Which attachment types can the AI suggested reply read, and how much?

For images it reads common formats (PNG, JPG, GIF and WEBP) and uses up to three images per ticket; very small images such as icons or tracking pixels are ignored, and when more than three qualify it keeps the largest ones. For documents it reads PDF, DOC and DOCX files, up to two per ticket, and uses a text excerpt from each (roughly the first several pages of a long PDF). Other file types are not read for the suggested reply.