For storytelling we summarize each exemplar as a social card. Typically, page authors supply metadata for the title, description, and striking image for each card. What if such metadata is not present? How do we approximate human behavior to fill these values? 🧵(19/)