AI-friendly by design

The choice underneath every other choice: a CTXR site must be readable and operable by an agent — Claude, or any assistant — without a connector, a plugin or a special interface. Not a feature added on, but the test every part of the system has to pass.

The choice

Most systems became “AI-ready” by bolting an integration onto an existing shape: an API for the assistant, a plugin into the database. CTXR asked what a site looks like when being understood by a machine is a constraint from the start. The answer: plain files, a public vocabulary, no interface between the content and whoever works on it, and machine views generated from the same source as the pages.

It comes first because it is a test, not a goal. “Can an agent read this without help?” sorts design decisions:

To… Needs help Needs none
read the content a database, via a query layer files
understand the model a custom model, via its manual Schema.org
change the content an admin panel, via a second path the files directly
learn how the system works a manual somewhere else the docs, through the same door as the files

Each time, the AI-friendly answer was also the simpler one.

One shape, many consumers

The live data is valid JSON-LD in Schema.org types. That single fact has a large consequence: the data a visitor’s browser renders, the data a search engine indexes and the data a language model ingests are the same data, in the same shape. There is no separate “API response format” to design, maintain and keep in sync with the site.

                    ┌─ a browser renders it
Live data      ─────┼─ a search engine indexes it
(JSON-LD)           ├─ a language model reads it
                    └─ an archiver stores it

Because the vocabulary is the one these systems already train and operate on, your content is on-distribution for them. A language model needs no custom parser to know that a Place has a geo and a Review has a reviewRating — it has seen millions of them. Summarising, categorising and answering questions over your content works without a bespoke integration on either side.

What it buys

Search engines and AI assistants find, index and quote the site, because the content that renders the page is the content a machine reads — and every entity is an address of its own under /data, the file for code and a readable form beside it. See AI and search.

You manage the site in conversation, because the agent works on the same files a developer would. See How you manage your site.

Durability. Interfaces age; files and a public vocabulary do not. A CTXR site from today will be readable by whatever assistant exists in ten years.

What it costs

Structure. Content is a typed entity with fields, not free-form text with a title. That is what makes the page findable, the agent precise and the machine views true. Your agent carries most of that weight.

Why “AI” is not in the name

Friendliness to machines is a property of the design, not a feature that can be switched on or go out of fashion. The name says what the engine is: ground to build on.