Why this research exists
AI engines rewrite their rules constantly. If your strategy is chasing the current rules, you are permanently one update behind. Our research question is different: which properties of content keep getting it found and cited even after the engine changes? Those invariant conditions — not this quarter's tricks — are what we are trying to measure.
The method: a hybrid model
The transformer side handles content representation — turning pages into features a model can reason about. The reinforcement learning side closes the loop: the reward signal comes from real citation outcomes across mainstream AI engines, not from hand-labeled guesses. The model learns which feature combinations actually buy citations, and keeps learning as the engines move.
Where it stands
The data pipeline and baseline models are running; we are now accumulating observations across engine update cycles. This page will stay honest: no numbers until we can stand behind them. When the findings hold up, we publish them.
What this means for clients
The GEO recommendations inside our marketing-growth practice are fed by this research pipeline — measured conditions, not folklore. Working with us means your visibility strategy updates when the evidence does. The research is still running, but the method has already done real work: a service business with fifteen years in its local market — six figures a month on marketing, beautiful traffic reports, and consistently zero customers from them — saw overall customer visits grow 109% in the first month after rollout, and over half of its traceable customers now arrive through AI search.
