The classic paper format is just ergonomically what many of us are good at handling effectively as readers. For example in ML typically they all have an abstract, a teaser figure with a caption, Fig. 2 with a method overview/architecture (boxes and arrows). An intro starting with the motivation and the problem with prior work, their key idea, their experimental evidence, then a dense restatement of the contributions as bullet points. Then related work overview, then the method description in detail, then the experiments, dataset descriptions, protocols, metrics, then the results and their interpretations, then the conclusion, i.e. what they conclude from the results.

Its fairly rigid and newcomers often complain that it's too repetitive but if you read such papers for years, you learn to very quickly navigate such a paper that adheres to these conventions and you quickly see if it's something you care about right now or not. Blog posts don't have the same formal structure and it makes the quick skimming and assessment much harder.