We only pick the one that have a structure that is easy to visualize. Which usually looks like headings and sub-headings or a list of things.

see the reply above ^

All the work was done manually. We just selected those essays which has a structure, summarized either in bullet points or headings. We edited each fact, reducing text, because with icon you don't need as much text. Regarding automation I experimented with text extraction (using abstractive text summarization) but haven't got good results yet.