Every recruiter has wondered about it at some point. The response is polished to a degree that feels off. The writing is structured in a way the candidate’s resume never suggested they could pull off. The answer covers every angle perfectly, like it was generated rather than written. The question that follows is almost automatic: did this candidate use AI to complete their assessment?
It’s a fair question. AI tools have made it easier than ever for candidates to produce impressive-looking responses that are far from their actual professional ability. For hiring teams trying to make decisions based on real evidence, that creates a genuine problem and raises a reasonable concern about the reliability of your hiring signal.
How to evaluate candidate skills in the AI era is a question TA teams can no longer afford to ignore.
But there is a second question behind that first one, and it matters just as much: if a candidate used AI, does that automatically mean they should be disqualified? Or does it mean you need a smarter way to evaluate what they actually bring to the role? This post walks through both sides: how to detect AI use during an assessment, and how to use that information to make better hiring decisions rather than just filtering candidates out.
Why Are So Many Hiring Teams Asking This Question Right Now?
The timing is not a coincidence. AI writing tools became mainstream in 2023, and by 2026 they are embedded into how people work, communicate, and yes, apply for jobs. Candidates are using them to write cover letters, polish CVs, and complete take-home assessments. That is the reality hiring teams are operating in.
By 2030, the World Economic Forum projects that nearly 40% of workers’ core skills will change dramatically.
Part of that shift is AI proficiency itself becoming a core competency. Every role, from customer success to finance to software development, now requires some degree of working alongside AI tools. The line between “cheating with AI” and “using AI the way you would on the actual job” is blurring, and hiring teams are caught in the middle.
The practical effect is that traditional written assessments and take-home tasks have become unreliable, when not conducted in a cheat-proof environment. A strong response no longer tells you whether a candidate can think. It tells you whether they know how to prompt a language model. That is a different skill, and one that deserves to be evaluated directly rather than caught out after the fact.
What Are the Signs a Candidate Used AI to Complete an Assessment?
Before getting to the detection tools available, it is worth knowing what AI-generated or heavily AI-assisted responses tend to look like in practice.
Common signals include writing that is noticeably more polished or structured than the candidate’s other written communication, such as their cover letter or email exchanges. Responses that address every possible angle of a question in a balanced way, with no clear personal voice or concrete examples from the candidate’s own experience. Generic phrasing that could apply to any company or role, with no specific detail tied to the actual job or task. Unusual consistency in formatting, where every paragraph follows the same structure regardless of the question type.
None of these signals is definitive on its own. Some candidates write well. Some are naturally structured thinkers. But a combination of these patterns, especially against a backdrop of minimal evidence of the same capability elsewhere in the process, is worth paying attention to.
Can Assessment Platforms Actually Detect AI Use During a Candidate’s Session?
Yes. Modern
skill assessment platforms have moved well beyond the honor system. Several specific capabilities are designed to flag or prevent AI-assisted cheating during a live assessment session.
ChatGPT detection flags responses that show strong indicators of AI generation and identifies when candidates paste external responses into a question field. Copy-paste detection identifies whether a candidate pasted text into a response field rather than typing it, which is a common method when working from an AI-generated draft outside the assessment window. Tab-switch tracking records when a candidate navigates away from the assessment, which can indicate they are consulting an external tool mid-task. Device and Location Tracking flags sessions where a candidate accesses or completes an assessment from multiple devices or locations.
Canditech’s anti-cheating suite combines all of these controls in one platform, and hiring teams can configure which ones apply to each assessment depending on the role and level of oversight required. For a task where independent work matters, you can layer multiple controls. For a task where using AI is expected, you can have the candidate use the embedded AI tools within the Canditech assessment, which gives you full visibility on how they work with AI including prompting, fluency, and critical thinking. The data these controls surface gives you something real to work with: not a guess about whether AI was involved, but a record of the session behavior and how it was used.
Is Knowing Whether a Candidate Used AI Enough to Make a Decision?
This is where the question gets more interesting. Detection tells you that something happened. It does not tell you whether what happened disqualifies the candidate or reveals a capability gap.
Consider two candidates who both used an AI tool on the same open-text task. One pasted in an unedited AI response without reading it carefully, resulting in an answer that contains a factual error and misses the specific context of the question. The other used AI to structure their thinking, added their own examples and judgment, caught the error in the draft, and produced a response that is both accurate and clearly personalized. Both candidates used AI. One demonstrated strong AI collaboration skills and critical thinking. The other demonstrated neither.
If your only data point is “did they use AI?”, you cannot tell these two candidates apart. If your assessment is designed to evaluate how they work, you can.
What Should You Actually Be Measuring When It Comes to AI?
The more useful question is not whether a candidate used AI. It is whether they can use it effectively in a work context. That is a skill, and like any skill, it can and should be tested directly, considering the age we live in.
AI proficiency assessment in 2026 covers three core competencies. The first is prompt engineering: the ability to give an AI tool clear, specific, contextually appropriate instructions that produce useful output. The second is critical evaluation: the ability to review AI-generated content, identify what is wrong or missing, and improve it before using it. The third is AI-human collaboration: knowing when to use AI and when to rely on independent judgment, and being able to blend both in a way that produces better outcomes than either alone.
These are not abstract concepts. They show up in how a candidate handles a task. A
job simulation that requires a candidate to work through a realistic, job-related scenario, using an AI tool as part of the process, will tell you far more about their actual capability than any after-the-fact analysis of whether they pasted text from ChatGPT.
How Can You Test AI Skills Directly Instead of Trying to Catch Their Use?
Canditech’s AI skills assessment is built specifically for this. Rather than asking candidates generic questions about AI or attempting to detect prohibited tool use, the platform lets you put candidates in a monitored environment where using AI is part of the task itself.
Canditech is the leading assessment platform with a dedicated AI Skills Library and the ability to embed ChatGPT directly inside an assessment task. A candidate can be given a realistic work scenario and access to an AI tool at the same time in the same assessment environment. What you are scoring is not whether they used AI. It is the quality of their prompts, the accuracy of their evaluation of the outputs, and the judgment they applied in deciding what to keep, what to change, and what to reject.
This approach solves the fundamental problem with trying to police AI use: it replaces suspicion with evidence. Instead of wondering whether a candidate’s polished response means they are skilled or just good at prompting, you design the task so that both are visible at once. The anti-cheating controls are still there as a baseline. But the goal shifts from catching candidates out to understanding what they can actually do and how they would use AI to succeed at their job.
What Does This Look Like in Practice for a Hiring Team?
A practical implementation looks something like this. For roles where AI proficiency is a requirement, you include one or two job simulation tasks that explicitly involve working with an AI tool. Canditech generates the assessment from your job description in seconds, pulling from a library of
500+ pre-built simulations and customizing the assessment to your role. You configure the anti-cheating controls to monitor session behavior without blocking AI use within the designated tasks. Responses are auto-scored using Canditech’s auto-scoring engine for closed answer questions, and
Canditech’s AI scoring engine, which evaluates open-text and video answers against criteria you define, giving you a ranked and consistent view of every candidate before a human reviews.
The candidates who do well are not the ones who figured out how to game a take-home task. They are the ones who can actually do the job and work with AI in the way the role demands. That is a meaningfully better hiring signal than anything you can get by trying to detect unauthorized tool use after the fact.
The question “how to tell if a candidate is using AI” has a real answer, and modern platforms give you the tools to find it. But the more valuable answer is the one you get when you stop treating AI use as something to catch and start treating it as something to assess. Ready to see how that works in practice?
Book a demo with Canditech and we will show you exactly how to build it for your roles.
