How to Conduct Online Exam Without Cheating Fairly

How to conduct online exam without cheating comes down to three things: a paper that cannot be easily searched or outsourced, controlled access for each candidate, and human review of any suspicious session. No online system is fully cheat-proof. A sensible process makes misconduct harder, creates useful evidence and avoids punishing students for ordinary technical problems.

Use lighter controls for a weekly practice test. For a final examination, admission test or hiring decision, add identity checks, recording, a clear exception process and a short viva when there is a credible concern. Match the control to the consequence of the result.

How to conduct online exam without cheating: use layers, not one tool

A webcam cannot show a helper sitting outside the frame. Browser monitoring cannot detect every mirrored screen or undisclosed phone. Be honest about those limits in your exam policy instead of claiming that software can see everything.

Use several controls that cover different risks. Individual access codes reduce impersonation. Randomised questions make answer-sharing less useful. Proctoring creates a record. Human review stops an alert becoming an automatic accusation.

Looking away, a voice in the room, a tab switch or a second-monitor alert is a reason to inspect the session. It is not proof of cheating. A student may be thinking, dealing with an interruption or responding to a browser permission prompt.

Set the review process before the exam starts. Name the person who reviews flags, define when a second reviewer is needed, and give candidates a way to explain a concern. This matters as much as the technology.

If you are revising a formal examination policy, check the current position against official UGC guidance and the Public Examinations (Prevention of Unfair Means) Act, 2024. Their scope differs, so neither should be copied as a rulebook for every internal assessment. 

Design a paper that is difficult to search or pass to ChatGPT

A fixed paper full of direct recall questions is easy to circulate, search online or put into an AI tool. Start with the assessment design before buying more monitoring.

Build a balanced question bank

Create more items than each candidate will receive. Draw questions randomly from the bank, and randomise both question order and answer options. Students can be tested on the same learning outcomes without seeing the same paper in the same order.

Check the bank for comparable difficulty. Random selection is not fair if one version contains several demanding calculations and another has none.

Set a time limit after a pilot

Give a prepared student enough time to read, think and submit. Do not leave enough spare time for searching each answer, messaging a helper or waiting for someone else to solve the paper.

Run the paper with teachers or a small pilot group at the expected ability level. Use their completion time to set the duration, then test the timer and submission flow on the actual devices candidates will use.

Ask for judgement, not copied definitions

AI assistants respond well to simple factual prompts. Ask candidates to interpret supplied data, correct a flawed working, choose between two practical actions or explain why an answer fits a case.

For example, give commerce students a small business payment schedule and ask which action would ease a short cash shortfall. Give science students an experimental result and ask them to identify one limitation in the method. The candidate must apply knowledge, not repeat a definition.

For descriptive answers, ask for workings and references to material included in the paper. Compare a doubtful response with earlier supervised work where appropriate. For an important result, a short follow-up viva can be more useful than watching a long recording.

State what resources are allowed

Decide whether the paper is closed book, open book or open web before you write it. Put that decision in the instructions and design the questions to fit. An open-book paper should assess interpretation and decisions, not facts available in a quick search.

For a closed-book exam, expressly ban ChatGPT and other AI assistants, web search, messaging apps, notes, unapproved headphones, helpers, second screens and mirrored displays. “No cheating” is too vague for consistent enforcement.

Lock down access, attempts and timing before sharing the link

Most avoidable failures begin in the settings. Have one person configure the assessment and another person check it before candidates receive their links.

Use a stable candidate identifier

Use an admission number, employee number or application number alongside the candidate’s name. A typed name is not an identity check. It can be misspelt, shared or entered differently on a second attempt.

Give every candidate an individual single-use code or an authenticated login. This is especially useful where siblings use one family phone or candidates register with a parent’s mobile number. In October 2026 assessments run on RootPro for Business, single-use codes stopped candidates from entering again under a slightly altered name.

Set one attempt and check the setting twice

Set the maximum attempts to one unless an authorised exception is recorded. In one October 2026 assessment, unlimited attempts were left on by mistake and several candidates retook the paper. A second-person check of the maximum-attempts setting would have stopped it.

Keep a one-page configuration record for each exam. Include the opening and closing window, duration, attempt limit, question randomisation, identity checks, allowed resources, devices, support contact and the person who can approve an exception.

Make the server enforce the deadline

The platform should enforce the time limit on its server, not rely on the clock in a browser or phone. In RootPro for Business, answers arriving after time expires are refused; a ten-minute grace period applies only to an answer delayed by a dropped connection, and unsubmitted papers are closed and marked automatically.

Test this behaviour before announcing the rules. A grace period for a delayed submission is not extra writing time after a candidate loses connection.

Pilot the full candidate flow

Do not test only on office Wi-Fi and a staff laptop. Try common Android phones, laptops and mobile networks. Check camera permission, photo ID, the timer, submission, the reviewer screen and the support process from start to finish.

A RootPro load test ran 1,100 simultaneous candidates with camera and session recording enabled and recorded zero failures. Your own pilot still matters because your students’ devices, networks and question types may be different.

Give candidates clear instructions before exam day

Send a single instruction note well before the exam and ask candidates to test their device and camera where possible. Say what they need: a charged device, reliable connection or backup data where available, a quiet well-lit space, front camera access and photo ID if you require it.

List permitted stationery, calculators, formula sheets and reference material. Also list what is not allowed. Do not leave invigilators to interpret rules differently while the exam is running.

Explain that recordings and alerts are reviewed by a person, that a candidate may be asked for an explanation or viva, and that decisions are based on the available evidence. This reassures honest students and makes your process easier to defend.

Pre-exam message for candidates

Your [exam name] is on [date] at [time]. Use only your own access code. Sit alone in a well-lit place, keep your face visible and keep any phone or second screen away unless the instructions allow it. This is a closed-book exam: do not use ChatGPT, web search, messaging, notes or outside help. Keep [photo ID] ready if requested. If you face a technical problem, contact [support contact] immediately. Suspicious activity is reviewed by a person, and you may be asked to explain it or attend a short viva.

Send the same note to invigilators, subject staff and technical support. Candidates lose trust quickly when one staff member allows a retry and another refuses it.

Run the exam with support and controlled exceptions

Keep a staffed channel open for login, camera and connection issues. Technical support can help a candidate enter the assessment. It must not explain a question, confirm an answer or grant academic extra time without authorisation.

Use photo ID or face checks, webcam recording and browser or session monitoring when the assessment warrants them. Before recording anyone, state why you are collecting the data, who can access it, how long it will be kept and how it will be deleted.

Review patterns rather than isolated events: repeated exits, long unexplained pauses, voices in the room, persistent attention away from the screen, tab changes, unusually rapid answer changes and attempts to reconnect under another identity. Several concerns together deserve closer examination.

Written rules alone cannot detect a helper. In one school exam in October 2026, a student kept facing the webcam while a parent sat outside the frame and supplied answers. Webcam recording and photo review exposed the pattern. For a high-stakes exam, add a second camera placed on a phone or require a follow-up viva for credible concerns.

Do not give automatic retries

Once a paper starts, do not offer a free retry automatically because questions may already have reached the browser. Let an authorised organiser approve an additional attempt only after recording the reason and checking the available session evidence.

Weak mobile internet needs a fair route. Candidates can get stuck at photo ID and lose time before the paper begins. Allow the organiser to permit one more verification attempt with a recorded reason. That is not the same as giving an unrestricted second exam attempt.

Log every support request, time adjustment, accommodation and retry decision. The record protects candidates with genuine problems and helps staff treat similar cases consistently.

Review evidence before deciding misconduct

Do not begin with a score and search for a reason to disqualify someone. Start with the evidence. Put the identity check, relevant recording clips, session events, answers, timestamps and attempt history in one review queue.

Use the same review questions for every case

• What happened, and when did it happen?
• How long did it continue?
• What evidence supports the concern?
• Does the answer pattern strengthen or weaken the concern?
• What explanation did the candidate provide?
• Who reviewed the case and made the decision?

A flag is context, not a verdict. A student may look away to read an allowed formula sheet. A tab switch may follow a browser prompt or connection problem. Each proctoring flag should go to a person for review.

Escalate serious cases to a second reviewer. Where the outcome matters, invite the candidate to respond and use a short viva to test whether they can explain their submitted answers. Review exceptions with the same care: similar technical problems should receive similar treatment.

After results are released, inspect identical answer sequences, abrupt unexplained score changes, possible question leakage and items that were too easy to answer through search or generative AI. Improve the next paper rather than simply adding more surveillance.

Printable online exam integrity checklist

Before publishing

• Confirm the candidate roster and stable identifiers.
• Issue unique single-use codes or authenticated logins.
• Set maximum attempts to one and obtain a second-person check.
• Confirm the exam window, duration and server-enforced timing.
• Turn on suitable question and option randomisation.
• Document allowed resources, devices, camera requirements and AI rules.
• Check proctoring settings, privacy notice, retention period and support contact.
• Complete a staff preview of the candidate flow.

Before exam day

• Send the candidate instruction note and device-test details.
• Pilot common Android devices and lower-bandwidth connections.
• Arrange approved accessibility accommodations.
• Train reviewers on the evidence and escalation process.

During the exam

• Log technical support without giving academic help.
• Record identity-check issues and action taken.
• Approve retries and time exceptions individually, with a reason.
• Monitor patterns but do not treat an alert as automatic misconduct.

After the exam

• Close attempts and preserve relevant evidence securely.
• Review flags against recordings, session history and answers.
• Use a second reviewer for serious cases.
• Offer a response or viva route for high-stakes decisions.
• Communicate decisions consistently.
• Delete recordings in line with the retention period given to candidates.

If your assessment process must connect with attendance, student records and parent communication, see how school management software supports examination administration. For monitoring and evidence review, read about online exam integrity controls and human review.

Choose one upcoming assessment, complete the checklist and run a pilot before sending the link. You can try RootPro for Business free for AI-proctored online assessments.

Frequently asked questions

Can online exams ever be completely cheat-proof?

No. A helper can remain outside a camera frame, and a candidate may use an undisclosed device or mirrored screen. Use layered controls, human review and a short viva for cases where the result has serious consequences.

How can schools stop students from using ChatGPT in an online exam?

Do not rely on a ban alone. Use application-based questions, randomised papers, a firm time limit, clear AI rules and a follow-up viva where an answer appears inconsistent with the student’s earlier work.

Should a proctoring flag automatically fail a student?

No. A flag shows that something needs review, not that cheating occurred. Check the recording, session events and answers, then give the candidate a chance to explain a credible concern.

What should happen if a candidate loses internet during an online exam?

Record the incident and review what the platform received before approving any exception. A delayed submission can be handled differently from a candidate who needs more writing time or a new attempt after seeing questions.

How do you verify identity when students share a family phone?

Use a stable admission number and give each student an individual single-use code or authenticated login. Add a photo ID or face check for higher-stakes assessments, and retain the evidence under your stated privacy policy.