This is wrong.
a) The article literally states we did the analysis with AI. We used semantic search to ID paragraphs relating to a topic, then grouped text into categories by their meaning.
b) The model outputs were, as below, boolean indicators about whether something fit a category.