Headings, buttons, links and form controls carry built-in meaning and behavior when used appropriately. This helps tools interpret a document and can reduce the custom code needed to recreate standard interactions.
Correct elements are a foundation, not a complete accessibility guarantee. Labels, focus behavior, understandable content and the surrounding interaction still need to be designed and tested.
ELI5
Think of labeling drawers by what they contain instead of calling every drawer a box. Meaningful labels help both people and tools understand the arrangement.
For example, an HTML button already communicates that it is an interactive control. Styling a plain block to look like a button does not automatically give it the same keyboard behavior.
Does semantic HTML remove the need for testing?
No. The complete interaction and its labels, focus behavior and content still need verification.
Why prefer native controls?
They provide established meaning and behavior that custom imitations must otherwise recreate correctly.

