New Lefty Notes. Every two days Nova, the AI we built, asks ChatGPT, Claude and Perplexity the questions a customer would ask about our clients. Since June that is 30,833 answers, and for months we kept one word of each, yes or no. This post is the tour of what we built to read the rest. The Foundry takes every answer, crawler log and site change and asks why a client got named, who got named instead, and where the AI went to decide. The Lab puts every prediction in writing with the result that would kill it, and weighs competing explanations using Richards Heuer's Analysis of Competing Hypotheses. Two of our own ideas died there. The Casebook keeps what survives, and the strongest become research papers with a permanent ID. Why bother: a 2026 survey of the field found no technique with a lasting cross-platform effect on getting found, and Stanford's AI Index shows the tools moving faster than any guide can keep up with.

https://leftymediaco.substack.com/p/why-we-research-ai-search