AI use in Reasoning Ability Tests

The rise of generative AI presents both opportunities and challenges for remote testing. A growing concern is that candidates may use AI tools such as ChatGPT while completing reasoning ability tests. These tests are designed to assess an individual’s problem-solving skills, and AI-assisted responses may compromise their purpose by masking the candidate’s true abilities.

At Hudson, we offer a wide range of mitigation strategies that can be deployed wherever needed or preferred. These include both preventive and detection-based measures to help organisations maintain the integrity of their tests and assessments.

The good news first

While AI tools are becoming more accessible, we currently see no major shifts in average reasoning test scores or clear signs of large-scale irregularities. However, the potential for misuse is real and may grow over time. For some hiring organisations it therefore may be important to apply appropriate measures to safeguard test integrity.

What can organisations do?

We recommend a dual approach:

Deterrence strategies: preventing unauthorised AI use

Built into Hudson’s test platform:

• Copy-pasting disabled: This technical restriction on our test platform prevents candidates from copying text from the test interface and pasting it elsewhere, such as into an AI tool. While not foolproof, it introduces a practical barrier against quick and easy AI consultation.
  
  
• Strict time limits: Our cognitive tests include strict time constraints, which serve as a natural barrier to external assistance. Using AI tools during the test involves several time-consuming steps: switching windows, copying questions, prompting the AI, reading and interpreting the response, and deciding whether to use it. These interruptions not only consume valuable seconds but also break concentration and increase stress.
  
  

Optional or tailored measures:

• Remote proctoring: Organisations may opt for remote proctoring solutions, which usually include live or automated monitoring using the candidate's webcam and monitoring of their screen activity. Where needed, Hudson can integrate third-party proctoring solutions with its test platform, offering clients flexibility in choosing the level and type of supervision that fits their process. Remote proctoring can effectively discourage unauthorised aid, but requires careful consideration of candidate experience, cost, and privacy concerns.
  
  
• Honesty agreement: An honesty agreement is a short declaration that candidates read and accept before starting the test. It reminds them of the importance of ethical behaviour and explicitly states that the test should be completed without external help. Research shows that this kind of ethical prompt can reduce dishonest behaviour, even in unproctored environments. On Hudson’s test platform, the honesty agreement is available as an optional feature that can be activated per client. The wording of the statement can also be tailored to reflect the organisation’s tone and values.
  
  
• Advance notice of detection or verification procedures: Simply informing candidates before they start the test, that irregularities may be detected—and that follow-up testing may occur—can serve as a deterrent.
  
  

Detection strategies: spotting suspicious activity

• Click tracking: On Hudson’s test platform, logs capture whether candidates click outside the main test tab. While occasional clicks are usually harmless, repeated switching may suggest attempts to consult external tools. This data can help flag potentially suspicious behaviour for further review.
  
  
• Data forensics: Upon request, Hudson offers detailed analysis of response patterns at both individual and group level, including item-level response times, accuracy patterns, and consistency across difficulty levels. While such analyses cannot confirm misconduct definitively, they can provide a basis for follow-up action.
  
  
• Verification testing: If irregularities are suspected, Hudson can advise clients in the use of verification measures. These may include a follow-up test, a live interview, a simulation exercise, or a short problem-solving task using a think-aloud protocol. This ensures the candidate’s true reasoning skills can be reassessed under closer scrutiny, without repeating the full test process.

No one-size-fits-all solution

Each organisation has its own context: the types of candidates it attracts, available budget, desired employer brand, selection goals, and level of risk tolerance. These elements must all be considered to strike a balanced approach that prevents or mitigates unauthorised assistance. The goal is to find the sweet spot between predictive accuracy, fairness, a positive candidate experience, and practical and financial feasibility. Our team is ready to help identify the most suitable strategy for your needs.

AI may be evolving rapidly, but so is our toolbox to ensure fairness and integrity in testing.