Notes on building the AI-native web.
Page 4 of 17
How to structure a page so AI search will quote it
Being quoted is a formatting decision as much as a content one. The anatomy of a page that models can lift from cleanly.
Choosing a database, without the religion
For most business software the answer is a relational database. The interesting questions are the ones after that.
Data migrations: how to move without losing a week
The migration is never the hard part. Reconciling what arrived against what left is where the week goes.
When your AI feature has to explain itself
A customer disputes a decision, or a regulator asks how it was made. Explainability is an architecture choice made long before the question.
Does a marketing site need search?
Under thirty pages, almost certainly not — a search box is a confession that your navigation failed. Past a hundred, it’s essential.
Should an AI employee handle complaints?
It can process them competently and shouldn’t answer them alone. The distinction is between the work and the relationship.
When your AI employee gets it wrong in front of a customer
It will happen. Whether it becomes a complaint or a non-event is decided by what you built before it did.
Which contact options should you actually offer?
Offering every channel looks generous and creates four queues you answer inconsistently. Choose deliberately.
How to write the job description for an AI employee
Write it like you’re hiring. The exercise surfaces every ambiguity that would otherwise become a production incident.
Content pruning: what to delete
Publishing is easy and removing is frightening. But thin, stale and duplicate pages actively dilute the ones you want cited.
Should your AI employee have a name?
A name is fine and useful. A fake person is not. The line is clearer than the debate suggests.
What to actually measure on a marketing site
Sessions and bounce rate feel like measurement and change no decisions. Five numbers that do.