For certain professions like media editing vector instructions help. But for your average Facebook / Netflix / Microsoft Word user, a kind of user that 95% users are, there are less benefits on vector instructions.
Are you saying Facebook, Netflix and Microsoft Word don't require media processing? Pretty sure you'd see plenty of SIMD instructions being executed in libraries called by those applications.
AVX is widely used in things as basic as string parsing. Does your application touch XML or JSON? Odds are good that it probably uses AVX.
Does your game use Denuvo? Then it straight-up won't run without AVX.
People are stuck in a 2012 mindset that AVX is some newfangled thing. It's not, it's used everywhere now. And it will be even more widely used once AVX-512 hits the market - even if you are not using 512-bit width, AVX-512 adds a bunch of new instruction types that fill in some gaps in the existing sets, and extend it with GPU-like features (lane masking).
Are you saying that iPhones and iPads are bad at Facebook, Netflix, and Microsoft Word? If they are, the end user certainly can’t tell. If they aren’t, then it doesn’t really matter does it?
Phones are much more reliant on having hardware decoders for things like video while desktops can usually get away with a CPU-based implementation, yes.
That's not really true. Single-threaded scalar performance is still super important for the everyday responsiveness of laptop/desktop systems. Especially for applications like web browsing which run JavaScript.