Hi,
I checked the database, because no HTML or any other codes should be present, those are dealt with within the index process.
There were indeed a lot of CSS tags and some other HTML related keywords. After debugging, I found that the Divi builder shortcodes were not rendered at all, printing effectively causing the issue.
After some research, I have found that there is no proper way of triggering them, only some hacks, and there are also known bugs with that, which the Divi team already knows about.
Anyways, I had to make a few direct changes to the plugin code to programmatically trigger the missing shortcodes from the Divi builder – now most of them seem to execute fine and their contents are being indexed.
After recreating the index, it shrinked to 1.6M from 1.9M, removing over 300K unwanted keywords.
I will make sure to include this change in the upcoming release, so you don’t have to worry about it. I sure hope Divi staff will address these issues as well, as these “hacky” solutions usually don’t work forever.