We get 1-5% response rates (which is normal; I've done this on a few sites now) but we have enough scale where we're getting enough feedback that I can use the data to identify the worst docs. The basic methodology is to prioritize the pages with the most pageviews + lowest helpfulness scores. Of course there's a risk that we're prioritizing our work based on data that isn't indicative of our overall audience and due to the general nature of the question ("was this page helpful?") we also don't know how exactly the doc is unhelpful but in practice it's usually easy to make an educated guess (e.g. one of the Lighthouse guides is very low rated, and when I checked on it recently I discovered it's just a stub with very little guidance; so the educated guess is to add more guidance to the page; and then a few months from now I can see if the ratings have improved). Another good pattern is to provide a freeform textbox for feedback. The only reason we don't have that is because Gooble considers it PII and I have to jump through hoops to store that data properly.