A third of the web already written by models, corporate race shifts gear

A new Pew Research study reveals the extent of AI-generated text on the open web: more than a third of web pages published since the launch of ChatGPT show clear signs of AI writing. The cumulative figure for 2026 stands at 9.4% of all commercial-domain (.com) pages, versus 1% of academic (.edu) and 0.8% of government (.gov) domains. The gap reflects a direct economic incentive, as each additional page generates immediate revenue, making a drafting-accelerator tool a competitive advantage that is hard to forgo.

API-spend data published by Ramp show that in the current third quarter OpenAI is growing faster than Anthropic in corporate-client token spend, 82% versus 76% quarterly growth. Until now Anthropic led quarterly growth in every period since Q2 2025, sometimes by a wide margin. The reversal indicates that the launch of GPT-5.6 Sol is beginning to affect organizational purchasing decisions, at least in the short term. Ramp’s data are updated daily and have closely tracked reported revenue growth of the companies in prior periods.

TARS Robotics presents AWE 3.5, a foundational embodied model developed with a “Born as One” approach—a single architecture that unifies action, perception, geometry and touch already in early training, rather than stitching separate modules together later. The format includes two priors that give the model a basic understanding of action patterns, spatial structure and physics laws before adaptation to a specific robotic body, and advanced training in which a World Engine runs simulations of future states to select better actions. The company reports closed-loop reasoning capability up to a horizon of a few minutes and task-execution efficiency double that of PI0.5.

A change in OpenAI’s search mechanism caused Reddit to lose almost all presence in ChatGPT citations. The platform, which had been a central source of up-to-date human discussions, no longer appears as a routine source in model answers. The practical implication is a further reduction of access to authentic community content in favor of sources OpenAI prefers or has licensing agreements with.

The emerging picture points to two parallel processes: the open web is filling with synthetic content at an accelerating pace, while major model providers compete for corporate customers through performance improvements and closed-data agreements. For developers and product managers, the implication is twofold—reliance on the open web as a clean training or evaluation dataset becomes harder, and dependence on closed models with shifting data-retention policies grows.