Key takeaways
- An interviewer gets three things from you: your camera, your microphone and your shared screen. Nothing else on your computer reaches them.
- Windows can exclude a window from screen capture completely, so it never appears in a share.
- Most AI use is spotted by people, not software: reading eyes, typing, pauses, a scripted tone and stories that fall apart on the follow-up.
- Coding pads such as CoderPad and HackerRank log pastes from outside the editor and time spent away from it. Vendors such as BrightHire and Fabric sell AI-cheating flags.
- Short notes you speak from in your own words show far fewer tells than a script you read.
Can interviewers detect AI?
Yes, often, but not the way most candidates fear. The worry is usually technical: can Zoom see my other apps, can the interviewer see my second window? The answer that matters more is human. Experienced interviewers have sat through a lot of calls. They know what thinking sounds like, and what reading sounds like.
So split the question in two:
- What can they technically see? Only what your camera, microphone and screen share send them, plus any activity a browser-based coding pad logs.
- What can they notice? Everything about how you deliver the answer: your eyes, your timing, your words, and whether you can go one level deeper.
The second is where AI use usually shows.
What can interviewers see on a video call?
A video interview is a narrow pipe. Here is what goes through it, and what doesn't.
| Channel | What the interviewer gets | What they don't get |
|---|---|---|
| Camera | Your face, eyes, posture, background and anything reflected in your glasses | Your screen, unless it's reflected |
| Microphone | Your voice, pauses, keyboard sounds, other voices in the room | Audio that plays only in your headphones |
| Screen share, one window | That window only | Every other window and app |
| Screen share, entire screen | Everything on that screen that is captured | Windows excluded from capture, other monitors you don't share |
| Browser coding pad | What you type, pastes from outside the editor, time away from the pad (on some platforms) | Which app or tab you went to |
| Your computer | Nothing directly in a normal video call | Running apps, processes, files |
Screen share shows only what is captured
When you share, the meeting app sends a captured image of your screen or window. A window that the operating system leaves out of capture isn't in that image. Microsoft documents this for Windows: with the WDA_EXCLUDEFROMCAPTURE setting, "The window is displayed only on a monitor. Everywhere else, the window does not appear at all." Microsoft's example use is a window of video recording controls, and the setting is supported from Windows 10 version 2004. macOS has a similar per-window content protection setting.
So an interviewer can't see a window that is excluded from capture, even when you share your entire screen. For how each meeting app handles this, read our posts on Zoom, Google Meet and Microsoft Teams.
Your camera and microphone give more away
The camera and microphone are the channels that matter. They carry your eyes, your timing and your keyboard. A hidden window doesn't hide the person looking at it.
Coding pads log what happens in the editor
Live coding rounds often run in a shared browser editor, and some log what happens in it. CoderPad's playback documentation says pastes from outside the pad "are highlighted to assist you in detecting potential cheating", and orange dots mark moments when the candidate's browser was no longer focused on the editor, with labels such as "Clicked away for 19 seconds". HackerRank's interview integrity signals, which an admin can turn on, flag switching to another browser tab, pasting from an external source, resizing the window and, in Chrome with display access, a multi-monitor setup. HackerRank's own article notes that copying and pasting code doesn't always indicate misconduct.
Note what these tools record: that you left the editor, not where you went.
How do interviewers tell if you're using AI?
They watch and listen. People Matters, citing internal Amazon interview guidelines seen by Business Insider, listed three signs Amazon's interviewers were told to look for: "typing while being asked questions", "wandering eyes that suggest screen-switching" and "overly rehearsed answers". Those three, plus a few more, are the signs to know.
Your eyes move like you're reading
Reading has a rhythm: eyes sweep sideways at a steady pace, drop a line, sweep again. Thinking looks different. People look up, away, or at nothing in particular. On a webcam, a steady sideways scan is easy to spot, especially when the eyes stop at the camera only at the end of a sentence.
You type while they're talking
Keyboard sounds during a question, or shoulders moving while you "listen", suggest you are entering the question somewhere. This is the classic tell of using a chatbot in another window.
A pause, then a perfect list
Fabric, which sells an AI interview platform with built-in cheating detection, says it listens for long pauses before polished answers, calling that "a common sign of someone consulting an external AI tool". It also lists silence followed by a list-like answer, an inconsistent tone and generic explanations. Real answers start a bit rough and get sharper; AI-fed answers often start late and arrive fully formed.
The answer sounds rehearsed or generic
Written English sounds different from spoken English. "There are several key considerations here, firstly…" is how a document reads, not how most people talk. Generic answers are the other half: correct, balanced and about nobody in particular.
The answer goes beyond your resume
If a three-year Java developer suddenly talks fluently about a Rust service they've never listed, the interviewer notices. Answers that drift beyond your resume are a signal on their own, and they invite exactly the question you can't answer.
The follow-up falls apart
This is the test that catches most people. CNBC reported in March 2025 that Google's interviewers are instructed to probe candidates on their answers. "Why that approach?", "What would you change?" and "How did you measure that?" are easy if the story is yours, and very hard if it came from a screen a minute ago.
Signals interviewers notice, and how to avoid them
Most of these fixes are also just good interviewing. They make you clearer whether or not you use any tool.
| Signal | What interviewers notice | How to avoid it |
|---|---|---|
| Reading eyes | A steady sideways scan, eyes off the camera for whole sentences | Glance at one line, then look at the camera and talk. Keep notes small and near the camera |
| Typing during questions | Keyboard noise or moving shoulders while they speak | Don't type questions in. Use something that listens to the call, or nothing |
| Pause, then a perfect list | Silence, then a fully formed, numbered answer | Start with a true first sentence straight away, then build the detail as you go |
| Scripted tone | Written-style phrases, flat delivery, no "um" or self-correction | Speak from points, not sentences. Rephrase everything in your own words |
| Answers beyond your resume | Tools, numbers or projects that aren't on your CV | Prepare only from your own experience. Drop any detail that isn't yours |
| Weak follow-ups | Long pauses or vague answers to "why?" and "how did you measure it?" | Rehearse the reasons behind each project before the day |
| Code that appears all at once | A large block pasted in, no thinking aloud | Type your solution yourself and explain each step as you write it |
| Leaving the editor | "Clicked away" marks or tab-switch alerts on the pad | Keep the editor in focus during coding rounds. Close everything you don't need |
The common thread: the more an answer sounds like you thinking, the fewer tells there are. A tool that gives you a paragraph to read creates most of the signals in this table. One that gives you a first line and a few points creates very few.
What AI detection tools do companies use?
A growing number of hiring platforms sell detection. We only list what each company's own pages say, checked on 11 November 2026.
Zoom and BrightHire
On 24 June 2026, Zoom's blog announced that BrightHire would flag fraud indicators including "deepfakes and AI-assisted cheating in interviews", with deepfake detection powered by Zoom's proprietary models. BrightHire's own post says it pulls "signals across behavior, identity, device, and context" into one view, with each signal linked to reviewable evidence. It is in early access for Zoom Workplace customers with an active BrightHire subscription. Neither page explains how AI-assisted answers are identified.
Fabric
Fabric describes itself as an AI interview platform with a built-in layer for detecting AI-assisted or dishonest responses. It says it relies on response timing, tone, phrasing and answer depth rather than gaze tracking or screen sharing.
CoderPad and HackerRank
As above, both log activity inside the coding environment: external pastes and time away from the editor on CoderPad, and tab switches, pastes, window resizing and multiple monitors on HackerRank when the employer turns the signals on.
What the tools have in common
None of them say they read your computer's other apps. They score what comes through the call or the editor: your face, voice, timing and keystrokes, the same evidence a sharp interviewer uses. HackerRank says it alerts only on meaningful clusters of suspicious activity: one glance away proves nothing, and patterns get flagged.
Can recruiters detect AI-written answers?
In a spoken interview there is no text to run through a detector. Recruiters judge how the answer sounds and what happens when they push on it. Vendors such as Fabric say they analyse phrasing and timing, but the strongest check is still the follow-up question. If you can explain every claim in your answer, it reads as yours, because it is.
Being hard to detect doesn't change what a round allows, though. Where an employer bans AI in interviews, using it is against the rules whether anyone notices or not. Our post on whether using AI in an interview is cheating goes through the rules round by round.
Why Helpwing
Helpwing is designed to keep the tells in the table above to a minimum, on rounds where you use it:
- No typing. It hears the interviewer through system audio, so you never type a question during the call.
- Off your shared screen. The overlay is 100% undetectable on screen share in Zoom, Google Meet and Microsoft Teams. On Windows the window is excluded from capture with
WDA_EXCLUDEFROMCAPTURE; on macOS it uses window content protection. It stays off the taskbar and the Dock, and no bot joins the call. - Fast enough to avoid the long pause. A short answer appears in the overlay in seconds.
- A shape you speak from, not read. Spoken questions get a 1–3 sentence "Say this" line plus 3–5 key points. You glance at the first line and talk in your own words, which avoids reading eyes and a scripted tone.
- Answers inside your resume. Tune Interview takes your resume, the company and the job description, so answers stay in your own experience. That cuts off the "sounds generic" and "beyond your resume" tells, and makes follow-ups answerable.
- Clear limits and simple pricing. Not for exams, online assessments or proctored tests. 15 free minutes, then pay-once packs from ₹299 for 60 minutes ($6 outside India). No subscription. macOS 14.2+ and Windows 10/11.
Here is the shape for a question a service-to-product switcher hears in almost every round. The details are illustrative.
Say this: I've spent four years building features to a client's spec, and I want to own a product end to end, including the metrics after it ships.
Key points: led the payments module for a banking client, but never saw usage data · started tracking error rates myself and cut failed retries · your team ships weekly and owns its dashboards · I want my next review to be about outcomes, not tickets closed.
You'd say the first line, then pick two points and tell them as your own story. There is no paragraph to read, so there is no reading rhythm for the camera to catch.
Weaker
The candidate reads a 120-word answer from a chatbot window, eyes moving across the screen, after typing the question in while the interviewer was still speaking. The follow-up, "What did the error rate go from and to?", gets a pause.
Stronger
The candidate glances at "Say this", looks back at the camera and talks through their own payments story. The follow-up is easy, because the numbers came from their own resume and notes.
How Helpwing helps in this round
Helpwing removes the tells that come from the tool itself: it listens through system audio so you never type, stays off your shared screen in Zoom, Meet and Teams, and gives you a short "Say this" line and 3–5 key points drawn from your own resume, so you speak instead of read. It runs on Mac and Windows, starts with 15 free minutes, then pay-once packs from ₹299. It is never for exams, online assessments or proctored tests.
Questions people ask
Can interviewers tell if you use AI?
Often, yes, through how you answer rather than through your computer. Reading eyes, typing while they talk, long pauses before polished answers and stories you can't expand on are the usual giveaways.
Can Zoom, Google Meet or Teams detect an AI assistant?
The screen share only shows what is captured, so a window excluded from capture doesn't appear in it. We cover each app separately in our posts on Zoom, Google Meet and Microsoft Teams.
Can interviewers see my other tabs or windows?
Only if you share them. If you share one window, they see that window. If you share your entire screen, they see what is captured on it. Browser coding pads can tell that you left the editor, but not which tab or app you went to.
Can recruiters detect AI-written answers in a live interview?
Not with a text detector, because a spoken answer isn't text they can paste anywhere. They judge tone, timing and detail, and they ask follow-ups. Some platforms, such as Fabric, say they also score timing and phrasing.
Do companies use AI detection software in interviews?
Some do. BrightHire offers AI-assisted cheating and deepfake flags for Zoom interviews in early access, Fabric builds detection into its AI interview platform, and coding pads like CoderPad and HackerRank log pastes and time away from the editor.
Can an interviewer tell if I'm reading from notes?
Usually. Reading produces a steady sideways eye movement and a flat, written-sounding delivery. Speaking from a few short points looks and sounds like normal thinking.
Does using a second monitor give it away?
It can. Your eyes turn away from the camera for long stretches, which interviewers notice. HackerRank's interview signals can also flag a multi-monitor setup in Chrome when you grant display access.
If they can't detect it, is it allowed?
No. Detection and permission are separate questions. If the employer bans AI in that round, using it breaks the rule whether or not anyone notices. Read is using AI in an interview cheating for the round-by-round rules.
Sources
- SetWindowDisplayAffinity function (winuser.h), Microsoft Learn, checked 11 November 2026
- Playback & Pad Summary, CoderPad Interview Docs, checked 11 November 2026
- Interview Integrity Signals, HackerRank Knowledge Base, checked 11 November 2026
- Zoom and BrightHire take on rising candidate fraud, including deepfake detection built into live interviews, Zoom Blog (24 June 2026), checked 11 November 2026
- A New Way to Detect Candidate Fraud in Your Interviews, BrightHire (24 June 2026), checked 11 November 2026
- Top AI Tools that Detect Cheating in Live Interviews, Fabric (14 November 2025), checked 11 November 2026
- Is it OK to use AI during a job interview?, People Matters (7 May 2025), checked 11 November 2026
- Meet the 21-year-old helping coders use AI to cheat in Google and other tech job interviews, CNBC (9 March 2025), checked 11 November 2026