Your best bet, as toolic and apl suggest, is to use an XML parser. Manually parsing XML is fraught with potential bugs and edge cases. However, assuming your XML is well-formed, you can accomplish this task using Look Around Assertions. Specifically, if you assume that your text field contains no < characters, you can do it with

s/EM(?=[^<]*</article>)/replacement/g.

This will match and replace any occurence of the letters EM that are followed by any number of non-< characters and then a closing article tag.