I've been doing a lot of my own research into this, and one idea that kept presenting itself was the following question:
"Is it better to be direct, or indirect? Visual or verbal?"
With regards to how you want to present articles, by showing text first one shows content that has a very obvious and deliberate scope (for the subsequent body of the article), while the other - images on top - infers content within the article itself. The latter is always going to be slightly less explicit than the former in identifying what the scope is about, however... this is where your users could come in. Are they predominantly visually oriented, or verbally?
Overall if I were do venture a guess, I might say the verbal-first pattern is more quickly parsable to users when thinking about the precise scope of the article, but that images might be faster for general scope.
Anyway, something else to consider too is that a very common pattern you'll see on news orgs today is that they will have
[image] [ headline > attribution > teaser ]
but for the actual article page will swap those two blocks so the text is on top, or in some cases on top of the image itself (within, like some of The Verge longform pieces).