LLM Addiction Scale (LLMAS-18)

An 18-item instrument for assessing addiction-like patterns associated with large language model use

The LLM Addiction Scale (LLMAS-18) is a concise, self-administered instrument designed to assess addiction-like behavioral, cognitive, and emotional patterns associated with the use of large language model (LLM)-based systems. These systems may include conversational artificial intelligence platforms, generative text applications, AI-assisted productivity tools, and other interactive language-based technologies that increasingly participate in academic, professional, creative, and everyday activities. The instrument does not propose LLM-related behavioral dependence as a formally recognized psychiatric disorder, but instead operationalizes addiction-like features associated with excessive or dysregulated engagement with LLM-based technologies.

The scale comprises 18 items organized across three theoretically grounded dimensions reflective of core behavioral addiction constructs adapted to the context of LLM interaction: compulsive engagement, cognitive dependence, and emotional-social reliance. The proposed domains were partially informed by behavioral addiction models described in the Diagnostic and Statistical Manual of Mental Disorders and the World Health Organization ICD-11 framework, while incorporating features more specific to LLM-mediated interaction, including cognitive outsourcing, compulsive consultation behavior, reassurance seeking, and AI-assisted emotional regulation.

Proposed Core Constructs for the LLM Addiction Scale (LLMAS-18)

Compulsive Engagement

Persistent and dysregulated interaction with LLM-based systems characterized by excessive use, urges to engage, preoccupation, impaired control, and interruption of other activities.

Cognitive Dependence

Progressive reliance on LLM-based systems for problem-solving, decision-making, task initiation, cognitive reassurance, and maintenance of perceived productivity or performance.

Emotional - Social Reliance

Use of LLM-based systems for emotional regulation, reassurance, comfort, loneliness reduction, and preferential discussion of personal or professional concerns with AI-based systems rather than with other individuals.

Below is the 18-item LLMAS-18, with 6 items per construct, rated on a 5-point frequency-based Likert-type response format.

LLM Addiction Scale (LLMAS-18)

Below is a list of statements related to the use of large language model (LLM)-based systems. Please indicate how often each statement applied to you during the past 3 months.

There are no right or wrong answers. Please respond as honestly as possible based on your typical experiences.

For each statement, select one response according to the following scale:

0 = Never

1 = Rarely

2 = Sometimes

3 = Often

4 = Very Often

For the purposes of this questionnaire, ‘LLM-based systems’ refer to artificial intelligence tools capable of generating human-like text responses through conversational interaction.

The LLMAS-18 consists of 18 items organized into three subscales:

Compulsive Engagement

Items: 1–6

Score range: 0–24

Cognitive Dependence

Items: 7–12

Score range: 0–24

Emotional–Social Reliance

Items: 13–18

Score range: 0–24

The total LLMAS-18 score is calculated by summing responses across all 18 items.

Total score range: 0–72

The The LLM Addiction Scale (LLMAS-18) and its associated materials (items, structure, scoring rubric, and interpretive guidance) are the intellectual property of Dr. Igor V. Pantić and are protected under applicable national and international copyright laws.

Academic and research use:

Use of the LLMAS-18 for non-commercial academic research purposes is permitted only with prior written permission from the copyright holder. Any publication, dissemination, or adaptation of the scale must include proper citation and acknowledgment of the author.

Commercial use:

Use of the LLMAS-18 or its derivatives for commercial purposes, including but not limited to integration into digital platforms, clinical software, corporate wellness tools, or monetized educational content, is strictly prohibited without a formal licensing agreement.

To request permission for academic or commercial use, please contact:

Dr. Igor V. Pantić

Email: igorpantic@gmail.com

Any unauthorized reproduction, distribution, or modification of this material is prohibited and may result in legal action.

Google Sites

Report abuse