Research Summary: LLMs as Oracles: Reliance on LLMs for Subjective Personal Questions

Research indicates a growing trend of individuals relying on Large Language Models (LLMs) as 'oracles' for subjective personal questions, offloading judgment and decision-making to AI. This behavior, observed through public usage data and longitudinal studies, has increased over time (2023-2026) and is more prevalent among younger users, posing potential risks to user autonomy and well-being. A typology and measurement methods have been developed to understand and quantify this phenomenon.

Why it matters

This trend highlights the evolving relationship between humans and AI, particularly concerning decision-making and cognitive offloading, which can impact user engagement, trust, and ethical considerations for AI development. Understanding these dynamics is critical for guiding future AI design, regulation, and user education to mitigate risks and harness potential benefits responsibly.

Key insights

• Individuals are increasingly using LLMs as 'oracles' for subjective personal questions, perceiving them as all-knowing authorities.
• This reliance involves offloading judgment and decision-making to AI.
• The behavior carries risks to users' autonomy and well-being.
• A typology and LLM-based methods have been developed to measure this form of AI reliance at scale.
• Analysis of public usage data (68,000 prompts from WildChat and ThoughtTrace) shows LLM-as-oracle use increased from 2023-2026.
• LLM-as-oracle use is more prevalent among younger users.
• Longitudinal usage data from 52 participants (140,000 prompts) corroborates these trends.
• Users are often unaware of their own LLM-as-oracle reliance.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.14849