Most coverage of artificial intelligence in daily life centers on workplace productivity and code generation, but recent analytics suggest users are shaping very different trends. A large-scale study by OpenRouter, involving over 100 trillion tokens, gives a rare look into how people interact with models such as ChatGPT, Claude, DeepSeek, Meta’s LLaMA, and other market leaders. This data challenges the prevailing assumption that large language models (LLMs) are mainly productivity tools for professionals. As AI models become increasingly accessible to non-Western users, the study’s demographic reach signals shifting dynamics worldwide. The gap between perceived and actual AI use raises questions about what users really want from these technologies.
Reports from earlier in 2024 and late 2023 identified industry dominance by US-based AI companies focusing on business tasks and developer productivity. Previous market analyses often predicted steady growth in coding assistance and document summarization. Only limited anecdotal evidence had referred to alternative uses, such as creative writing or roleplay, and few studies tracked the significant rise of Chinese models like DeepSeek or Qwen. Current findings highlight a move toward more diverse usage and signal stronger adoption in Asia-Pacific, contrasting with the previous US-centric narrative.
Why Do Users Prefer Roleplay and Storytelling?
The study reveals that more than half of open-source AI model traffic now centers on roleplay and creative storytelling, not productivity. Interactive fiction, character-driven conversations, and simulated gaming scenarios occupy far more time than previously assumed. OpenRouter noted,
“Many users engage with these models for companionship or exploration.”
Structured roleplay activity exceeds even programming tasks, suggesting a significant shift in the relationship between users and generative AI models.
How Has Programming Usage Shifted?
Programming queries have seen the fastest growth: what accounted for 11% of interactions at the beginning of 2025 rose to over half of all activity by the end of the year. As reliance on AI for debugging, architectural review, and code analysis increases, prompt lengths also surged, sometimes exceeding 20,000 tokens per query. Anthropic’s Claude remains especially prominent, with over 60% share in coding-related interactions, though Google, OpenAI, and others intensify competition in this space.
What Does Rising Asian Adoption Mean for Global AI?
Chinese AI models, including DeepSeek and Qwen, now represent 30% of global usage—a dramatic increase from 13%. Simplified Chinese is now the second-most common interaction language, while Singapore trails only behind the United States in overall usage. This expansion outside the US, together with rising adoption in other Asian markets, reduces Western companies’ exclusive dominance and broadens the AI market’s focus and requirements.
The report also highlights the move from single-step queries to multi-step, agentic reasoning tasks. These advanced uses see models executing complex workflows, integrating with tools, and carrying out longer, more persistent interactions. User loyalty turns out to hinge less on general availability and more on models that effectively address unmet needs early—a phenomenon OpenRouter researchers dubbed the “Glass Slipper Effect.” As OpenRouter puts it,
“The median LLM request is no longer a simple question or isolated instruction. Instead, it is part of a structured, agent-like loop.”
With pricing proving relatively inelastic and both premium and budget models maintaining strong user bases, the AI landscape currently allows room for a wide range of offerings. For anyone implementing or evaluating AI tools, it is important to recognize that adoption is driven by unique, locally-relevant use cases, not simply general productivity claims or aggressive pricing. The balance between creative engagement and technical complexity in real-world use means providers may need to support both gaming-style interactions and high-level programming support. Diverse user needs worldwide challenge providers’ standard assumptions and suggest further market segmentation is likely. For those developing AI, awareness of region-specific adoption and strong early engagement for “unsolved problems” can be as crucial as feature set or price in securing lasting loyalty.
