What the major models recommend when someone asks what to buy — and how that moves as brands get better at being the answer. Each question is asked many times per model, once from the model's own weights and once with web search on. The gap between those two is where GEO is being won today.
Access
Password required
Panel
Questions
Models
* Modes
this calls every model for real and spends OpenRouter credit. run it now?
Who gets recommended, across every model and question
Weights vs. web search
How often each brand is recommended with the model answering from training alone, against the same question with search enabled. A brand far ahead on the right is winning on content the model can retrieve — the thing GEO work actually buys. Marked significant only where the difference clears a two-proportion test.
from weightswith web search
By model
Share of voice per brand per model. A brand carried by one model is a fragile position however high its overall share.
Where the answers come from
Domains cited in the web-search arm, as a share of those answers. This is the practical read: it's where content has to exist to be picked up.
Spot check
A spread of raw answers beside the brands extraction pulled from each one. The charts above are only as good as this reading, so it is worth skimming after a run — especially a first run on a new model.