I find lineage-linked databases to be inadequate in cases like this, especially Ancestry's online trees.  In Family Historian, I prefer to link people as Associated Persons.  This allows me to record the relationship which is actually stated in the source, and I don't have to use a dummy relationship to attach them to the tree ahead of the evidence. 

On Ancestry's online tree, if two people have the same surname, and the source says they are brothers, I create a no-name father with the same surname and link them that way, but otherwise, I create the person as a descendant of someone in the tree and then unlink them, leaving them in the tree as an unrelated person.  

I don't like creating dummy intermediates for more distant relations, because there isn't any good way on Ancestry's online tree to flag those links as a hypothesis.  

Because I do not want to mis-lead other researchers, most of my Ancestry trees are private (the exception is a public tree for a same-name family which is NOT my family, created to demonstrate why I do not think their records belong to family I am studying).

For the cases where I have a hypothesis I want to test, I create a smaller private tree separate from my main tree.  This allows me to mine Ancestry's hint system without putting a false lead into my main tree.  Be sure to leave yourself some kind of note about what you have done -- Ancestry does allow you to make notes in the Tree Overview page, on each person's profile, and in the comments, so I encourage you to leave yourself and others breadcrumbs about what you are trying to do.  The real danger, as I see it, is that by creating too many dummy people, in addition to mis-leading other researchers about what you are doing, it is easy to forget what you have done, and mis-lead *yourself.*

This is why I prefer to keep material like this in research notes instead of 'cementing' the relationships with links in online trees.