Blog: copy, traffic and testing for sites built with AI
Practical writing for people who shipped a site with an AI coding tool and now want it to work. Each post is useful without IntentGrid, shows its working where there are numbers, and says where its advice runs out.
Last updated 2 October 2026.
Your landing page has no error message
A broken deploy tells you it is broken. Copy that nobody understands looks exactly like copy that works. Here is how to find out which one you have, and the order of work that avoids the problem.
Read the postTraffic problem or first-screen problem? How to tell
Before you go looking for more visitors, find out what the ones you have are doing. Split your numbers by where people came from, and the problem often turns out to be the first screen rather than the traffic.
Read the postStock phrases that make an AI-built site read as AI-built
Visitors do not read your source code to decide whether a model wrote your site. They read the first screen and recognise the defaults. Here are the common ones, and what to put in their place.
Read the postToo little traffic for an A/B test? What to do instead
An A/B test tells you which page performs better, with a known chance of being wrong. That guarantee is paid for in visitors. Here is the arithmetic, worked through, and what to do when you do not have them.
Read the postHow to write an llms.txt, and what it will and won't do
An llms.txt is a short Markdown file that tells a language model what your site is and where the important pages are. It is quick to write. Whether AI crawlers use it is a separate and much less certain question.
Read the post