[mrandish]

When LLM results started being shoved into standard result pages, I immediately added them to my ever-growing Userscript, CSS mods and uBO filters for https://www.google.com/search*.

That said, I DO sometimes prompt explicit web searches via LLM because for certain searches, an LLM can be the best tool for the job. It's almost always cases where it's hard to target terms with inclusions, exclusions and booleans (even if Google hadn't partially nerfed them). This is usually due to English language ambiguities in word meaning and LLMs are actually ideal for that because they already have discrete tokens for "Wear" (as in clothes) and "Wear" (as in tires).

My conflict with Google product managers arises because I'm an advanced search user and I know when an LLM is the best tool for a search and when it's not. And when it's not, LLM results are always cognitive noise that just gets in the way and eats screen-space. I understand that most users may not be able to reliably make that distinction but Google refuses to offer any option for users like me to enable a functional experience, even buried in an advanced menu.

I don't think this gap is just monetization driven. I've worked as a senior group PM in a FAANG-ish tech company and I suspect the problem is related to a deeper issue I saw in many UX designers and product managers - an inability to accurately model the frequency and quantity of ways "automatic" or default options can fail to serve users. It's an insidious issue because it tends to appear as the absence of signal in normal interface usage analytics. And it's hard to detect in focus group and observational studies, unless you aggressively look for it. And they usually don't because there's an innate bias pushing them toward believing the automatic "it just works" solution serves everyone as well as they've designed it to. They insist it must be right because it's exactly the sort of elegant, simple solution their D-School profs graded highly.