Unlocking LLM User Habits: A Publisher’s Guide
▼ Summary
– LLMs are not replacing search engines but are shifting how people access information through conversational interfaces.
– The top three use cases—Practical Guidance, Seeking Information, and Writing—account for 80% of all ChatGPT conversations.
– Asking (49%) and Doing (40%) queries dominate usage, with Asking queries growing faster and rated higher in quality.
– Non-work-related usage of LLMs has increased significantly, from 53% to over 70% by July 2025, showing habitual adoption.
– Publishers need to create valuable, linkable assets and tools to address user intents like “Doing” rather than just publishing content.
Understanding how people actually use large language models is crucial for publishers navigating today’s digital landscape. The most comprehensive research to date reveals that LLMs aren’t replacing search engines but are fundamentally changing how people access information. Asking questions and completing tasks dominate user interactions, while practical guidance, information seeking, and writing assistance account for the overwhelming majority of conversations.
At their core, chatbots function as statistical models designed to generate text responses based on text inputs. The training process typically occurs in multiple stages. During initial pre-training, these systems learn to predict the next word in sequences, developing what might be described as a latent understanding of language patterns. The subsequent phase involves refining responses through techniques like reinforcement learning, where the model receives feedback to improve answer quality over time.
The concept of temperature settings plays a critical role in how these systems operate. Without temperature adjustments, LLMs would produce identical responses to the same prompts repeatedly. Higher temperature settings introduce more randomness and creativity, making them suitable for content creation tasks, while lower temperatures yield more predictable and precise outputs ideal for technical applications like coding.
Recent data provides fascinating insights into actual usage patterns. Rather than decreasing traditional search activity, frequent LLM users actually conduct more searches, supporting the theory that these tools expand rather than replace existing information-seeking behaviors.
Three primary use cases dominate the landscape. Practical guidance, straightforward information requests, and writing assistance collectively represent four-fifths of all human-AI interactions. Interestingly, most writing-related queries focus on editing existing content rather than creating new material from scratch.
Personal usage has seen remarkable growth, with non-work-related messages increasing significantly over recent years. These tools have become habitual companions for decision-making both inside and outside professional contexts. Within workplace environments, writing support remains the most common application, particularly for refining emails and documents.
Contrary to popular assumptions, computer programming represents a relatively small portion of overall usage. This suggests that specialized tools designed for specific domains often outperform general-purpose models for technical tasks.
User demographics have undergone substantial shifts. While early adopters skewed heavily male, the gender gap has narrowed considerably, reaching near parity in recent usage statistics. Age distribution reveals another important trend, with nearly half of all adult messages originating from users under twenty-six years old.
The nature of queries also provides valuable insights. Asking and doing-related messages constitute the vast majority of interactions, with asking-type queries growing faster and receiving higher quality ratings. Meanwhile, despite anecdotal reports of AI therapy sessions, relationship and personal reflection topics represent a minimal portion of actual usage.
For publishers, these findings suggest several strategic implications. The emergence of “doing” or “generating” as a distinct intent category represents a significant shift from traditional search patterns focused on navigation, information, commerce, and transactions. Success requires moving beyond simple content creation toward developing valuable tools and linkable assets that address these new user needs.
Building trust with younger, more AI-receptive audiences becomes increasingly important. The most effective approaches combine AI capabilities with human expertise, creating content and resources that machines cannot easily replicate. Programmatic SEO strategies and practical tools that repeatedly answer user questions can deliver substantial value in this evolving landscape.
The challenge lies in scaling quality while maintaining distinctive value. Publishers with established audiences, strong branding, and expert perspectives remain well-positioned to thrive by focusing on what AI cannot easily synthesize, authentic insights, verifiable information, and practical resources that genuinely serve user needs.
(Source: Search Engine Journal)
