Which part of text of a patent do you use for the semantic search?

Overall, we work on a family basis, therefore, we take the newest full text available for a family, but we have a certain prioritization on which authorities we take the text from. This is due to our search engines working in English and hence we try to use as many original English documents. These are simply preferable in quality compared to machine translations. E.g. if a family has a US and a ES family member, we would keep a US A1 text over an ES A2 text. The text we use would change, once there is a newer publication from one of the English-speaking countries e.g. if an AU B2 was published (or, of course, a US A2).