XML or HTML with an 'is_html' property.  If this property doesn't exist,
the default content-type-sniffing behavior is used.