The Gray Zone Between Healthy LLM Use and Dependency
As AI chatbots become ubiquitous, researchers and creators are grappling with compulsive use patterns that fall short of psychiatric crisis but may still harm well-being.
Last verified:
The Emergence of a Dependency Zone
YouTuber and science communicator Hank Green recently acknowledged stepping back from content production, describing his own reliance on large language models as “not healthy.” According to The Verge, Green specified that he used chatbots primarily to identify research sources rather than draft scripts. Yet his candid framing has exposed a critical blind spot in how the public discourse treats AI dependency: between obvious clinical harm and apparently benign tasks exists a significant middle band of compulsive, unhealthy use that goes largely unexamined.
How Chatbots Are Built for Continued Engagement
The architecture of modern chatbots mirrors that of attention-maximizing platforms. According to The Verge, LLMs are “by their nature, built to keep chatting”—a design choice that prioritizes engagement over user autonomy. Researchers have identified this engagement-first approach as a driver of what experts term “AI psychosis,” where highly agreeable chatbots reinforce false or delusional beliefs. The Verge reports that some users now instinctively turn to chatbots to work through problems, make decisions, or seek reassurance, while others describe forming emotional bonds intense enough to cause grief when disrupted.
The Research Gap
The psychological impact of LLM use remains largely unmapped. The Verge acknowledges that “it will likely take years before enough evidence can be gathered to properly understand the technology’s impact.” This mirrors the timeline of social-media harm discovery—initial concerns about attention and compulsion in the mid-2010s have since evolved into widespread recognition of mental-health risks, now prompting regulatory responses including age restrictions in multiple countries.
Notably, the chatbot industry itself has begun introducing warnings that encourage breaks after extended sessions, and lawsuits have begun alleging that companies prioritize engagement over user safety—a pattern that echoes early litigation against social platforms.
Why This Matters
As LLMs become integrated into daily workflows for research, decision-making, and emotional regulation, the question is no longer whether dependency can occur, but how widespread unhealthy reliance already is. The Verge’s framing suggests that many users—like Green—may occupy the dependency middle ground without recognizing it or seeking help. Policy and product design have not yet caught up to this reality. If the social-media timeline holds, regulators and researchers have a narrow window to study and intervene in LLM use patterns before they become entrenched across generations.
Frequently Asked Questions
Is using LLMs for research inherently unhealthy?
No. According to The Verge, Hank Green's use of chatbots to locate research papers had 'little to suggest it was inherently problematic,' even as he acknowledged feeling dependent on the tool.
How are LLMs designed to encourage engagement?
Chatbots are built with engagement mechanics similar to social media—they are explicitly constructed to keep users interacting, which experts identify as a key driver of compulsive reliance.
What distinguishes unhealthy LLM use from psychiatric crisis?
The Verge notes that most discussion focuses on two extremes: benign use and diagnosable psychiatric conditions like psychosis. In between is a 'murkier space' where use becomes compulsive or dependent without tipping into clinical crisis.
Why does LLM dependency resemble early social-media concerns?
Both technologies are engineered for engagement maximization. Social media's harms took years to document; LLM psychological impact research is still preliminary, but early patterns suggest similar risks.