Scientists Build the First Validated Scale to Measure AI Addiction

Artificial intelligence has slipped into nearly every corner of daily life, from chatbots that draft emails to assistants that plan meals, write code, and keep lonely people company late at night. As these tools become more persuasive and more pervasive, psychologists have grown increasingly worried that some users may be developing relationships with AI that look less like convenient tool use and more like behavioral addiction. Until now, however, researchers lacked a standardized, validated instrument to measure that risk. A team of Turkish psychiatric nursing researchers has moved to fill the gap, publishing the development and psychometric evaluation of the BERTA Artificial Intelligence Addiction Scale, or B-AIAS, in the journal Discover Psychology.

The study, led by Belgin Varol of the University of Health Sciences in Ankara, together with Yalçın Kanbay of Artvin Çoruh University and Burcu Arkan of Bursa Uludag University, set out to do something deceptively simple: create a short, reliable questionnaire that can distinguish ordinary enthusiasm for AI tools from problematic patterns of use that resemble the compulsions seen in gambling, gaming, and social media addiction. The researchers note that while AI technologies offer significant benefits, their excessive use may trigger problematic patterns that mirror established behavioral addictions, and that no standardized instrument previously existed to assess these AI-related addictive tendencies.

What makes the new scale notable is the rigor of its construction. Rather than simply drafting questions and releasing them online, the team followed the COSMIN framework, an internationally recognized consensus-based standard for selecting health measurement instruments. That framework dictated a multi-stage pipeline: item generation from the theoretical literature, expert review to prune weak or ambiguous candidates, exploratory factor analysis to uncover the underlying structure of the data, confirmatory factor analysis to test that structure on a fresh sample, assessments of convergent and discriminant validity to ensure the scale measures what it claims and nothing else, item discrimination analysis, and test–retest reliability to confirm that scores are stable over time.

The data came from online surveys distributed through social media platforms, targeting adults in Turkey who reported using AI tools. The researchers split their sample in a methodologically important way. One group of 308 participants provided data for the exploratory factor analysis, the statistical fishing expedition that reveals how items cluster together. A separate group of 306 participants supplied the confirmatory data, allowing the team to test whether the structure discovered in the first sample held up in an independent one. This two-sample design guards against the common pitfall of building a scale that fits one dataset by statistical accident but fails to generalize.

The final instrument is strikingly lean: eight items rated on a five-point Likert scale ranging from zero, meaning not at all, to four, meaning completely. Despite its brevity, the scale captures two distinct psychological dimensions. The first, labeled Deprivation, comprises four items and probes the distress and craving users feel when separated from their AI tools, a hallmark of dependence that echoes withdrawal symptoms in substance and behavioral addictions. The second, labeled Sensitivity, also comprises four items and reflects heightened emotional reactivity tied to AI use. Together, the two factors explained 60.2 percent of the variance in responses, a substantial share for a brief behavioral measure and an indication that the eight items collectively tap a coherent underlying construct.

The reliability statistics are equally strong. Cronbach’s alpha, the classic measure of internal consistency, came in at 0.88, while McDonald’s omega, a more modern coefficient that is less sensitive to violations of statistical assumptions, reached 0.90. Both figures sit comfortably above the conventional 0.70 threshold for acceptable reliability and approach the 0.95 ceiling beyond which items may be redundantly measuring the same thing. Notably, none of the items required reverse coding, which simplifies scoring and reduces the respondent errors that reversed items often introduce. Test–retest analysis confirmed that participants’ scores remained stable across repeated administrations, a prerequisite for any instrument intended to track changes in behavior over time or to evaluate interventions.

The implications reach well beyond Turkish universities. As generative AI companions proliferate, clinicians and public health researchers have warned that the design of these systems, engineered for engagement and emotional resonance, may exploit the same psychological levers that make slot machines and infinite-scroll feeds so compelling. A validated scale gives researchers the measurement tool they have been missing. With it, they can estimate the prevalence of problematic AI use in different populations, identify risk factors, test whether people who score high on the B-AIAS show the functional impairment characteristic of behavioral addiction, and evaluate whether therapeutic or design interventions actually reduce dependence. The authors explicitly position the instrument as a foundation for future research on technology-related behavioral addictions and for cross-cultural investigations of AI use patterns.

The study also carries practical weight for health care. All three authors are psychiatric nursing specialists, and the work was approved by the Health Sciences Research and Publication Ethics Committee of Bursa Uludag University in May 2025, with written informed consent obtained from every participant in line with the Declaration of Helsinki. A brief, freely usable screening tool fits naturally into clinical workflows, where long diagnostic interviews are impractical. Because the scale is short enough to complete in under a minute, it could be embedded in routine mental health assessments, digital wellness apps, or large-scale epidemiological surveys, giving clinicians and policymakers an early-warning signal as AI saturation deepens.

Caveats remain, and the authors are careful about scope. The B-AIAS was developed and validated among adults in Turkey, and psychometric properties do not automatically transfer across languages and cultures; cross-cultural validation studies will be needed before the scale can be deployed globally with confidence. The two-factor structure of Deprivation and Sensitivity also invites further research into whether these dimensions map onto the craving and withdrawal constructs used in established addiction frameworks, or whether AI dependence follows its own distinctive trajectory. And as the technology itself evolves, the behaviors the scale measures may shift with it, meaning the instrument will likely require periodic re-evaluation as AI tools change form.

Still, the arrival of the B-AIAS marks a threshold moment in the psychology of human–machine interaction. For decades, researchers have built validated measures for gambling, internet gaming, and smartphone dependence, and each of those instruments proved essential for turning vague cultural anxieties into testable science. The B-AIAS now extends that tradition to the most immersive technology yet. As hundreds of millions of people form daily habits around conversational AI, the question is no longer whether problematic AI use exists but how common it is, whom it affects, and what can be done about it. With a validated eight-item scale in hand, researchers finally have a way to find out, and the answers may shape how societies manage the technology that is rapidly becoming their constant companion.

Subject of Research: Development and psychometric validation of a scale measuring addictive patterns of artificial intelligence use

Article Title: Development and psychometric evaluation of the BERTA Artificial Intelligence Addiction Scale (B-AIAS)

Article References: Varol, B., Kanbay, Y., & Arkan, B. (2026). Development and psychometric evaluation of the BERTA Artificial Intelligence Addiction Scale (B-AIAS). Discover Psychology. https://doi.org/10.1007/s44202-026-00921-2

Image Credits: AI Generated

DOI: 10.1007/s44202-026-00921-2

Keywords: artificial intelligence, behavioral addiction, psychometrics, scale development, validity, reliability, factor analysis, mental health, problematic AI use, Turkey, Development, psychometric