The web was designed for human readers, but that assumption is breaking down as AI agents increasingly consume online content. According to a recent arXiv paper, the current architecture still ships content structured for human browsing, leaving publishers without a native mechanism to expose the underlying semantic units that machine readers need. The abstract notes this gap as a core problem, though the paper's full proposal is not detailed in the available text.
The authors frame the issue as a transition “from crawled to published.” Instead of relying on search-engine-style crawling to extract meaning from pages, they envision an architecture where publishers control how their content is made available to AI-native systems. This would give publishers a direct role in shaping what agents can access, rather than leaving it to external crawlers to infer structure and significance.
Because the source is a single arXiv preprint, the summary reflects only its stated motivation and direction. The abstract cuts off before explaining how this publisher-controlled model would be implemented, so the proposal's technical mechanics and feasibility remain unclear from the provided text alone.