If you don't want to use metadata descriptions (btw, this is exactly what they are for), you have a lot of research and work to do. Essentially, you have to guess which part of the page is content and which is just navigation/fluff. Indeed, Google has exactly that; note however, that extracting valuable information from useless fluff is their #1 competency and they've been researching and improving that for a decade.
You can, of course, make an educated guess (e.g. "look for an element with ID or class maincontent
" and get the first paragraph from it) and maybe it will be OK. The real question is, how good do you want the results to be? (Facebook has something similar for linking to websites, sometimes the summary just insists that an ad is the main content).