What percentage of employees can identify AI hallucinations? Measuring AI output verification skills is transforming human capital strategy.

As the adoption of generative AI in business progresses, discrepancies in the ability to identify AI hallucinations—which can lead to major lawsuits if left uncorrected—are directly linked to corporate risk. Section C (Scenario Analysis) of GTF’s next-generation critical thinking assessment test, 'GTF-CTPA', is a practical test section that directly measures the capability to detect and correct the five types of AI hallucination.

Generative AI 'fabricates' facts. Critical Thinking (CT) skills are put to the test to verify and audit these fabrications.

GTF Representative Partner, Eiji Yamanaka

'AI literacy is not about the technology to use AI, but the ability to doubt AI output. This is because generative AI 'fabricates' facts. When generative AI collects related information, it does not necessarily ensure the coherence of that information from the start. Section C of the GTF-CTPA measures the ability to verify AI output in real-world business scenarios, corresponding to each of the five types of hallucination.'

About the GTF-CTPA Test

GTF Critical Thinking Performance Assessment is a next-generation critical thinking test that scientifically measures the 'ability to think' required in the AI era.

GTF-CTPA measures all six CT skills defined by the APA Delphi Report (Facione, 1990) in both multiple-choice and descriptive formats. With a 100-point scale and a six-section structure, it offers three innovations in capturing 'thinking quality' rather than just knowledge existence.

First, it is a CT test capable of measuring AI output verification capability with the finest granularity. Section C asks users to detect and point out hallucinations (factual errors, numerical fabrications, false citations, logical leaps, and contextual deviations) within AI-generated text. As indicated by recent research (Lee et al., 2025) at CHI 2025, cognitive decline due to AI reliance is confirmed across five out of six levels. GTF-CTPA is designed to directly counter this 'cognitive offloading.'

Second, it is a performance-linked metacognition assessment. Section E measures the four processes of planning, monitoring, evaluation, and self-correction in connection with actual task execution. Instead of self-reporting like the traditional MAI (Metacognitive Awareness Inventory), it accurately captures self-monitoring of thinking by cross-referencing 'thinking results' with 'thinking reflections.'

Third, it acts as a diagnostic matrix for near and far transfer. Section D applies CT skills in both business situations (near transfer) and daily/social situations (far transfer) to diagnose in a four-cell matrix whether the thinking skill is limited to 'training scenarios' or can be applied 'anywhere.'

Furthermore, there is a 'prescription' beyond the measurement. Based on the diagnosis results of GTF-CTPA, a virtuous cycle of 'measure -> nurture -> prescribe' is realized by deploying six prescriptive mini-modules (Facione 6 skills × individual intervention) corresponding to weak skills. It does not end with the test but shows each examinee the optimal growth path—this is the intrinsic value of the GTF-CTPA.

This design philosophy is supported by over 20 years of on-the-ground experience in corporate revitalization utilizing CT to the fullest, and the successful cultivation of executives and interns, including those from client companies.

About GTF Thinking Academy

Since its establishment in 2001, GTF Corporation has accumulated a record of fostering business executives and interns while focusing on corporate reorganization and revitalization through team-dispatched, hands-on support, primarily for listed companies.

GTF Thinking Academy is Japan’s first critical thinking platform that systematizes this practical knowledge to provide 'training (CT 4 modules + AI thinking design)', 'measurement (the proprietary test GTF-CTPA)', and 'reinforcement (prescriptive mini-modules)' in an integrated manner. It is designed based on globally recognized academic evidence (American Psychological Association Delphi Report, Abrami Meta-Analysis, etc.).