One of the biggest barriers to the semantic web is the barrier to entry. Scraping web pages are hard. Parsing HTML (which probably doesn't validate) is hard. Ex

One of the biggest barriers to the semantic web is the barrier to entry. Scraping web pages are hard. Parsing HTML (which probably doesn't validate) is hard. Extracting semantic meaning from parsed HTML is hard. Even once you've piled on the libraries and extracted the bit of information that you need, what do you do with that data? You process it a bit and store it in some kind of data structure. But at this point, you've could have just pinged the website's API and gotten the same data (and more) in a data structure. It turns out it's a heck of a lot easier to return a blob of JSON than it is to process text in markup on a page. And smaller, as well: JSON often takes up far less space than the corresponding markup for the same content. That's a big deal when you're processing a very large amount of information. There's the promise that AI will someday make this easier: if you eliminate the parse-and-process-into-a-data-structure step and just assume that an AI will do that part for you, you're in good shape. But that's nowhere near being a practical reality for virtually all developers, and APIs eliminate this step for you. Even if you use something like HTML microdata, there's very few consumers of the data. Some browsers give you access to it, but that doesn't make it extremely useful: if you generated the data on the server side, why not just make it into a useful interface? Or expose the data as raw data to begin with? Going through the extra effort to use these APIs is a redundant step for most use cases.