How to Handle ChatGPT Use in Hiring Tests

Last updated: 23 October 2026 · By the Asuraa Team

Quick answer: You cannot fully stop candidates using ChatGPT in remote tests, so set a clear policy (ban, allow, or allow with disclosure), redesign questions to match, and verify shortlisted candidates in a live round. Proctoring tools such as HackerRank's Proctor Mode flag tab switches, block paste and capture screenshots, and Google said in 2025 it would require at least one in-person round for some roles.

Key takeaways

• Remote proctoring reduces casual AI misuse but cannot fully prevent it, so a live verification round for shortlisted candidates is the most reliable check.
• HackerRank's Proctor Mode flags tab switches, disables copy-paste in the editor, captures screenshots every 15 seconds and runs AI plagiarism detection by default.
• Anthropic's candidate policy, reported in July 2025, bars AI in most assessments unless instructions allow it, while Canva since June 2025 expects AI use in technical interviews.
• Google CEO Sundar Pichai said in 2025 that at least one in-person interview round would be required for certain roles such as those with coding tests.
• Proctoring collects webcam and screen data, so Indian employers should give clear notice and safeguards in line with the DPDP Act 2023 and Rules notified on 14 November 2025.

You cannot reliably stop every candidate from using ChatGPT in an unsupervised online test, so the practical answer is to choose a clear AI-use policy (ban, allow, or allow with disclosure), redesign the test to match it, use proctoring only where the stakes justify it, and confirm every shortlisted candidate's ability in a live round. Detection tools reduce casual misuse; test design and live verification are what actually protect your hiring decision.

Can you stop candidates using ChatGPT in online assessments?

You can make AI misuse much harder, but not impossible, in a remote test. Proctoring platforms combine several controls. HackerRank's Proctor Mode documentation, for example, lists webcam presence checks, flags for leaving full-screen or switching tabs, disabled copy-paste in the code editor, periodic screenshots that a model classifies as suspicious or not, and an AI plagiarism check that is on by default when Proctor Mode is enabled (HackerRank). Candidates also have to acknowledge the prohibited actions and grant webcam and screen-sharing permissions before they start.

These controls catch the obvious cases: a second tab open on ChatGPT, pasted code, someone else in the room. They are weaker against a phone held below the webcam or a second device. That is why large employers have added a supervised step. Google's CEO Sundar Pichai said in June 2025 that at least one in-person round would be required for certain roles such as those with coding tests, as reported by CXO Digital Pulse (CXO Digital Pulse).

What AI-use policy should you set for hiring tests?

Pick one of three policies for each test and state it in the invitation, the instructions and on the first screen. Companies have landed in different places: Anthropic's candidate guidance, as reported by Fortune in July 2025, bars AI in most assessments and live interviews unless instructions say otherwise but allows it for refining applications and preparing (Fortune), while Canva announced in June 2025 that it expects backend, frontend and machine learning candidates to use AI coding tools in technical interviews (Canva).

Canva's redesign is a useful model for the "allow" option. It replaced classic algorithm puzzles, which AI tools solved quickly, with bigger and more ambiguous problems, and it now looks at whether candidates ask clarifying questions, break the problem down, and catch and fix problems in AI-generated code (Canva).

How do you redesign a test so ChatGPT matters less?

Design questions where the answer depends on context the AI does not have, or where the score comes from reasoning you observe. Practical changes, in order of effort:

1. Replace recall questions with applied ones. "What is a primary key?" is instantly answerable; "Here is our orders table with duplicate rows; write the query that finds them and explain how you would stop it happening" is less so.
2. Use your own material. A real (anonymised) dataset, a short internal-style spec or a bug from your codebase gives the candidate context a chatbot lacks.
3. Time-box tightly for banned-AI tests. Short questions with reasonable time limits leave less room for copying a question into another device and waiting for an answer.
4. Ask for explanations alongside answers. A two-line "why" under each answer is easy for someone who understood it and awkward for someone who did not.
5. Randomise from a large pool. Different candidates get different variants, which limits shared answer keys in campus WhatsApp groups.
6. Add a live follow-up. A 15 to 20 minute call where the candidate explains or extends one answer is the strongest single check. For take-homes, see our guide to fair take-home assignments for freshers.

Should you use proctoring software for fresher tests?

Use proctoring for high-volume, banned-AI screening tests where you cannot supervise in person, and tell candidates exactly what is monitored. For small hiring rounds, a short live session is often simpler and fairer than heavy monitoring.

Proctoring collects sensitive personal data: webcam images, screenshots of the candidate's screen and behaviour logs. HackerRank's documentation says screenshots are captured every 15 seconds, or every five seconds after suspicious activity, under Proctor Mode (HackerRank). In India, the Digital Personal Data Protection Act 2023 and the DPDP Rules notified on 14 November 2025 require a clear notice of the specific purpose, consent that can be withdrawn, and reasonable security safeguards (PIB). Check your vendor's data handling and retention settings, and describe the monitoring in your test invite. This is general information, not legal advice.

Candidates also need to know what a warning means. Many freshers panic over a tab-switch flag; our explainer on what happens after a tab-switch warning shows what they are told, and you should treat flags as a prompt for review, not automatic rejection.

What do most guides on stopping ChatGPT in tests get wrong?

Most guides treat this purely as a detection problem, looking for the right proctoring setting or AI-text detector. The bigger issue is that many hiring tests were measuring the wrong thing: if ChatGPT can pass your screening test, the test was mostly measuring recall that the job no longer requires unaided.

Two other corrections:

• A flag is not proof. Proctoring tools output signals such as integrity ratings and flagged screenshots. HackerRank's own documentation describes integrity results and session replays for review (HackerRank). A human should review flagged sessions before any decision, and a candidate who is flagged but strong in a live round should not be auto-rejected.
• Bans without redesign punish honest candidates. If the test can be beaten by a hidden phone, the candidates who follow your rules are at a disadvantage. Either supervise properly or move to "allow" or "allow with disclosure" and raise the difficulty.

For background on how candidates experience these tools, see can online assessments detect cheating and our Mercer Mettl proctoring guide.

How do you verify skills after an online test?

Verify skills with a short live round for everyone you shortlist, built on the same questions for each candidate. Pichai's reasoning for Google's in-person round was to make sure the fundamentals are there (CXO Digital Pulse), and the same logic applies to a startup hiring five freshers.

A simple setup:

1. Pick one question the candidate answered in the online test.
2. Ask them to walk through it in their own words, then change one constraint and ask how their answer changes.
3. Score on a written scale, as you would in a structured interview.
4. If the live performance is far below the test score, discuss it with the panel before deciding.

The wider picture of AI in hiring, from screening to interviews, is covered in our AI hiring and recruitment guide.

Related guides

• AI Hiring and Recruitment: A Guide for Candidates and Employers

FAQ

Can online assessments detect ChatGPT use?

Partly. Proctoring platforms combine webcam checks, tab-switch flags, blocked copy-paste, screenshots and code-similarity or AI plagiarism checks. HackerRank's Proctor Mode, for example, captures screenshots every 15 seconds and runs AI plagiarism detection by default. These catch obvious misuse but are weaker against a second device, so a live follow-up round is still needed.

Should employers ban AI tools in hiring tests?

Ban AI only where you are measuring fundamentals a new hire must have unaided, and only if you can enforce it with proctoring or supervision. Otherwise a ban mainly disadvantages honest candidates. Many employers now allow AI in some rounds, as Canva does in technical interviews, and assess judgement and debugging instead.

What is an AI-use policy for candidates?

It is a short statement telling candidates where AI tools are banned, allowed, or allowed with disclosure in your hiring process. Anthropic, as reported by Fortune in July 2025, allows AI for refining applications and interview preparation but bars it in most assessments and live interviews unless instructions say otherwise. State your policy in invites and on test screens.

How do you design a hiring test that ChatGPT cannot easily solve?

Use applied questions built on your own data or scenarios, ask candidates to explain their answers, time-box tightly where AI is banned, randomise from a large question pool, and add a short live follow-up. Canva replaced classic algorithm puzzles with larger, more ambiguous problems that cannot be solved with a single prompt.

Is it legal to proctor candidates with webcam monitoring in India?

Proctoring is widely used, but it collects personal data such as webcam images and screenshots. Under the DPDP Act 2023 and the DPDP Rules notified on 14 November 2025, employers should give a clear notice of the purpose, obtain consent that can be withdrawn and keep reasonable security safeguards. This is general information, not legal advice.

What should employers do when a candidate is flagged for cheating?

Review the flagged session before deciding. Proctoring tools produce signals such as integrity ratings, flagged screenshots and session replays, not proof. Check what triggered the flag, and if the rest of the evidence is unclear, invite the candidate to a short live round on the same material. Reject on evidence, not on an automatic flag.

Final thoughts

Treat ChatGPT in hiring tests as a design choice, not just a cheating problem: state a policy, build the test around it, proctor where it is proportionate, and always confirm skills live. That keeps your process fair to honest candidates and useful for you. You can post your fresher roles, with the assessment steps described, on Asuraa Jobs.