call ends, and hands the recording, transcript, and score back to your own system. Instead of sending candidates to a separate screening tool, you build the first round into the ATS, staffing platform, or marketplace they already use, under your own brand, with the results landing wherever you already track pipeline.

This guide is for whoever is deciding whether that is worth building: a CTO weighing an API against a turnkey platform purchase, or a head of talent trying to size what faster screening and a cleaner pass through rate actually buy them. The question underneath both is the same. Does the first round interview belong inside your product's data flow and brand, sized to your own numbers, or is it fine living on someone else's login page with someone else's benchmarks attached to it.

Why AI interview software moved from platform to API

The category has settled around a simple frame, an AI interviewer for automated candidate screening, and nearly every recruiting tech vendor has followed that lead. AI recruiter tools, AI video interview platforms, and structured interview products now sit on nearly every ATS vendor's roadmap, because the phone screen is the most repetitive, most schedule dependent step in hiring and the easiest one to hand to a system that never gets tired of asking the same ten questions well.

But a platform is a destination. A recruiter has to leave the ATS they live in, open a separate tool, invite the candidate to yet another link, and check a separate dashboard for the result. Every ATS, staffing platform, and marketplace now wants that first round running inside its own product, under its own brand, writing straight into its own pipeline view, not sitting behind a tool a recruiter has to remember to check.

That is the gap an API closes. Everything a screening platform sells under the hood, an interviewer persona, a scripted question flow, a scoring rubric, transcripts and analysis, gets exposed as building blocks you call from your own backend and render in your own interface. You keep the data. You keep the brand. The candidate never leaves your product, and the score never lives somewhere your own systems cannot see it.

The workflow this replaces is usually a recruiter blocking thirty minutes on a calendar for every single candidate in a funnel, chasing no shows, and repeating the same opening questions dozens of times a week. That does not scale past a handful of requisitions at once. An API means candidate two hundred this quarter gets the same structured interview as candidate one, scored the same way, without booking a single calendar slot.

The hiring math: automated candidate screening, time to hire, and offer accept

Before the anatomy and the build, it is worth doing the arithmetic that justifies a build or buy decision, because "candidates get screened faster" is not a budget line on its own. A few levers turn automated candidate screening into a funnel a recruiting leader actually feels, and the math holds from first principles, no vendor's number required.

Screening capacity per recruiter is the ceiling today. A recruiter running thirty minute phone screens, with scheduling back and forth and the inevitable no shows, realistically clears somewhere around fifteen to twenty candidates a week per open requisition before something else on their desk suffers. An AI interviewer does not have a calendar. It runs whenever the candidate is free, at midnight or on a weekend, and the fifteenth interview of the day costs the same setup time as the first. Run this against your own requisition load and recruiter headcount, not the illustrative number above, before you decide what it is worth.

Time to hire compresses when the first round stops waiting on a calendar match. Published case studies across this category report screening time cut by half or more once the first round moves off a shared calendar. You do not need someone else's case study to trust the direction. If a candidate today waits an average of five business days for a first screen, mostly waiting on a recruiter's open slot, and an always available interviewer collapses that wait to same day, the time to hire compression is just the wait time you removed, multiplied by however many requisitions are open at once. Every day a requisition stays open is a day of lost productivity on the team waiting to fill it.

Pipeline pass through improves because more candidates actually get screened. A recruiter who can only reach fifteen candidates a week filters the funnel by who got a calendar slot, not who was the best fit. An API embedded interviewer that is available around the clock screens closer to everyone who applies, which means the humans making the next round decision are choosing from a wider, more consistent set of scored transcripts instead of whoever happened to answer the phone.

Cost per screen drops from a loaded recruiter hour to a fraction of one. Picture a recruiter's loaded cost at roughly forty dollars an hour. A thirty minute phone screen, plus the scheduling overhead around it, is closer to forty five minutes of that recruiter's time per candidate. An API call that runs the same structured conversation costs a sliver of that, freeing the recruiter's actual hours for the judgment calls only a human should make, which round to advance, how to sell the offer, how to read a candidate's hesitation. Replace the illustrative number with your own recruiter cost before you present this internally.

Twenty four seven availability changes the candidate experience, and candidate experience is upstream of offer accept. A candidate who can interview the same evening they apply, instead of waiting a week for a slot, forms an impression of the company before a human ever speaks with them. Faster response time is one of the most consistently cited drivers of offer accept rate in hiring research generally, and an always available first round is the most direct way to compress that response time without adding headcount.

None of the specific figures above are Waterr's verified numbers. They are illustrative calculations meant to be rerun with your actual requisition volume, recruiter cost, and current time to hire. The mechanism holds regardless of whose numbers you plug in: an always on first round screens more candidates, faster, at a lower marginal cost, and hands the humans in the loop a cleaner set of scored transcripts to choose from.

The anatomy of an AI interview

The category has settled on shared vocabulary. Here is how the pieces a screening platform sells map onto what you would build.

Once you see the mapping, the rest of the decision is about who owns each piece and where the results live, not what each piece does.

Send the ready agent, or build a custom one

None of this requires authoring anything to start. Every new account is auto seeded with two ready to use scenarios: a personalized conversational meeting and a Requirement Gathering agent that interviews you to scope a project, and the interview category has its own set of ready made templates on top of that, a behavioral interview, an HR culture fit interview, a technical screen for software engineers, a machine learning focused technical screen. None of those need a script written from scratch or a persona designed from a blank page. Personas can also draw a face and a voice from a curated catalog rather than commissioning custom assets, so an interviewer that looks and sounds the part is a few clicks, not a design project.

That means the honest starting point for most teams is not a build project. It is sending a candidate the ready first round template today, on a real requisition, and watching what the transcript and scores look like on a handful of actual candidates before committing engineering time to anything custom.

Send the ready agent when:

• You are piloting the format and want real candidate reactions this week, not next quarter
• The role needs a generic first round screen: background, motivation, basic qualifying questions
• You are running a scoping or intake conversation rather than an evaluation, which is what the auto seeded Requirement Gathering agent already does
• You need a live demo for stakeholders before asking for integration budget

Build a custom agent when:

• Your rubric has specific, weighted scoring criteria unique to a role or team
• The interview needs to go deep on a domain, a specific tech stack, a specific compliance requirement, a specific product
• Candidates need to see your brand end to end, not a generic template experience
• You are running a multi round loop where round two needs to remember what round one covered
• Your organization has a proprietary interviewing methodology a template cannot approximate

The ramp path most teams actually take is send first, customize later. Start with the ready agent, run it against real candidates, see where the transcripts and scores fall short of what a human interviewer would have caught, and only then invest in a persona, script, and rubric built for exactly your role. Most of what looks like a build decision on day one turns out to be a configuration decision once the data from a real pilot is in hand.

How an integration comes together

Standing up an AI interviewer as a feature of your own product follows the same shape regardless of vendor.

1. Define the interviewer. Describe the persona your candidates will meet, a technical lead screening for backend depth, a hiring manager assessing culture fit, a staffing recruiter running a standardized first round.
2. Script the interview. Write the structured conversation the AI should run, the opening, the topics to cover in order, how to handle a vague answer, and when to wrap.
3. Set the scoring rubric. Define the goals, the three or four criteria the AI grades the transcript against once the interview ends, invisible to the candidate throughout.
4. Launch the session. Your product creates the interview and gets the candidate into it, no separate login, no separate app.
5. Results land in your ATS. When the interview ends, the recording, transcript, and scores come back automatically and write into your own pipeline view, not a third party dashboard your team has to check separately.

That is the whole loop. A new interview type after the first is a configuration change, a new persona and script, not new engineering work. If your engineering team wants the field level build, the technical reference lives at docs.waterr.ai, and the hands on build guide is the version to hand them once you have decided this is worth building rather than buying off a shelf.

Live conversation versus asynchronous video interview

Most tools marketed as an AI video interview platform are still one way. The candidate records themselves answering a fixed list of questions, alone, on camera, and a human or a model reviews the tape afterward. That is an asynchronous video interview, and it is closer to a video screener than an actual interview.

The difference shows up the moment an answer is vague. A real interviewer does not just move to the next question when a candidate gives a one sentence answer about a past project. It asks what specifically they built, what broke, and what they would do differently. A recording cannot do that. It asks, waits, and stops, because there is no listener on the other end while the candidate is talking.

A live, two way AI interviewer behaves like the second kind. It follows up on vague answers in real time, adjusts its next question based on what the candidate just said, and can see a screen share or a camera feed while the conversation is happening rather than reviewing a static video after the fact. That distinction is the entire reason to build on a conversational engine instead of a recording pipeline. Async is cheaper to build and easier to review at a glance. Live is the only format that produces a transcript worth scoring the way a real interview deserves to be scored.

What you can build on this

Once persona, scenario, goals, session, and analysis are primitives you control, the ambition widens past a single screening step. Everything below is buildable on documented pieces of the same system, not a hypothetical roadmap.

• First round screening inside an ATS. Embed the interview directly into the requisition workflow your recruiters already open, so the score shows up next to the resume instead of in a separate tab.
• High volume hourly hiring flows. A booking page or a public share link lets a candidate self schedule an interview the moment they apply, which matters most for roles with hundreds of applicants and thin recruiter time per hire.
• Staffing agency interview as a service. White label the entire candidate facing experience so a staffing agency's clients see the agency's brand, not a third party tool bolted onto the placement process.
• Marketplace vetting. The same persona, scenario, and goals structure that screens a job candidate can verify a freelancer, tutor, or driver joining a marketplace, with a live conversation standing in for whatever manual vetting call the marketplace runs today.
• University admissions interviews. A structured, scored conversation format applies just as directly to an admissions screen as it does to a hiring one, with the same rubric based scoring behind it.
• AI mock interviews for candidate prep products. Vision is explicitly built for this: the AI observes body language and gives feedback on eye contact, posture, and confidence signals, which is the core loop of an interview practice product.
• Multi round loops with participant memory. A signed in candidate returning for round two does not start cold. The system carries forward what they covered in round one, scoped to that scenario, so round two builds on round one instead of repeating it.
• Multi language global hiring. The same scenario runs in a candidate's native language, auto detected, with evaluation that understands language specific nuance, so a global requisition is not forcing every candidate through an English only screen to get a fair read.
• White label under your own brand. On enterprise plans, the interview interface, share links, and end of session page carry your brand instead of the vendor's, so a candidate inside your ATS sees your logo, not a tool a recruiter had to explain away.

None of these require new API surface. They require deciding which combination of persona, script, and delivery method matches the hiring problem in front of you.

Structured interviews at scale: screening flows that work

Interviewing everyone in a funnel, not just the candidates who happened to get a calendar slot, is the whole point of moving screening onto an API. A few flows come up most often.

AI phone screen automation: the first round screen

The classic first round, ten to fifteen minutes, covers background, motivation, and one or two qualifying questions specific to the role. Scripted well, the AI does not just read questions off a list. It asks the follow up a recruiter would ask when a candidate's answer is thin, and it ends the call the same way every time, so every candidate in the funnel gets the same structured interview instead of whichever version of the pitch a tired recruiter gave on their tenth call of the day.

AI video interview for high volume hourly hiring

Retail, logistics, and hospitality roles generate applicant volume no recruiting team can phone screen one by one. A self scheduled, always available interview removes the scheduling bottleneck entirely: the candidate applies, picks a slot or starts immediately, and the score is waiting by the time a hiring manager opens the requisition the next morning.

Beyond those two, the same structure covers the rest of a hiring funnel:

• Technical screening with vision. With screen share enabled, the AI reads code as a candidate writes it and asks about the specific approach on screen, not a description of one.
• Culture and values conversation. A scenario scripted around the company's actual values, not generic behavioral questions, surfaces answers a template question never would.
• Language specific screening. The same scenario, run in the candidate's own language, for a distributed or international applicant pool.
• Second round loop. A deeper, more technical conversation that picks up on unresolved threads from round one instead of re asking the basics.

What separates a screen that produces real signal from one that wastes a candidate's time is specificity in the script, not which vendor built the underlying model. "Ask about their experience" produces a generic answer any candidate can recite. A script that asks for the exact project, the exact failure, and the exact fix produces something a hiring manager can actually evaluate.

Embedding in your ATS: integration and candidate experience

The reason to build this on an API instead of pointing candidates at a separate platform is that the interview, and everything it produces, stays inside your product.

Embed. The interview drops directly into a requisition page or a candidate portal as an inline widget or a floating start button, so candidates never see a separate login screen mid application.

White label. On enterprise plans, the interview interface, invite emails, and share links carry your brand instead of the vendor's, so a candidate applying through your platform sees your logo, not a third party tool.

Continuity across rounds. A candidate coming back for a second interview in the same scenario does not start from zero. The system carries forward what they covered last time, which is the API equivalent of a platform's multi round feature, except every round writes into a pipeline you already control.

Screening report, pushed into your ATS. Scores write into your system automatically the moment an interview is scored, delivered reliably so a flaky endpoint on your side does not silently drop a candidate's result. The deeper reference on defining a scoring rubric is worth a look if your team is designing the goals before committing to a specific structure.

That combination, embedded, branded, continuous, and automatically synced, is what separates automated candidate screening built into your product from a separate tool your recruiters have to remember to open.

Platform versus API: an honest comparison

Neither is wrong. If your recruiting team just needs candidates screened this quarter and a dashboard to go with it, a platform gets there faster, and the category numbers above are a reasonable proxy for what to expect. If screening needs to live inside your ATS's requisition flow, your staffing platform's placement workflow, or a marketplace's vetting step, an API is the only path that does not bolt a third party brand, and someone else's benchmark, onto your product.

There is also a middle case worth naming: teams that start on a platform to prove the idea, then move the scenarios and rubrics that worked onto an API once they know which questions and personas actually predict a good hire. Nothing about the persona, scenario, goals structure above is vendor specific, so that migration is mostly re scripting, not re architecting.

What this looks like end to end

A staffing platform that builds screening into its own product typically sets it up once: a persona for each role family, a handful of scenario scripts covering the questions that matter most for those roles, three or four goals per scenario, and results wired straight into the client facing pipeline view. A candidate applies, clicks a link inside the platform they already applied through, completes the interview whenever suits them, and the score shows up next to their resume, no separate tool, no separate login, no recruiter chasing a calendar slot. That is the shape screening takes once it stops being a platform a recruiter visits and becomes a feature of the product candidates already touch.

Frequently asked questions

What is an AI interviewer and how does it work? An AI interviewer is a persona that conducts a live, two way conversation with a candidate, following a scripted structure, asking follow up questions when answers are vague, and scoring the transcript against a rubric once the call ends. It runs over a real time video and voice connection, not a recorded, one way video upload.

Does an AI interviewer replace recruiters? No. It screens and produces a scored, timestamped record so a recruiter or hiring manager can evaluate more candidates, more consistently, in less time. The AI runs the conversation and the scoring pass. A human still decides who advances and who gets the offer.

Can I customize the interview questions and scoring? Yes. The interview script, its opening, its topics, its follow up rules, is fully custom per role, and the scoring rubric is a set of goals you define yourself, each with its own name, description, and scoring instructions, invisible to the candidate during the call.

What does the recruiter see after the AI interview? A per goal score, written feedback tied to specific moments in the conversation, an overall average, the full transcript, and the recording, all delivered back to your own system automatically once the analysis finishes.

Is AI interviewing fair to candidates? Fairness here means transparency, not a claim about bias elimination. A configurable consent step tells candidates they are speaking with an AI and whether the session is recorded before it starts, and the scoring rubric a candidate is evaluated against is defined by your team, not hidden inside a vendor's black box. The hiring decision itself stays with a human.

A platform gets a team screening candidates this quarter, with someone else's benchmark attached to it. An API gets the interview living inside the product candidates already touch, scored the way your rubric says it should be, reported wherever your pipeline already lives, and measured against your own time to hire instead of a case study on a vendor's website. For the broader category this sits inside, what an AI meeting API is covers the same primitives powering sales roleplay and testimonials, and the sales roleplay API guide is the closest sibling if your next build is training reps rather than screening candidates.