I recently audited top 25000 sites for security & seo. This answer is based on what I learned auditing those sites.
Though HTML is a structured, within html as such there is so much of unstructured data. I couldn't create a schema to store data extracted from these html, especially when you have to account for erroneous data out there in the wild.
Simple example: there are many sites that don't close the tags. What if you want to capture them and analyze for types of errors?
So to capture such data, use nosql.
But when I wanted to analyze the data, I couldn't use nosql. I had to depend on sql (may be because I only know sql queries). There is no empirical data to prove, but I believe sql dbs are faster for querying on thousands of rows of data (may be because they are declarative).
So experiential hypothesis: use nosql dbs for schemaless data; use sql dbs for analysis.