My observations: it can be for both reasons, though I think most modern ones will be better than the human average unless you ask them to add typos. Which they can do, but they have a hard time being consistent with it, especially if they’re trying to pose as the same person over multiple responses (e.g. a reddit or fedi bot), they won’t have the same type and frequency of typos that a human would have consistently across posts.