AI can retrieve and generate answers, but the quality of those answers still depends on the information beneath the model. If content is poorly structured, weakly governed, or divided into the wrong units, a more capable LLM does not automatically solve the problem.
Patrick Bosek, co-founder and CEO of Heretto, joins me to examine how structured content shapes AI context. We begin by defining structured content and comparing the different approaches, from DITA and XML to headless content management systems. Patrick explains why the right structure depends on scale, interoperability, reuse, and the ways an organization needs to deliver information.
This conversation builds on episodes Documentation Is Becoming the Interface and The Work AI Cannot See. If documentation is becoming a surface through which people interact with products—and AI cannot infer context it has never received—then content architecture becomes part of the AI system itself. This episode moves one layer deeper to examine that foundation.
Patrick walks through chunking, retrieval-augmented generation (RAG), vector search, metadata, and knowledge graphs without treating AI as a replacement for established information practices. Metadata still provides deterministic controls. Traditional search may remain the better choice in some situations. Governance and provenance still matter when content comes from product, engineering, support, and documentation teams.
We also consider the practical side. A lone writer can begin with a small style guide, consistent Markdown, useful metadata, and an AI agent that reviews content against established rules. At enterprise scale, however, tools alone are not enough. Information modeling, information architecture, governance, and deliberate implementation have to come first.
In this episode, we discuss:
Season 8 of Inside Tech Comm is sponsored by LavaCon. Use discount code ITC26 to save $200 on registration for LavaCon 2026.
Show Credits
Interested in being on the show? Let's connect.
