Take the AI Screening Interview Yourself

On August 28 we put a link on the HireQwik site that lets any visitor get screened by our production AI interviewer. No signup, no sales call, no credit card. You give a name and a work email, you talk for about five minutes, and a few minutes later your inbox has the same kind of scorecard a recruiter would read off a shortlist.

If you want to try an AI screening interview demo yourself, that page is at hireqwik.in/ai-interview-demo. This post is about why we built it, what it deliberately does not hide, and what we think you should pay attention to while you are in the call.

Why a screening vendor should let you be the candidate

Every vendor in this category claims their AI holds a real conversation. We claim it too. The problem is that the claim is unfalsifiable from the outside. You sit through a 45 minute call, you watch a recording of somebody else’s interview, you see a dashboard on a shared screen, and at the end you have an impression rather than evidence.

That gap is the reason buying decisions in AI screening take three months and still go wrong. The buyer never experiences the product the way the person it acts on experiences it. A TA director evaluates a screening tool entirely through the recruiter’s view, then deploys it against two thousand candidates who will only ever see the other side.

The cost of that blind spot is now measurable. Greenhouse’s 2026 candidate AI interview report puts the abandonment rate at 38%, meaning that share of job seekers has walked out of a hiring process specifically because an AI interview was in it. A matched 38% finished a well run one with a better opinion of the employer, while 34% finished a bad one with a worse one. That spread is not a fact about AI, it is a fact about implementation, and the buyer who never sat in the chair has no way to tell which half they just purchased.

So the argument we are making with this page is simple. If the conversation feels robotic to you, it will feel robotic to your candidates, and you should not buy it. We are comfortable letting the interview make that case without a salesperson standing next to it.

There is a second reason, less noble and worth admitting. We were tired of describing the product in prose. A blog post explaining how a voice agent handles an interruption is worse than thirty seconds of being interrupted.

What actually happens in the five minutes

The flow is short enough to describe completely.

You enter a name and a work email on the page. The email matters because your scorecard goes there, and nothing beyond those two fields is asked of you. The page posts to a public demo endpoint on the app, gets back a join URL for an interview room, and forwards you into it. The interview runs in the browser over WebRTC, which is the same transport a real candidate uses when they join from a hostel room on a phone. You need a microphone and a reasonably quiet spot. Chrome, Edge and Safari all work, on desktop or mobile. There is nothing to install.

The AI interviewer greets you and asks structured questions about your own hiring: what roles you are filling, what volume you are dealing with, what screening costs you today. Those questions exist because they are the honest version of a discovery call, and because they give the evaluation layer something real to grade. You are not asked to pretend to be a fresher applying for a developer role.

Then it wraps up. A few minutes later the scorecard email lands.

One deliberate constraint: we cap how many live demos run per day. When the day’s slots are full, the page says so and offers a booked walkthrough instead of queueing you into a failure. That cap exists because concurrency in voice screening is a real resource, not a marketing number, which is a point we have made at length in the concurrency question vendors dodge.

The four things worth noticing while you are in it

Most people take a demo interview and come out with a vague feeling. Here is what to actually watch for, in the order it becomes observable.

Does it interrupt you correctly? Real candidates trail off, restart sentences, and go quiet mid thought. A voice agent that treats every pause as a finished answer will cut off half your applicants. A voice agent that waits too long feels dead. Try trailing off deliberately and see what happens in the gap.

Does the next question come from what you just said? This is the single highest signal moment in any AI interview demo, and it is the whole subject of a separate post on telling an adaptive interviewer from a scripted one. Give a deliberately incomplete answer and see whether the follow up probes the gap or moves to the next item on a list.

How does it handle your accent, your pace, and your filler words? Speak fast. Use “um” freely. Indian English, regional pronunciation, code mixing: all of it is what your actual candidate pool sounds like. Speech evidence enters our scoring asymmetrically: it has the power to rescue someone sitting on the boundary, and no power at all to pull down a candidate who answered well. The reasoning behind that design is in the asymmetric speech blend.

What does silence do? Stop talking for fifteen seconds. An agent that panics into the gap, loops, or hangs up will behave identically when the silence belongs to a nervous fresher rather than to you.

None of those four things show up in a feature list. All four of them decide whether a drive completes.

What arrives in your inbox afterwards

The scorecard is the part that surprises people, because it is not a summary. It is the artifact the recruiter would work from.

It carries a per question evaluation with quotes pulled from your own answers, so every score points at the specific thing that produced it rather than at a number with no provenance. It carries a read on communication that comes from the audio itself and not only the transcript, which is a distinction most tools in this category cannot make because they score from text alone. And it carries an overall verdict tier, the Strong Go, Go, Hold or No Go band a shortlist actually sorts by.

Reading that document critically is a skill, and it is a different skill from taking the interview. We wrote a whole companion piece on what your own demo scorecard tells you, including what a weak one looks like, because a scorecard that scores you highly is not automatically a scorecard that would survive a hiring committee.

One honest caveat about your own result: it will probably flatter you, and a flattering verdict proves very little about accuracy. What the document can genuinely settle is whether the reasoning underneath the verdict is followable. We took that apart properly in the companion piece below.

Who else on your team should take it

One person taking a demo interview produces an opinion. Three people taking it produces something you can act on, and the three should not be three recruiters.

The recruiter who will actually run the queue. Not the head of TA, the person who will open the review screen every morning during the drive. They are the only one who will notice whether the output is workable at 8 AM with 300 rows in front of them. Their objections will be operational and specific, and they are the objections that decide whether the tool gets used after month two.

The hiring manager for your hardest role. Give them no context and let them answer as themselves. What you are testing is whether the evaluation dimensions match what that manager actually cares about. If your manager for a field sales role reads the communication score and shrugs, you have learned that the default rubric is not calibrated for the role, which is a fixable problem but only if you find it before the drive rather than during it.

Somebody who will do badly. This is the uncomfortable one and it is the most informative. Ask a colleague to deliberately give thin, hesitant, unstructured answers, the way a nervous fresher does. Then read what the system said about them. A screening tool’s character shows in how it describes a weak candidate, not a strong one. Look for whether the language stays factual and evidence linked, or whether it drifts into judgments about the person. You are going to send rejections based on this reasoning to thousands of twenty two year olds, and some of them will read the wording aloud to their friends. Surveys of AI hiring in 2026 consistently find that a large share of rejected applicants never receive a reason at all, which is the low bar you are trying to clear.

Run the three sessions inside a week and compare notes in one meeting. The disagreements between your own three people are the useful output, more than any of the individual scores, because they surface the fact that your team does not currently agree on what a good first round answer sounds like. That disagreement exists whether or not you buy anything. Most high volume first round screens in India are calibrated by nobody in particular, and a free five minute interview is the cheapest way to make that visible.

Where this fits against a normal vendor evaluation

We already publish a thirty minute vendor evaluation checklist, five questions to put to a salesperson before they start presenting. That checklist is still the right instrument for the things you cannot see: per JD rubrics, audit trails, the handling of a formal complaint six weeks after a rejection, and whether any Indian drive at your volume has actually completed on their platform recently.

The live demo covers a different surface. It answers the experiential questions the checklist cannot reach, because those questions have no verifiable verbal answer. “Does your agent handle interruptions gracefully?” gets a yes from every vendor on earth.

Use both, in that order. If a vendor cannot let you do the first one at all, that is itself information worth having.

What we are not claiming

Two things this page does not prove.

It does not prove the system works at volume. A single conversation says nothing about what happens when eight hundred candidates hit a booking page in one evening, which is a scheduling and concurrency problem rather than a conversation problem. Our own pilot numbers are a 3,000 candidate evening and an 89% cut in the HR hours each candidate consumes, measured against manual phone screening. Interrogate both. Neither is something the five minutes you just spent could have shown you.

It also does not prove the rubric is right for your role. The demo asks about your hiring because that is a fair conversation to have with a stranger. A real screen runs against a job description, with the disqualifier questions that end a call in the first two minutes and a scoring rubric built from that specific JD. That is a different and much more consequential thing, and the only way to judge it is to run one against a live role of yours.

Which is the actual next step, and the reason the page ends with an offer of a walkthrough rather than a signup form. Bring a real job description, and we will screen against it.

The take

The category has trained buyers to evaluate screening software the way you evaluate a CRM: features, integrations, a demo environment somebody else drives. That framing is wrong for this product. Screening software is the first impression thousands of candidates form of your company, and the only way to judge a first impression is to have one.

Five minutes, in your browser, with nobody watching. Take the interview, then decide whether the rest of the conversation is worth your time.