Experiments, including the ones that fail
Most of what gets written about AI search is repeated rather than tested. This section is the other thing: methods published in full, predictions written down before the data arrives, and results reported whichever way they land.
Studies
Each study has its own page holding the question, the predictions, the full method and, when it closes, the results. The pages it runs on are grouped with it.
Every AI crawler user agent, and what blocking each one actually does
Twenty two agents across seven companies, what each one controls, which two ignore robots.txt on purpose, and what blocking each one costs you. Built entirely on provider documentation.
How a page reaches an AI answer, and where it usually breaks
Fetch, keep, retrieve, generate. Three of the four steps are documented. The one in the middle is invisible from outside, and most advice quietly pretends otherwise.
What each AI provider actually documents about how it picks sources
Every official statement, quoted and linked, including the four things Google explicitly says you do not need to do. Plus the column marked "not addressed", which is the longest one.
Google lets you mark a site as a preferred source. Its own documentation says content from a source you have chosen is more likely to appear in Top Stories, and in AI Mode and AI Overviews.
It is one of the very few controls a publisher can point a reader at that touches AI answers directly, and it needs no news credentials or Google News listing. It also only does anything for people who actively click it, so this is a request rather than a tactic.
Every piece here follows the same structure: what I wanted to know, what I predicted before the data, the method in enough detail that someone could repeat it and catch me out, the results, what the results do not prove, and what I changed in my own work because of it. The fifth section is the one most studies skip.