Well, that's in some sense what these new large language models are all about: if all I want is the information from a website, it makes much more sense for me

Well, that's in some sense what these new large language models are all about: if all I want is the information from a website, it makes much more sense for me to have an AI read it and give me a summary than to wade through ads... but that means the only ad impressions are from AI bots... which means more attempts at enforcing that actual human eyeballs see ads... which means more captchas and more digital rights management. And, lo and behold, right around the time that this become an issue, we see Google starting to lead the charge on merging WebDRM into Chrome, and every social media company locking down their content with login walls and refusing alternative clients. This use case is just going to get bigger, though: the premise of ad blockers has a manifest destiny on just looking at the page and picking out the ads, which is possible and not particularly difficult to do even in the ridiculous case of a web page returning its content as a single giant PNG file, even if the ad is a watermark, and even if the "ad" is just a bunch of conspicuous mentions of Mountain Dew strewn throughout every time a placeholder noun is required. (I am legally obligated to note here that this comment was sponsored by Mountain Dew.)