Kevin White’s Post

Turns out good ol’ HTML is the preferred language of LLMs. Over the last few weeks, both Google and Anthropic have more or less said so themselves. Google’s recent guidance on AI search optimization (and the broader shift toward AI-native search experiences coming out of I/O) makes it pretty explicit what these systems want: -- clear structure -- semantic HTML -- readable content -- crawlable content -- minimal technical cruft -- strong technical foundations They’ve also been pretty clear on what doesn’t matter nearly as much, dispelling AI search myths like: -- llms.txt (does nothing for ya) -- “chunking” content (models already parse nuance) -- rewriting just for AI (unnecessary step as models interpret meaning) What’s more, the HTML train is not just coming out of Mountain View. On the latest “How I AI” podcast, Thariq Shihipar, an engineer at Anthropic, shared how HTML is replacing Markdown as the preferred interface for working with agents because it’s more structured, navigable, and easier for models to reason at scale. This is something we’ve been advocating at Scrunch for a long time now (linking our CTO’s post from February in the comments). In practice, that means: ✦ clear structure and semantic hierarchy ✦ minimal technical or navigational cruft ✦ text alternatives for rich media ✦ complete server-rendered content visibility ✦ schema reflected in readable page content ✦ information-dense, comprehensive pages With this in mind, we built AXP (Agent Experience Platform)—an optimized experience layer for a new kind of web visitor: AI agents. AXP automatically applies these best practices to the pages you already have (and the ones you create in the future) without forcing teams to rebuild their site architecture, maintain separate AI versions of content, or rethink how their website works. Not only that, it works in practice. Companies like Akamai are seeing massive lifts in AI visibility across AXP-optimized pages to the tune of: 364% growth in brand presence and 85% more citations.