Can AI tell if you cheat or use notes in a video interview?

No AI scans your desk for notes. A standard one-way interview records your answer and sends it to a person. A small set of proctored tools flag patterns linked to cheating, like a steady off-screen gaze, a second face, or tab switching. None of that targets an honest candidate glancing at a bullet point.

The fear is understandable. You get a recorded interview link, you read that it is “AI-powered,” and your mind jumps to a webcam that follows your eyes, counts how often you glance down, and silently builds a cheating score. That picture is mostly myth. The reality is a lot more boring, and a lot more reassuring, once you separate what these tools actually do from what the word “AI” makes people imagine.

This page walks through it honestly. What detection genuinely flags, what it cannot see, where proctoring is real and where it is not, and the simple way to stay on the right side of all of it. If your specific worry is notes, the companion page on whether you can use notes in a one-way interview goes deeper on that one question.

First, what most video interviews actually do

Start with the common case, because it covers the majority of recorded interviews you will ever take.

A typical one-way interview shows you a question, gives you a short window to think and record, and uploads the file. A hiring team watches it later. That is the whole loop. The “AI” in most of these tools does two unglamorous jobs: it transcribes what you said so a recruiter can skim it, and it helps organize and compare answers across candidates. It surfaces. A person decides.

In that common case, there is no proctor. Nothing is tracking your eyes. Nothing knows whether you have a notepad beside the keyboard. The company has no way to see the sticky note on your monitor, because the only thing it receives is the video you chose to submit. This is the single most important fact on this page, and it is true far more often than the scary version.

So the honest answer to “can AI tell if I use notes” starts here: in a standard one-way interview, usually not at all. The detection people fear is a feature of a minority of tools, not a default of the format.

Where real proctoring does exist

Now the exceptions, because they are real and you deserve the straight version.

Some platforms add proctoring and fraud detection as features an employer can switch on. Jobma is the best-known example, and it is known specifically for this. When those features are enabled, the session can do more than record. Depending on the setup, it may:

• Verify your identity before you begin, so the person answering matches the person who applied.
• Watch for a second face in frame, or for the face changing partway through.
• Notice sustained off-screen attention, meaning your gaze parked off-camera for long stretches, not a quick glance.
• Flag tab switching or leaving the window during a recording.
• Mark a session for a human to review more closely, rather than auto-rejecting anyone.

Read that list carefully, because the theme is consistent. Every one of those signals is aimed at the same thing: confirming that the recording is really you, doing your own thinking, in real time. None of it is aimed at the honest candidate who glanced at a note to remember a date.

Why it exists is not a mystery. Interview fraud is a genuine employer problem. The classic versions are one person sitting the interview while a different person shows up to the job, or someone reading a fully scripted answer off a second screen. More recently, recruiters have raised alarms about deepfake video filters in live interviews, with some moving to third-party identity verification in response. Proctoring is the employer’s attempt to trust what they are looking at. It is built to catch the small number of people gaming the process, and for everyone else it fades into the background.

What detection can actually flag, plainly

Strip away the mystique and there are really only a handful of things any system, AI or human, can notice. Here they are, honestly.

Your eyes, when they read. This is the big one, and it does not need fancy software. Reading aloud changes you. Your eyes move in lines, left to right, and your delivery flattens into a recital. A recruiter who reviews these interviews put it bluntly: “you can literally tell if someone is reading an answer to you.” A proctoring tool may quantify that as sustained off-screen gaze. A human reviewer just feels it. Either way, reading a full script is the most detectable thing you can do, and it is detectable precisely because it changes your behavior, not because a camera is psychic.

A second face or a voice that is not yours. If someone else is feeding you answers in frame or off-mic, that is the cheating proctoring is designed to find. It is the core fraud case.

Your identity not matching. Where identity verification is on, the system checks that the person answering is the person who applied. This is an identity check, not a judgment of how you look or perform.

Over-polished, generic language in the transcript. A scripted answer reads as scripted. It is safe, smooth, and forgettable, and it often matches phrasing the reviewer has seen ten times that day. This is not “detection” so much as a tell that any experienced reviewer picks up.

That is close to the full list. Now the other side.

What it cannot see, and the myths to drop

Just as important is what these tools do not do, because the myths cause more harm than the reality.

• It cannot see your desk. No standard tool reads a paper note, a second monitor’s content, or the cheat sheet taped beside your camera. It only has the video feed.
• A glance up to think is not a flag. Looking away briefly to gather a thought is normal, and reviewers expect it. Proctoring is tuned for sustained or repeated off-camera attention, not a natural pause. Do not freeze your face trying to avoid blinking the wrong way.
• It is almost never judging your face for scoring. The idea that AI scores your micro-expressions or “confidence” from your face is largely outdated. Facial analysis for scoring has been stepped back across the industry. HireVue, one of the largest vendors, discontinued facial analysis in 2021. For the full version of this, see do AI interviews use facial recognition.
• A flag is not a verdict. Where proctoring exists, an unusual signal typically routes the interview to a person, not straight to a rejection. The AI raises a hand. A human looks.

The reason the myths persist is that “AI interview” sounds like surveillance. In practice, the format is closer to a video drop-box with a transcript than to a lie detector. If you are unsure whether your invite is even an AI-driven interview at all, here is how to tell.

The honest candidate’s playbook

Here is the practical part. Because the same behavior keeps you clear whether or not any detection is running, you do not have to guess which kind of interview you have.

1. Look at the lens, not a screen of notes. Talking to the camera dot reads as eye contact and, not coincidentally, looks nothing like reading. This single habit covers most of what people worry about. More on where to look during a video interview.
2. Use notes as prompts, never a script. A few keywords beside the camera steady your nerves and are completely fine. A full paragraph you read aloud is what flattens your delivery and moves your eyes in lines. Keep prompts to single words or short phrases.
3. Answer in a quiet room, alone. No second person feeding you lines, no second screen you are reading from. This is the behavior fraud detection is built to find, and the one thing worth genuinely avoiding.
4. Be yourself on identity checks. If the tool verifies who you are, that is routine. You are the person who applied. Good light and a steady frame make it pass cleanly.
5. Speak in specifics. Name the project, the tool, the number. Concrete, lived detail is the opposite of a generic script, and it reads as credible to a person and a transcript alike.

Do those five things and you are not gaming anything. You are just doing the interview honestly, which is exactly what every version of this format, proctored or not, is built to reward.

So, can AI catch you?

The fair summary: AI, and the human watching with it, can notice when you are clearly reading a script, when a second person is involved, or when your identity does not match. It cannot see a glance at a bullet point, it is not scoring your face, and most one-way interviews are not watching for any of this at all.

That means the cheating fear and the honest-prep question have the same answer. Refer to a few notes, keep your eyes on the lens, and talk like a person. Worry less about being caught and more about being clear, because clarity is what actually moves a reviewer.

If a proctored tool is what you are facing, the Jobma candidate guide covers what those checks mean in practice. To prepare for the answers themselves, read how to prepare for an AI interview and how to pass a one-way video interview.