We measure what the model won't tell you.

Selectorate started with a pattern the founder kept seeing from two angles. First, doing interpretability research on closed-source models at Princeton — getting a rare look at how these systems actually reason inside. Then, working as an ML engineer in GEO/AEO, watching those same systems quietly decide which products got discovered and used.

The through-line: models were already shaping real product outcomes, and the companies affected had almost no visibility into what the model was doing or why. Search had SEO. Answer engines got AEO. But the fastest-growing channel — a coding agent choosing and wiring up your product on a developer's behalf — had nothing.

So we built the instrument. We run real agents against real products, measure the decision, and hand teams the transcript. Not a score, not a checklist — the actual moment the agent chose, and a ranked list of what to change so it chooses you next time.