$ cat ai-seo-research.md # updated 10 Aug 2026

AI SEO Research

When you ask ChatGPT, Perplexity or Gemini a question, something decides which pages get read, quoted and credited, and none of them document how. So I read the raw network traffic underneath the answers, on my own logged-in accounts, and write down what each engine is actually doing.

Then I keep it current, because these systems change every few weeks and a dated article is stale the moment it publishes. Findings read straight off the wire are marked structural (high confidence, one clean capture proves them). Frequencies from my small, single-account sample are directional, the shape of the thing, not the measurement.

Pick an engine to see what is true now and every change since. Want to read it on your own session in one click? That is what FanoutFox does.

$ ls engines # 3 tracked

$ cat method.md

It is HTTP inspection after the browser decrypts the traffic, not packet sniffing. A small hook reads the JSON each engine streams to my own session as it answers, so nothing leaves my browser and no account but mine is touched. It grew out of three ChatGPT teardowns (part one, part two and part three), then became this because the wire keeps moving.

The taxonomy is the contribution, not the percentages. You can argue with a number; you cannot argue a field out of existence.