For over a year, artificial intelligence has been widely celebrated as a revolutionary productivity tool—assisting with writing emails, generating code, and summarizing documents. However, a recent data-driven study by OpenRouter uncovers a much more complex and nuanced picture of how AI is actually used in real-world scenarios.
OpenRouter’s analysis, which examined metadata from billions of interactions across more than 300 AI models provided by over 60 companies worldwide, processed over 100 trillion tokens. This extensive dataset includes user interactions with well-known large language models (LLMs) such as ChatGPT, Claude, and various open-source alternatives, offering an unprecedented view into how AI serves diverse user needs globally.
Unanticipated Dominance of Roleplay and Creative Storytelling
Contrary to popular belief that AI usage is primarily oriented towards productivity, the study reveals that over half of all open-source AI model interactions revolve around roleplay and creative storytelling. Users frequently engage in character-driven dialogues, interactive fiction, and gaming scenarios, far surpassing the volume of programming assistance requests.
This discovery challenges the prevailing narrative that large language models are mainly tools for writing code, emails, or document summaries. Instead, many users leverage AI for companionship, exploration, and entertainment. Approximately 60% of tokens used in roleplay contexts relate to gaming or creative writing, highlighting a substantial yet often overlooked segment of AI interaction.
Rapid Growth of Programming-Related AI Usage
Despite roleplay’s prominence in open-source platforms, programming has emerged as the fastest-growing application across all AI models throughout 2025. Coding-related queries surged from 11% of total AI usage at the start of the year to over 50% by year-end.
This trend reflects AI’s deeper integration into software development workflows. The average length of prompts for programming tasks expanded significantly—from about 1,500 tokens to over 6,000 tokens—sometimes exceeding 20,000 tokens, indicative of complex codebase analysis and debugging sessions. Developers now rely on AI for sophisticated architectural reviews and multi-step problem-solving rather than simple code snippets.
In this domain, Anthropic’s Claude models hold a leading market share, accounting for over 60% of programming-related usage during much of 2025, although competition with Google, OpenAI, and open-source models is intensifying.
China’s Expanding Influence in AI Usage
The study also highlights a significant shift in the global AI landscape, with Chinese AI models now representing approximately 30% of worldwide usage—a more than twofold increase since early 2025. Providers like DeepSeek, Alibaba’s Qwen, and Moonshot AI have rapidly gained traction, with DeepSeek alone processing over 14 trillion tokens.
Simplified Chinese has become the second most common language for AI interactions globally, constituting 5% of total usage behind English at 83%. Asia’s share of AI spending more than doubled, rising from 13% to 31%, with Singapore emerging as the world’s second-largest AI usage market after the United States.
The Emergence of Agentic AI and Complex Reasoning Tasks
One of the most transformative developments identified is the rise of “agentic” AI—models that execute multi-step tasks, utilize external tools, and engage in extended reasoning processes rather than merely providing isolated answers.
Interactions optimized for reasoning jumped from negligible levels at the start of 2025 to over 50% by year-end. This shift marks a transition from AI as a simple text generator to autonomous agents capable of planning, execution, and continuous contextual understanding.
For example, instead of requesting AI to “write a function,” users now task it with debugging entire codebases, identifying bottlenecks, and implementing solutions—demonstrating remarkable advances in AI capabilities.
The “Glass Slipper Effect” in User Retention
The study introduces the concept of the “Glass Slipper Effect,” where AI models that are first to effectively solve critical user problems retain users longer and build durable loyalty. Early adopters of Google’s Gemini 2.5 Pro in June 2025, for instance, maintained a 40% retention rate after five months, outperforming later user cohorts.
This finding refines the understanding of AI competition, emphasizing that being first to market is less important than being first to solve significant challenges, which embeds models deeply into user workflows and raises switching costs.
Price Sensitivity in AI Usage Is Limited
Contrary to assumptions about cost driving AI adoption, the research shows AI usage displays relative price inelasticity. A 10% price drop results in only a 0.5% to 0.7% increase in usage.
Premium models from Anthropic and OpenAI charge between $2 and $35 per million tokens yet sustain high usage, while budget-friendly models like DeepSeek and Google’s Gemini Flash operate below $0.40 per million tokens, coexisting successfully in the market.
This illustrates that AI users value quality, reliability, and capability over mere cost, and the LLM market has yet to commoditize fully.
Implications for the Future of AI
The OpenRouter study underscores that real-world AI usage is far more varied and complex than industry narratives suggest. While AI is undeniably transforming professional programming and productivity, it also fosters new forms of human-computer interaction through roleplay and creative outlets.
Geographically, the market is diversifying with growing influence from China and other Asian countries. Technologically, AI is evolving from simple text generation toward sophisticated agentic systems capable of multi-step reasoning and tool integration.
Moreover, success in AI depends on solving real problems effectively, not just on being first or cheapest. As AI becomes increasingly embedded in daily life, understanding these nuanced usage patterns will be critical for developers, businesses, and policymakers.
Related Articles
Anthropic’s Claude AI Experiences Major Outage Affecting Users Worldwide
OpenAI Secures $110 Billion in Historic Private Funding Round
OpenAI Unveils GPT-5.3-Codex-Spark, a High-Speed Coding Model Optimized for Real-Time Programming
Anthropic Launches Searchable Library of Practical Use Cases for Claude Generative AI
