If you order by id as a tie breaker it's guaranteed that ever new inserted row has a higher id than the last id in the dataset.
If the id is a GUID it's possible to add a new row between the paginated API calls with a GUID lower than that of the last received row. Or does the cursor prevent that the new row is skipped?
What if the pagination is by order by id only?
Seems to me that you're assuming that everything should be paginated after being sorted by "ID". Why should it be? The user is not interested in your internal representation. An "ID" does not imply semantic meaning, or ordering, and you should not treat it as such.
If you want to order by creation time, then have a column with creation time. With resolution down to nanoseconds it'll be unique. And if not, well your query should be deterministic and "order by ctime,id".
But the problem is always possible. Doesn't matter what method you use. It applies to OFFSET … LIMIT, guid, name, or ID.
It's up to you to decide what you want the user experience to be if someone is on page 1, presses "next page", but before you click someone deleted the entire first page. And maybe there is no page 2 anymore.
These questions are application dependent, I'd say. And while I may have my opinions on the best way to handle it, it's an orthogonal question to what this article is talking about.
Edit: Oh, and there's GUID type 1, so "GUID" doesn't inherently mean "not sortable".
