What does semantic HTML give you that a div with ARIA attributes does not?

Native elements come with behavior as well as semantics: a button is focusable, fires on Enter and Space, participates in forms and exposes the right role and state for free. A div with role button gets the label but none of the behavior, so you add tabindex, key handling and disabled state yourself, and each one can go wrong. The first rule of ARIA is not to use ARIA when HTML already does the job.

ARIA only changes what gets exposed to the accessibility tree. It does not make anything focusable, it does not add key handling, and it does not manage focus. So the moment I write a div with role button I have signed up for tabindex zero, an Enter and Space handler, an aria disabled state that I also have to enforce in the handler, and I still do not get form submission. A real button gives me all of that and it stays correct when the browser changes. So my default is native first: button, a with a real href, input with a matching label, dialog for modals, details for simple disclosure. I reach for ARIA when there is genuinely no native equivalent, like a tab set or a combobox, and then I follow the authoring practices pattern rather than improvising roles. The bug I see most often is a clickable div in a table row that a keyboard user simply cannot reach, and it passes every automated check because nothing in the markup is technically wrong.