AI Usage Statement
Where AI does the work on this site, where it deliberately doesn't, and how the whole thing gets built.
Last updated: 2026-08-22
How this site is built and maintained
This site is built and maintained with Claude, Anthropic's AI assistant — including the August 2026 rebuild that replaced thirty-three hand-edited pages with the templated site you're reading now. Claude writes the templates, the build tooling, and most of the code in this repository, and works alongside Neil on the copy across the service pages, the sector briefings, and pages like this one.
None of it publishes itself. Every change — code or copy — is directed by Neil, reviewed by Neil, and approved by Neil before it goes anywhere near the live site. The ideas, the research, the argument, and the final sign-off are his.
Where AI actually runs on this site
One feature calls an AI model directly, live, while you're using it: Your Plan, in the Practice Library.
You describe a situation in a sentence or two. That description is sent to Claude (Anthropic's AI model — currently one of its Sonnet-generation models), via Netlify's AI Gateway, in two steps:
- A scope check first. A quick call decides whether what you've described is something the Practice Library's actual published content can help with. If it isn't — or if it reads as needing real professional support rather than a reading list — the tool says so plainly and stops there, before generating anything
- Then, only if that passes, a plan. Claude is given your description and the complete list of items currently published in the library, and asked to select three to six that genuinely fit and explain why, in each case, in its own words
Before anything reaches your screen, every item Claude selects is checked against the real, published library. Anything that doesn't resolve to a card that actually exists is dropped. The tool cannot show you advice it invented — only advice already published here, chosen and explained for your situation.
What you type is used to generate that one response and isn't stored by this site afterwards. Netlify and Anthropic may keep standard request logs for a limited period under their own infrastructure policies, the same as any API call — see the Privacy Policy for how third-party providers are treated more generally.
Where it deliberately doesn't
Two things on this site that might look AI-powered aren't:
- The contact form. Nothing you send through it is read, sorted, or drafted by AI. It goes straight to Neil's own inbox as plain email, via Microsoft Graph — no AI is involved at any point
- Predictive Purpose, the self-assessment diagnostic linked from the Practice Library, runs entirely in your own browser. Its forty statements and its scoring are fixed — transcribed from the book's appendix — and every calculation happens on your device. Nothing is sent to a server at all: your answers are saved only to your own browser's local storage, so you can close the page and come back to it later, and no AI is used anywhere in it
Questions
If anything here is unclear, or you want more detail on how a specific part of this site works, email nc@neilcatton.com.
Review cadence
This statement is reviewed whenever how the site uses AI changes materially, and at least once a year regardless. The date at the top of this page shows when it was last revised.