# How this site works, and why it has a map of itself

> The thinking behind this website: a graph of everything on it, a layer written for AI assistants, publishing straight from my notes, and an assistant that answers questions.

Author: Arun Wilson (https://arunwilson.co.uk/about)
Published: 2026-10-14
Canonical: https://arunwilson.co.uk/writing/how-this-site-works

This site has the usual jobs: say what I do, show the work, make it easy to get in touch. It's also a working example of the things I build for clients, so a few of its choices are worth explaining.

## The graph is the structure of the site, drawn

Every service, project and article here carries the same set of tags: skills and themes, defined once and shared everywhere. Those tags power the normal pages, such as the related work under each service. They also mean the whole site is a network, and the [graph](/graph) simply draws it.

I studied geography and I still think in maps. A graph like this is only a gimmick if the connections are made up. Because these come from the content itself, it's an honest picture of how the work fits together.

## It's written for AI assistants too

More and more people ask ChatGPT, Claude or Perplexity before they search. So alongside the normal pages there's [llms.txt](/llms.txt), a plain summary of who I am and what's here, and a full-text version for assistants that want more depth. Every article is also available as plain markdown. If you sell AI work, being easy for AI to read is the minimum.

## Publishing starts in my notes

I write in Obsidian. When a note is ready, I move it to a publishing folder, and a small script tidies the formatting, checks the tags and files it into the site. Case studies work the same way. No content management system, no copying and pasting, which is the difference between a blog that grows and one that stops after three posts.

## The assistant in the corner

The compass in the bottom corner is an assistant that answers questions about my work using only what's published here. It can point you to the right page, suggest whether a call makes sense, or pass a message on to me. It runs on a small, cheap model with tight limits, which is how I'd set one up for a client too.

## The general principle

None of this needed a big budget. It needed the basics done properly: clear structure, fast pages, structured data, and content treated as a system rather than a pile of pages.