I assume that it somehow scrapes the page for some sort of header text, but this doesn’t seem to always work.
For example, titles are suggested for spacenews.com articles, but not for europeanspaceflight.com articles. What differences result in suggested titles for one but not the other?
Could be. Though Lemmy could also use the title property (the one users see in their tab header) as a fallback option if it wanted to, or some other fallback.