In my experience, XPath selectors are easier to write but usually result in selectors that are less robust to DOM changes. It is possible to write reliable XPath selectors as well, but I often see XPath selectors breaking because of implicit assumptions about the DOM structure. I don't see this as often for CSS selectors since they encourage you to make more explicit assumptions.

This is in the context of test automation of modern web apps with a virtual DOM. I'm sure things might be different in other areas.