Because the standard gave the element semantics that would retroactively justify this presentation.
Here’s a sad truth:
<small> is not actually a semantic element, not really. It merely pretends to be one. When it was first standardised in HTML 3.2, the
<small> tag was purely presentational: its definition stated simply ‘SMALL places text in a small font’. Back then, the separation of presentation and semantics wasn’t considered important.
The big push towards semantic markup started with HTML 4. The W3C released this standard in three versions: Frameset, Transitional and Strict, and it was intended that everyone would eventually move to the strict version, which removed most presentational elements. A couple presentational tags were, however, left in, like
<small>, but it was intended that a future standard would replace them with purely semantic elements. And indeed XHTML2 did away with
<small> completely. An intermediate standard, XHTML1, reformulated HTML 4 as an application of XML with no other changes.
This didn’t quite work out; adoption of the new standards was low. Web designers didn’t use XHTML for a number of reasons: it wasn’t supported very well by browsers, especially by the dominant Internet Explorer, which had no support for it at all. Another reason few people robust, which made them ill-suited to generating the much more strict XHTML, where any syntax error would prevent the page from rendering at all.
Browser makers didn’t much feel like implementing the XHTML standards either. Instead, they formed an organisation of their own, the WHATWG, and promoted an alternative standard, HTML 5, which embraced a philosophy of ‘backwards compatibility at all costs’. It defined the parsing of ill-formed markup based on reverse-engineering the behaviour of Internet Explorer and on de-facto usage. It also brought back presentational tags that W3C standards removed. To placate the proponents of semantic markup, the editors of the HTML5 standard redefined
<small> to be ‘semantic’; but the semantics they came up with are so vague that all they are good for is justifying the tags’ historical presentation.
In the end, ‘semantic web’ largely died out in the 2010s. These days, web designers predominantly approach hypertext as a purely visual medium, with no attention paid to how the page is interpreted other than in a GUI browser with the website-provided style sheet. The semantics of individual tags became largely irrelevant.