Presumably this will only start to happen more when LLMs are being trained on this kind of data. For example, every training corpus weights Wikipedia way higher than random websites/forum posts, so sticking an ad for your product on some random article that no one looks at will get it into the model.