We understand the world through connections. When you learn something new, your brain doesn't file it away in isolation, it integrates it into existing knowledge, creating a web of relationships that gives meaning to individual facts. A data infrastructure provider recently shared with me that their metadata now signifacatlly outweighs their actual data, indicating that even in our information systems, relationships matter more than raw content. The internet is rich with these connections, but they're not always accessible or relevant. Now, as AI agents become primary consumers of web content, are we structuring our sites so these systems can build the same rich understanding that humans naturally create?

In web development, beyond creating content, webmasters must get it in front of their intended audience. In previous decades, websites evolved around search engine optimization (SEO), and that feedback loop influenced how we created content. Now AI will drive further changes in how we write and present ideas. Webmasters should maintain control over how their sites are perceived regardless of the visitor. This requires rethinking how we structure content and the context we provide. To explore this space, I'm experimenting with my own website, designing for all audiences. I'll help you refine this article. Here's my edited version with improvements to clarity, flow, and impact:

Human-Centric Web Design

A beautiful website has traditionally combined attractive visuals, thoughtful layout, and elegant prose tailored to its audience. Or, when optimized for traffic, SEO drove keyword stuffing and link building tactics designed to game search algorithms. But AI agents don't think like search crawlers. They ingest the context, relationships, and semantic meaning. They are trained to know not just what your content says, but what it means and how it connects to everything else.

Websites optimized for human consumption are typically built on frameworks that benefit developers. But examine the underlying CSS and JavaScript, and they're hardly human-readable. LLM agents trying to understand your content must parse through navigation menus, sidebar widgets, and footer links. With minimal extra work, websites can be beautiful for both humans and machines. New standards, like llm.txt, are adapting the web for AI. We've spent years developing languages to make sites look beautiful to humans, then years more developing algorithms to parse them for search engines. Now its time to make them understandable to all.

AI-Centric Design

An AI-readable site should allow LLMs to make connections and understand how content fits into a broader context. This requires structured data and semantic markup to make a metadata layer that makes content LLM-readable without affecting the human experience.

Schema.org markup is an established standard that transforms raw HTML into a rich knowledge graph with semantic meaning. However, a new approach using an llms.txt file is gaining traction, championed by Jeremy Howard from Answer.AI (previously of Kaggle and fast.ai), designed specifically for LLM agents.

The llms.txt standard is based on markdown—simple, human-readable structured content that's easy for LLMs to process (also my favorite text format which I use daily). Its widespread adoption means language models are already trained to understand markdown formatting. Importantly, markdown maintains semantic structure through headers, lists, and other elements that organize information hierarchically, allowing LLMs to quickly navigate and extract relevant details. Markdown is uniquely suited for contexts where both human authors and AI systems need efficient access to the same information. It has been adopted by developers for documentation and wikis making that content easily searchable by AI coding agents.

My Schema.org Implementation

Article Schema: My blog posts now include structured metadata—headline, author, publication date, description, and word count. AI agents can instantly understand content hierarchy and attribution.

Organization Schema: The site gains an identity with contact information, social profiles, and a clear organizational description.

Website Schema: This enables AI systems to understand how to search the site and what content they'll find.

The Emerging llms.txt Standard

While Schema.org handles individual pages, the llms.txt file addresses site-wide context. It provides a single markdown file that explains your entire site to AI agents.

My implementation includes:

  • A concise site overview and mission

  • Content categorization and recent posts

  • Technical implementation details

  • Usage guidelines for AI systems

  • Links to feeds and APIs

Think of it as the AI-directed preface for your entire site.

llms AI readable website

Practical Implementation

Since this is a static blog, I enhanced my static site generator to:

  • Generate JSON-LD structured data following the Schema.org format

  • Create comprehensive robots.txt files for AI crawler management

  • Implement dynamic llms.txt generation

Every build now generates fresh structured data reflecting current content. New blog posts automatically receive Article schema, the sitemap updates with proper change frequencies, and the llms.txt file reflects the latest content categorization.

Results and Future Implications

The structured data now validates in Google's Rich Results Test, giving AI agents multiple pathways to understand and cite my content appropriately. Most importantly, it allows me to influence how LLMs interpret my work.

These changes position the site for the emerging AI-first web. It's now accessible to:

  • Search engines that use AI to understand and surface content

  • Voice assistants that need structured data for accurate responses

  • Research tools that rely on semantic markup to attribute sources correctly

As a next step, I'll be integrating an AI filtering layer that lets users consume content in their preferred format while maintaining my editorial control.

The Dual Optimization Challenge

My primary goal is to reach other humans and share ideas. But I also want to experiment with these new channels of communication. That requires optimizing for both humans and AI and since LLMs have been trained on web content it's a relatively light lift.

The changes I made can actually benefit humans too. Proper structured data enhances rather than detracts. Rich snippets in search results look better. Clear content hierarchy helps both screen readers and AI agents navigate. Semantic markup improves accessibility. And this framework put my positioning of my messages at the forefront. Together, they create a better knowledge scaffold upon which ideas can be connected and created.