Earlier quoted context omitted.
PDF:s are not great on mobile. And you can’t easily translate them (I often translate restaurant menus when they are on a website with just 2 clicks)
Translating anything that renders on my screen is the same two clicks to open an LLM with the screen contents. I expect that will become an increasingly universal experience as LLM features get shoved into every nook and cranny of tech.
For example: if there's a dish name with a 2 line description below it and some allergy symbols below that, in HTML you can imagine the document structure that produces that. In PDF terms that might be 4 separate objects and, in particular, the eyes can see the two lines are adjacent so they fit together but the document structure doesn't really represent it taht way, necessarily.
This might also not work with translation because the lines are set for the size of the text they contain. Same for resizing the font.
Put another waay, PDF should be viewed as a typeset and layout format, not a document format.