Interview integrity
Aced the interview.Cannot do the job.
Candidates are reading answers off a second screen while you take notes. Zavura captures their screen at moments they cannot predict, then hands the frames to you.
À la carte or enterprise · $5 per interview or less
Where hiring is right now
- 1 in 4
- Candidate profiles projected to be fake by 2028, driven by generated text, synthetic voice and deepfake video.
- Gartner
- 59%
- Hiring managers who suspect candidates have used AI to misrepresent what they can actually do.
- Industry survey, 2025
- 15% to 35%
- Measured growth in candidates using live interview assistance across a single year of hiring data.
- Fabric, Jun to Dec 2025
- $20
- Monthly cost of a subscription that listens to your questions and writes the answers on their screen.
- Public pricing
What is actually happening on your calls
Four ways a candidate gets fed answers while you are talking to them
None of this requires technical skill from the candidate. It requires a card on file. The reason a good interviewer still misses it is that the signals are timing, gaze and sentence structure, and you are busy listening, taking notes and thinking of your next question.
- 01
A layer drawn on top of their screen
The tool paints answers over whatever the candidate is already looking at, and excludes itself from the video the meeting software sends you. They read straight ahead while appearing to study their own work.
TellEyes track left to right in even lines instead of drifting while thinking.
- 02
Your voice, transcribed and answered
Your question is captured through a virtual audio device, run through speech to text, passed to a language model, and returned as a structured answer. The whole loop takes about two seconds.
TellEvery answer arrives after the same short pause, easy or hard.
- 03
Your coding problem, read by machine
For technical screens the tool reads the problem off the screen with optical character recognition, solves it, and supplies a complexity analysis and a plain English walkthrough to recite.
TellA clean optimal solution with no exploration, no dead ends, no rework.
- 04
A phone parked below the camera line
When the interview software gets stricter, the answer moves to a second device. The tool runs on the laptop and pushes text to a phone positioned just under the webcam frame.
TellA repeated downward glance at a fixed spot, always to the same place.
Why the usual answers do not hold
Predictable checks get prepared for. Constant recording gets ignored.
Approach A
Scheduled integrity checks
A room scan at the start, a check at the halfway mark. Anything a candidate can anticipate is something a candidate can prepare around. The overlay closes for ten seconds and reopens.
Fails because it is a pattern
Approach B
Record the whole session
Continuous capture of screen and camera. It catches things, in theory. In practice nobody watches forty minutes of footage per candidate, and you now hold a large pile of personal data you have to store, secure and justify.
Fails because nobody reviews it
Zavura
Sample at random, review what matters
A handful of stills taken at moments nobody can predict, including us. Small enough that you will actually look at all of them. Small enough that the data you are holding stays defensible.
Works because there is no pattern to learn
How it works
Random beats continuous
The value is not in seeing everything. It is in the candidate not knowing which moments you will see. A person who cannot predict the next capture has to behave as though every second is the one that gets recorded.
What a capture contains
- The candidate's shared screen at that instant
- A timestamp relative to session start
- A sequence index
What it does not contain
- No webcam feed
- No microphone audio
- No keystrokes
- Nothing captured outside the session window
- 01
Send one link
Schedule the interview the way you already do. Zavura gives you a session link to send alongside it. Nothing to install on your side, no calendar integration to configure, no help from IT.
Setup: about a minute
- 02
The candidate consents and starts
They open the link, read what will be captured, and agree before anything begins. Capture is scoped to that session only and stops the moment the call ends.
Consent recorded and timestamped
- 03
Capture happens on no schedule
Stills are taken across the session at intervals drawn at random. There is no fixed cadence, no countdown, and no indicator that a capture is coming. Not even we know when the next one lands.
Typically 20 to 40 frames per hour
- 04
You get the frames when the call ends
A single session view with every frame in order, timestamped, ready to scan in under a minute. Frames showing more than one active window are surfaced first so you are not hunting.
Available immediately after the call
The arithmetic
You do not need every second. You need enough draws.
A candidate leaning on a tool has to keep it on screen long enough to read from it, over and over, across the whole interview. Each capture is an independent draw against that exposure. The odds stop favouring them quickly.
At least one capture lands while the tool is visible
84.0%
- Total exposure
- 2m 40s
- Share of the session
- 5.9%
- Model
- 1 − (1 − e/T)ⁿ
The model treats captures as independent draws spread evenly across the session and assumes exposure windows do not overlap. Real sessions are messier in both directions. We are showing you the formula rather than a marketing number so you can argue with it.
Evidence, not verdicts
The tool does not decide anything about a person
This is a deliberate limit, not a missing feature. The moment a system scores candidates and acts on that score, it becomes a decision system, and in a growing list of jurisdictions that carries audit duties, disclosure duties and real liability. A folder of screenshots that a human reviewed does not.
You are the one who knows what the role needs, what the question was actually asking and whether a second window was a stack trace or a chat thread. Zavura does not pretend to know any of that.
No score, no rating, no verdict
Zavura does not tell you whether someone cheated. It has no opinion about a candidate and produces no number that could be mistaken for one. It shows you what was on the screen and stops there.
Nothing is rejected automatically
There is no threshold that removes a candidate from your pipeline, because there is no threshold. Every decision stays with the person who ran the interview.
You can explain the decision afterwards
If a candidate, a client or your own legal team asks why someone was passed over, the answer is a timestamped image you looked at, not the output of a model you would have to defend.
Worth saying plainly: a screenshot is a starting point for a conversation, not proof of anything. The right next step is almost always a follow up question, not a rejection.
Privacy posture
Holding less is the point
A session produces a small number of stills rather than hours of video and audio. That is better for the candidate, and it is better for you, because everything you collect is something you have to store, secure, disclose and eventually delete.
Candidates are told what happens before it happens. In our experience that is also the part that makes the deterrent work: most people are not going to run an overlay in an interview where they have just read that the screen gets sampled.
- What is captured
- Stills of the screen the candidate is sharing. Nothing else.
- What is never captured
- Webcam video, microphone audio, keystrokes, clipboard, browsing history.
- When capture runs
- Between session start and session end. There is no background process before or after.
- Candidate disclosure
- Shown before the session begins, in plain language, with consent recorded and timestamped.
- Default retention
- 30 days, then automatic deletion. Configurable down to same day.
- Candidate access
- Candidates can request the frames captured in their own session.
Formal certifications are on the roadmap and not yet held. We will not claim otherwise.
Runs beside the call you are already on
- Google Meet
- Microsoft Teams
- Webex
- Any browser based call
Zavura does not join your meeting, does not appear in the participant list and does not need admin rights on your video conferencing account. These are environments it works alongside, not partners.
Request a demo
See Zavura on the interviews you already run
We will set up a walkthrough on your terms and show you exactly what your team sees: when captures happen, what gets flagged, and what the report looks like.
- Available à la carte or on an enterprise plan
- A direct line to the two people building it, not a ticket queue
- Your interview workflow shapes what gets built next
- Leave whenever you like and take your data with you
Questions people actually ask
Before you put this in front of legal
Sometimes, and we would rather be straight with you than find out during your evaluation. Some overlay tools set an operating system flag that excludes their window from capture, and where that flag holds, a still will not show the overlay. What a still does show is everything around it: the second browser window, the chat thread, the notes document open beside the editor, the reflection in a monitor. Most cheating is not a compositor exploit. It is a second window and a phone.
Yes, before the session starts, in plain language, and they have to agree before anything runs. Covert monitoring is a bad idea legally and a worse idea for your employer brand. It also removes the deterrent, which is the part that does most of the work.
Then it shows their own notes, and you carry on. Zavura does not classify what it captures and does not flag people. It marks frames where more than one window is active so you can look at those first, and that is the extent of the interpretation.
Less than you would expect. Candidates are used to being recorded, and being told exactly what is captured and what is not tends to land better than a vague recording notice. Nothing about the format of your interview changes.
No. There is nothing to install on your machine and nothing to configure in your video conferencing account. If you can send a calendar invite you can run a session.
The stills from the session, their timestamps and the consent record. Default retention is 30 days with automatic deletion, and you can shorten it. No webcam, no audio, no keystrokes, and nothing at all outside the session window.
They would have to know when a capture is coming, and the interval is drawn at random with no indicator. Closing the overlay permanently is, from your point of view, a completely acceptable outcome.
Both à la carte and enterprise plans are available. À la carte is $5 per interview, and enterprise plans bring that lower depending on the plan.