What is Online Proctoring? Types, Features & How it Works

Online proctoring has become a core part of how assessments are run today.

As hiring moved online, organizations needed a way to maintain integrity without controlling the environment. Proctoring filled that gap by bringing structure, oversight, and consistency to remote evaluations.

But the landscape hasn’t stayed the same.

Assessments are now happening at scale, across locations, devices, and time zones. At the same time, candidates have access to tools that can assist them in real time, often without leaving obvious traces.

This creates a new challenge.

In this guide, we’ll break down:

• What online proctoring is and how it works

• The different types of proctoring systems

• Key features that enable monitoring and control

• Where traditional approaches fall short today

• And how evaluation is evolving beyond surface-level signals

If you’re using online proctoring, or thinking about it, this will help you understand not just what it does, but where it fits in a modern evaluation stack.

What is Online Proctoring?

Online proctoring is the process of monitoring and evaluating candidates during an online assessment or interview to ensure the integrity of the interview process.

At its core, it tries to answer a simple question: Is the person being evaluated actually doing the work on their own?

In a traditional setting, this role was handled by a human invigilator in a physical room. Online proctoring replaces that environment with a mix of technology and oversight by using webcams, screen monitoring, audio tracking, and behavioral signals to observe what’s happening remotely.

But that definition only scratches the surface.

Today, online proctoring is no longer limited to exams in classrooms or certification centers. It has expanded into remote hiring processes, where candidate fraud in hiring is becoming harder to detect.

What actually gets monitored?

Depending on the setup, online proctoring systems can track multiple layers of activity during an assessment:

• Video feed: to verify presence and detect suspicious behavior

• Screen activity: to monitor tab switching, external tools, or copy-pasting

• Audio input: to identify conversations or external assistance

• Browser behavior: to restrict navigation or unauthorized access

• Device signals: in some cases, detection of additional screens or devices

Instead of relying on a single signal, modern systems combine these inputs to build a more complete picture of candidate behavior.

From supervision to signal detection

One of the biggest shifts in online proctoring is how monitoring is approached.

Earlier systems focused on continuous supervision, essentially trying to replicate a human watching a candidate at all times.

Newer systems are moving toward signal-based detection, where:

• Behavior is tracked continuously

• Suspicious patterns are flagged automatically

• Only relevant moments are reviewed

This shift matters because scale has changed.
When hundreds or thousands of candidates are being evaluated simultaneously, manual oversight alone doesn’t hold up.

Where online proctoring fits in

Online proctoring now sits at the intersection of three things:

• Assessment integrity - ensuring results can be trusted

• Scalability - evaluating large volumes without manual overhead

• Remote access - enabling candidates to participate from anywhere

For organizations, it acts as a safeguard.
For candidates, it creates a structured environment, even outside a physical room.

Why Online Proctoring Matters Today

Online proctoring has become important because the environment around assessments have completely changed.

A few years ago, most evaluations happened in controlled settings such as classrooms, offices, test centers. The rules were clear, the environment was predictable, and supervision was straightforward.

That’s no longer the case.

Today, assessments happen anywhere, bedrooms, co-working spaces, cafés, across time zones. And that shift has introduced a new layer of complexity that traditional systems weren’t designed for.

The shift to remote evaluation

Hiring and assessments have moved online faster than most teams anticipated.

What used to be:

• In-person interviews

• On-site technical rounds

• Controlled exam environments

Has now become:

• Remote interviews across geographies

• Asynchronous assessments

• High-volume screening processes

This shift brought scale and accessibility.
But it also removed the one thing that made evaluation reliable: control over the environment.

Online proctoring stepped in to fill that gap.

The rise of AI-assisted cheating

Today, a candidate doesn’t need to prepare the same way they did before. They can:

• Generate structured answers in seconds

• Get real-time help during interviews

• Use second devices without being noticed

• Rely on AI tools that listen and respond instantly

This changes the nature of the problem.

It’s no longer about someone trying to look up an answer once or twice.
It’s about continuous external assistance throughout the evaluation.

And most traditional interview formats are not built to detect that.

👉 10 Ways to Prevent AI Cheating in Remote Interviews

The gap between performance and actual ability

As a result, a new kind of gap is emerging:

• Candidates can sound highly competent

• Without necessarily having the underlying depth

Polished answers are easier than ever.
But independent thinking, problem-solving, and real understanding haven’t changed.

This creates a risk for hiring teams, often resulting in bad hires due to AI interview cheating.

• Strong interview performance doesn’t always translate to strong on-the-job performance

• Decision-making becomes harder to trust

• False positives increase

Online proctoring, when done right, helps reduce this gap by adding behavioral context to answers.

Scale is amplifying the problem

Another reason this matters now: volume.

Companies are no longer interviewing a handful of candidates. They’re evaluating:

• Hundreds of applicants per role

• Across multiple stages

• Often simultaneously

At that scale:

• Manual monitoring breaks down

• Inconsistencies increase

• Edge cases get missed

Without structured oversight, it becomes difficult to maintain fairness and integrity across all candidates.

Online proctoring introduces consistency, every candidate is evaluated under a similar set of rules and signals.

The cost of getting it wrong

A weak evaluation process impacts business outcomes.

When integrity breaks down:

• Hiring decisions become unreliable

• High-potential candidates get filtered out

• Underqualified candidates move forward

• Teams spend more time correcting bad hires

Over time, this compounds into:

• Increased hiring costs

• Lower team performance

• Slower execution

Which is why proctoring is becoming part of the core evaluation infrastructure.

Types of Online Proctoring

Not all proctoring works the same way.
The approach you choose depends on what you’re trying to balance such as accuracy, scale, cost, and candidate experience.

Most systems fall into four broad categories. While they often get listed together, the differences between them are significant in practice.

1. Live Proctoring

This is the closest equivalent to a traditional exam setting.

A human proctor monitors candidates in real time through video and audio feeds. They can intervene if something looks off, like ask the candidate to adjust their camera, pause the test, or flag the session.

Where it works well:

• High-stakes exams

• Certification tests

• Low-volume, high-risk evaluations

What you get:

• Direct human judgment

• Immediate intervention when needed

• Higher confidence in edge cases

Where it starts to break:

• Doesn’t scale easily

• Expensive to run across large volumes

• Prone to inconsistency between different proctors

In hiring contexts, this approach is rarely used at scale. It slows things down and adds operational overhead.

2. Automated (AI) Proctoring

This is the most widely used model today, especially for large-scale assessments.

Instead of a human watching every session, the system tracks behavior and flags anything unusual. These flags are based on predefined signals, like tab switching, multiple faces on camera, or suspicious eye movement.

No one is actively monitoring in real time. Everything is recorded, analyzed, and scored.

Where it works well:

• High-volume hiring assessments

• Early-stage screening

• Standardized tests

What you get:

• Scalability across thousands of candidates

• Consistent rule-based monitoring

• Lower operational cost

Limitations to be aware of:

• Context can be missed

• False positives can happen

• Doesn’t always capture more subtle forms of assistance

This is where most modern systems sit but the quality of detection varies a lot depending on how signals are interpreted.

3. Recorded Proctoring (Record & Review)

In this model, sessions are recorded in full, but not actively monitored during the test.

After the assessment, flagged segments are reviewed, either by a human or through additional analysis.

It sits somewhere between live and automated approaches.

Where it works well:

• Medium-stakes assessments

• Situations where auditability matters

• When real-time intervention isn’t necessary

What you get:

• Full session visibility

• Flexibility in review

• Less pressure on real-time infrastructure

Tradeoffs:

• Issues are only caught after the fact

• Review still requires manual effort

• Slower turnaround for final decisions

This model is often used when organizations want a record of the session without committing to live monitoring.

4. Hybrid Proctoring

Hybrid setups combine elements of automated detection with human oversight.

For example:

• AI flags suspicious behavior in real time

• A human proctor reviews or intervenes only when needed

Or:

• Automated monitoring runs throughout

• Human review is triggered for high-risk sessions

Where it works well:

• High-stakes hiring

• Technical interviews

• Scenarios where both scale and accuracy matter

What you get:

• Better balance between efficiency and judgment

• Reduced manual workload

• Higher confidence in flagged cases

Challenges:

• More complex to implement

• Requires coordination between systems and reviewers

• Costs can vary depending on how much human involvement is added

Choosing the right approach isn’t straightforward

There’s no single “best” type of proctoring. Each comes with tradeoffs.

In practice, teams often end up using a mix:

• Automated systems for early-stage filtering

• More controlled setups for later stages

The key is understanding what you’re optimizing for:

• Speed

• Accuracy

• Cost

• Candidate experience

Most of the problems with proctoring don’t come from the technology itself, they come from using the wrong approach for the wrong stage.

How Online Proctoring Works

From the outside, online proctoring can feel like a black box.
A candidate logs in, takes a test, and gets a result.

Behind the scenes, there’s a structured flow that most systems follow. The details vary by platform, but the core steps remain fairly consistent.

Step 1: Identity Verification

Before the assessment begins, the system needs to confirm that the right person is taking the test.

This usually involves a combination of:

• Uploading a government-issued ID

• Capturing a live photo through the webcam

• Matching the live image with the ID

• In some cases, verifying email, phone, or login credentials

Some systems also introduce liveness checks, simple actions like turning your head or blinking, to ensure it’s not a static image being used.

The goal here is straightforward: Make sure the candidate is who they claim to be.

Step 2: Environment Check

Once identity is confirmed, the system looks at the candidate’s surroundings.

This step is often quick, but it plays an important role.

Candidates may be asked to:

• Turn their webcam to show the room

• Adjust lighting or camera angle

• Ensure their face is clearly visible

• Remove any unauthorized materials from the desk

In more controlled setups, the system may also:

• Check for additional screens

• Verify that no other person is present

• Ensure the workspace meets certain guidelines

This step sets the baseline for what’s considered a “clean” environment before the assessment starts.

Step 3: System & Browser Setup

Before the test begins, certain controls are put in place on the candidate’s device.

Depending on the platform, this can include:

• Restricting access to other tabs or applications

• Enabling a secure browser environment

• Disabling copy-paste functions

• Preventing screen sharing or recording

Some tools go further and monitor:

• Running background applications

• Connected devices

• Network behavior

This layer is less visible to the candidate, but it’s critical.
It reduces the chances of obvious forms of external assistance.

Step 4: Real-Time Monitoring

Once the assessment starts, monitoring runs continuously in the background.

Multiple signals are tracked at the same time:

• Video feed: facial presence, movement, multiple faces

• Eye direction: frequent looking away from the screen

• Audio: background voices or unusual sounds

• Screen activity: tab switches, window changes

• Keyboard and mouse patterns: unusual behavior

Individually, these signals don’t mean much.
But when combined, they start to form patterns.

For example:

• Repeated tab switching + long pauses + off-screen glances

• Consistent audio disturbances during key questions

These patterns are what the system pays attention to, often revealing behavioral signs of cheating during remote interviews

Step 5: Behavior Analysis & Flagging

This is where raw data turns into usable signals.

Instead of recording everything blindly, the system identifies moments that stand out. These are flagged based on predefined rules or behavioral models.

Common flags include:

• Multiple faces detected

• Candidate leaving the frame

• Suspicious eye movement patterns

• Attempts to switch tabs or minimize the window

• Unexpected audio activity

Each flag is usually tagged with:

• A timestamp

• A short clip or snapshot

• A severity level

Some systems also assign an overall credibility score based on how the session unfolded.

This makes it easier to quickly understand which sessions need attention.

Step 6: Review & Decision Making

After the assessment ends, the flagged data is reviewed.

Depending on the setup:

• Fully automated systems generate reports directly

• Hybrid systems route flagged sessions for human review

• Live proctoring setups may already have notes from real-time monitoring

Reviewers don’t watch the entire session.
They focus only on the flagged segments, which saves time and keeps the process efficient.

Based on this:

• The session may be cleared

• Marked for further review

• Or flagged as compromised

What actually matters in this process

While the steps look linear, the effectiveness of online proctoring depends on a few deeper factors:

• Signal quality: Are you tracking meaningful behavior or just surface-level activity?

• Context awareness: Can the system distinguish between suspicious and normal behavior?

• Noise vs insight: Are you generating useful flags or overwhelming reviewers with false positives?

Two systems can follow the same steps and still produce very different outcomes.

At a high level, online proctoring is not just about watching candidates.
It’s about collecting signals, identifying patterns, and turning them into decisions.

Key Features of Online Proctoring Software

Not all proctoring tools are built the same. On paper, many of them list similar features but how those features are grouped and used makes a big difference.

Instead of looking at a flat list, it’s more useful to break features into layers based on what they actually do.

1. Identity & Authentication

This is the first layer, making sure the right person is taking the assessment.

Common features include:

• ID verification: Candidates upload a government-issued ID which is matched against their live image

• Facial recognition / face match: Compares the candidate’s face at the start (and sometimes throughout) with the initial capture

• Liveness detection: Prompts simple actions (like blinking or head movement) to prevent spoofing

• Login authentication: Email, OTP, or secure credentials to restrict access

If identity itself isn’t reliable, everything that follows becomes questionable. This layer sets the foundation for trust.

2. Environment & Presence Monitoring

Once identity is confirmed, the focus shifts to the candidate’s surroundings and presence.

Features here include:

• Webcam monitoring: Continuous video feed to ensure the candidate remains visible

• Multiple face detection: Flags if another person appears in the frame

• Candidate absence detection: Detects when the candidate leaves the seat or moves out of view

• Environment scan (pre-check): Room scan before the test begins

• Lighting and visibility checks: Ensures the candidate is clearly visible throughout

A controlled environment reduces obvious risks, like someone else assisting off-camera or stepping in during the test.

3. Screen & Device Monitoring

This layer focuses on what’s happening on the candidate’s system.

Key features include:

• Screen recording: Captures on-screen activity during the assessment

• Tab switching detection: Flags when candidates move away from the test window

• Application monitoring: Detects unauthorized apps running in the background

• Copy-paste restrictions: Prevents content from being copied out or brought in

• Multi-monitor detection: Identifies additional connected screens

A large portion of cheating attempts happen through the device itself, such as switching tabs, referencing material, or using external tools.

4. Browser & Access Controls

These are preventive features designed to limit what candidates can do during the test.

They include:

• Secure / locked-down browser: Restricts navigation outside the test environment

• Disable right-click, shortcuts, and extensions: Reduces ways to access external help

• Full-screen enforcement: Prevents minimizing or hiding the test window

• Session control: Blocks multiple logins or parallel sessions

Instead of detecting issues after they happen, this layer tries to prevent them from happening in the first place.

5. Audio & Behavioral Signals

This is where things move beyond basic monitoring into pattern detection.

Features include:

• Audio monitoring: Detects voices, conversations, or unusual background noise

• Eye movement tracking: Flags repeated off-screen glances

• Head pose detection: Identifies frequent shifts in attention

• Unusual interaction patterns: Long pauses, inconsistent typing, or erratic behavior

These signals don’t prove cheating on their own. But when combined, they help identify patterns that deserve attention.

6. AI-Based Detection & Flagging

This is the layer that ties everything together.

Instead of leaving raw data for manual review, the system highlights what actually matters.

Typical capabilities include:

• Automated flagging of suspicious events

• Timestamped clips for quick review

• Severity levels for each flag

• Session summaries or credibility scores

Some systems also:

• Prioritize high-risk sessions

• Reduce noise by filtering out low-signal events

Without this layer, proctoring becomes unmanageable at scale. With it, reviewers can focus only on the moments that matter.

7. Reporting & Insights

After the session, everything is compiled into a format that teams can act on.

This usually includes:

• Detailed session reports

• Flag summaries

• Video snippets of suspicious activity

• Audit logs for compliance

In hiring contexts, this may also tie into:

• Candidate evaluation workflows

• Decision-making dashboards

Data is only useful if it’s easy to interpret. Good reporting turns raw monitoring into clear signals.

Features alone don’t define effectiveness

Most tools will check off many of these features.

But in practice, what matters more is:

• How accurately signals are captured

• How well noise is filtered out

• How easy it is to act on the output

A system with fewer, well-calibrated signals can outperform one that tracks everything but overwhelms you with irrelevant flags.

At a glance, online proctoring looks like a feature-heavy category.
In reality, it’s about how these features work together to create reliable, usable signals.

Where Online Proctoring Falls Short And What Comes Next

Most proctoring systems were built for a different problem.

They focus on what’s visible:

• Is the candidate on camera?

• Are they switching tabs?

• Is someone else in the room?

That worked when cheating looked obvious.

But today, the problem has shifted.

Candidates don’t need to break rules anymore.
They can stay within the system and still get help.

• AI tools generating answers in real time

• Copilots running silently during interviews

• Second devices sitting just outside the frame

From the outside, everything looks clean.
No tab switches. No visible violations. No interruptions.

And yet, the evaluation is compromised.

The real gap: surface signals vs actual thinking

Traditional proctoring is built on rule-based detection:

• Trigger an alert when something “wrong” happens

• Flag based on predefined conditions

But modern interview fraud doesn’t always trigger those rules.

It shows up differently:

• Answers that are too structured, too consistent

• Delays that don’t match the complexity of the question

• Sudden jumps in clarity or articulation

These aren’t violations.
They’re patterns.

And most systems aren’t designed to read them.

What Sherlock AI actually does

Sherlock AI approaches this problem from a different angle.

Instead of just monitoring activity, it focuses on verifying authorship and reasoning in real time.

During an interview, Sherlock AI:

• Joins the session automatically through calendar integrations

• Observes the interaction using video, audio, and behavioral signals

• Detects signs of AI assistance or external help as the conversation unfolds

• Sends real-time alerts so interviewers don’t have to second-guess what they’re seeing

It’s active analysis, running alongside the interview without disrupting it.

From rule-based flags to behavioral models

One of the key differences is how detection works.

Most systems rely on rules.
Sherlock AI uses a multimodal approach:

• Combines device activity, audio environment, and behavior

• Models what “natural interaction” looks like

• Identifies when behavior has been subtly engineered

Instead of asking:

“Did something suspicious happen?”

It asks:

“Does this look like genuine thinking?”

That shift matters because the hardest cases to detect are the ones that look normal on the surface.

Built for how interviews actually run today

Sherlock AI fits into real workflows:

• Works with tools like Zoom, Google Meet, and Teams

• Integrates with calendars and interview schedules

• Runs in the background while interviews happen normally

Interviewers don’t have to change how they evaluate candidates.
They just get better visibility into what’s actually happening.

Beyond cheating: understanding how candidates use AI

There’s another shift happening.

Some teams don’t want to block AI entirely. They want to understand how candidates use it.

Sherlock AI supports that too.

• Tracks AI usage patterns during interviews

• Evaluates how effectively candidates use tools

• Helps teams distinguish between assistance and actual capability

So the conversation moves from: “Was this allowed?” to “Was this meaningful?”

What this changes for hiring teams

When you move beyond surface monitoring, a few things become clearer:

• Which candidates are genuinely strong

• Which performances don’t fully hold up

• Where deeper evaluation is needed

Instead of relying only on answers, teams get context behind those answers.

And that’s what’s been missing.

Read more: How Sherlock AI Stops Invisible Interview Cheating

Conclusion

Online proctoring started as a way to replicate in-person supervision. Monitor the candidate, control the environment, flag obvious issues.

And for a while, that worked.

But as assessments moved online and candidate behavior evolved, the limits of that approach became clearer.

Today, proctoring still plays an important role:

• It helps establish baseline integrity

• It brings consistency to remote evaluations

• It enables assessments at scale

But it’s no longer sufficient on its own.

Because most modern risks don’t look like traditional cheating. They don’t trigger clear violations or visible disruptions.

They show up in more subtle ways such as how answers are formed, how consistently someone performs, and whether that performance reflects real understanding.

Online proctoring is still a critical piece of the stack. But as interviews and assessments evolve, it needs to be complemented by systems that go deeper.

That’s where solutions like Sherlock AI come in, extending beyond surface-level monitoring to help teams understand not just what happened during an evaluation, but how it happened.

Because in the end, the goal isn’t just to run assessments. It’s to trust the decisions that come out of them.