The read-versus-cited split for Reddit (95% retrieved, 8 of 625 cited) is the most useful number here. It suggests Reddit shapes what the model thinks while docs and review sites get the credit. Did you see any difference for tools that had their own comparison or alternatives pages? Given that documentation-shaped pages get cited, that seems like the cheapest thing a new tool could test.