How does online proctoring work?

Who can benefit from online proctoring? Anyone delivering a remote assessment — a university running finals, a certification body issuing a license, an employer testing new hires, a government agency training staff. All of them face the same problem: how do you supervise an exam when the test-taker isn't in the room?

This guide walks through how online proctoring actually works, using Constructor Proctor — an AI-powered proctoring platform — as the working example: the technology it runs on, the steps a candidate goes through, what gets flagged as a violation, and the different modes institutions can choose between.

What is online proctoring?

Online proctoring is the practice of supervising a remote exam using software instead of, or alongside, a person in the room. It combines identity verification, live monitoring through a webcam and microphone, and desktop activity tracking to reproduce what an in-person invigilator would catch, without requiring anyone to be physically present.

It's used across a wide range of settings:

• Universities and schools — entrance exams, term exams, remote finals.
• Certification bodies and ministries — licensing and national exams.
• Employers — pre-hire assessments and internal certifications.
• Government training programs — qualifying exams for public-sector roles.

How online proctoring works, step by step

The details vary by provider, but most systems, Constructor Proctor included, follow the same sequence from login to final report.

Getting started: browser-based access, no installs, LMS integration

Proctoring runs as a browser extension or in-browser session, with no software installation required. It's delivered as SaaS, with the exam infrastructure hosted in the cloud and integrated directly into the institution's learning management system (LMS). For the candidate, this means opening a browser, following the link provided, passing the system and identity checks, and starting the exam — nothing to download beforehand.

System checks: camera, microphone, internet, second-screen detection

Before the exam opens, the system requests access to the camera and microphone and confirms both work, along with checking the stability of the internet connection. It also checks for a second monitor and for any attempt to simulate or spoof the screen. A candidate who fails one of these checks is prompted to fix the issue before the exam can begin.

Identity verification: photo ID and face match

Once the system checks pass, the candidate confirms their identity by photographing themselves alongside a form of ID — a driver's license, passport, or student ID card — which the system matches against their face. Only after this match succeeds does the exam unlock.

During the exam: combined data from camera, microphone, and desktop

From the moment the exam starts, the system pulls in data from three sources at once: the camera feed, the microphone, and activity on the desktop itself — which windows are open, what's been clicked, whether the candidate has switched tabs.

After the exam: consolidated session record and review

Once the candidate submits, the system merges all three data streams — camera, microphone, and desktop — into a single session record. That record becomes the basis for everything that happens next: automated scoring, human review, or both.

The online proctoring technology behind the scenes

Underneath the candidate-facing browser session sits a stack of AI models trained to recognize specific behaviors: gaze direction, the number of faces in frame, voice activity, and patterns in how windows and tabs are used. Because the whole platform runs as SaaS, none of this depends on what's installed on the candidate's machine — the models run against the video, audio, and desktop-activity streams as they arrive.

Rather than presenting every anomaly with equal weight, most systems reduce a session to a single risk score. Constructor Proctor sorts each session into a green, yellow, or red zone based on the overall probability of a violation — green for low probability, yellow for medium, red for high — so a review team knows immediately which candidates to check first instead of working through every session in order.

For any flagged session, reviewers can go back and watch or listen to the camera, microphone, and desktop recordings together, step through every flagged moment in chronological order, and check exactly what happened at that timestamp before making a final call.

What online proctoring flags as a violation

Exactly what counts as a violation is configurable — a host might allow a calculator in one exam and require a clear desk in another — but most systems flag two broad categories of behavior.

Automatically detected by AI, without a human proctor involved:

• Changes in gaze direction — the candidate looking away from the screen.
• The face missing from the frame, or not fully visible.
• Other people appearing in frame (even a portrait on the wall can trigger this, so removing them beforehand is worth doing).
• Candidate substitution — the person in frame doesn't match the registered candidate.
• Any human voice, including the candidate's own, such as reading questions aloud.
• Switching windows or tabs, such as leaving the exam window for a search engine or another application.

When a live human proctor is also monitoring the session, additional violations can be reported that automated detection alone tends to miss:

• The use of phones, tablets, or other devices.
• Remote-access or screen-sharing software such as TeamViewer, Skype, or RemoteAdmin.
• Virtual machines or thin clients used to let a third party sit the exam.
• Books, notes, or drafts.
• Calculators, where prohibited.
• Non-verbal communication — winks, hand signals, nods — directed at someone off camera.

Every flagged entry is logged with the type of violation and a timestamp, feeding into the session record described above.

The four modes of online proctoring

Institutions don't have to pick one proctoring style for every exam — most platforms offer four modes, trading off cost, speed, and how much human judgment is involved.

• Automatic (AI-only): everything is handled by AI, with no human proctor involved. It's the cheapest mode to run, but it has real limits: high-stakes exams generally shouldn't rely on a machine for the final call, subtle violations can slip past even an accurate detection model, ordinary interruptions — a parent opening a door for a few seconds — can trigger false flags, and some exams require a human to walk a candidate through a specific protocol, such as checking their hands are empty or that scratch paper gets torn up afterward, which automation can't perform.
• Asynchronous: a human proctor reviews the recorded session after the exam has finished, checking flagged moments and confirming or dismissing them.
• Synchronous: a human proctor monitors the exam live and can intervene in real time as it happens.
• Identification-only: a human proctor's only role is confirming the candidate's identity before they're let into the exam; everything after that runs on AI.

Each mode has its own interface and workflow for the proctors running it.

Computer vs. phone: what devices work for online proctoring?

Desktop and laptop computers are the only devices that produce reliable results, and most providers, including Constructor Proctor, recommend them for that reason — a mobile device gives a candidate far more room to work around the system. The exception is field or remote workers who genuinely have no access to a computer, such as staff based at a remote plant or facility, where a mobile setup may be the only realistic option.

How to choose online proctoring technology

• Configurable rules — can you set different permitted materials, like a calculator or notes, per exam rather than one fixed policy for everyone?
• Violation coverage — does it flag both automated signals (gaze, voice, tab switching) and the things only a live proctor tends to catch (remote-access software, virtual machines, physical notes)?
• Proctoring modes — can you choose between automatic, asynchronous, synchronous, and identification-only depending on how high-stakes the exam is?
• Review tools — does it give reviewers a risk score plus synchronized camera, microphone, and desktop playback, or just a flat list of alerts?
• Device flexibility — does it support desktop as the default while still accommodating candidates who can only access a phone?
• LMS and SaaS integration — does it plug into your existing learning management system without requiring installs for candidates?
• Data protection — is it clear what's collected, how long it's kept, and who can access it?

How Constructor Proctor fits in

Constructor Proctor runs as the SaaS platform described throughout this guide: browser-based, with no installs, integrated into more than 30 LMS platforms, and built to scale from a handful of candidates to over 10,000 simultaneous sessions.

On the technology side, it monitors 100+ behavioral and technical parameters, reports around 90% accuracy in detecting violations, and lets a single live proctor oversee up to 150 candidates at once through its dispatching interface, with device detection and secure-browser lockdown covering much of the automated side.

Princess Nourah University's English Language Institute switched from paper-based exams to a digital system built on Constructor Proctor's AI-driven proctoring. The result: an 85% increase in student engagement, a 70% reduction in administrative workload, and a 90% drop in academic dishonesty.

You can read more about Constructor Proctor here.