I came across this article on 10 Useful Usability Findings and Guidelines. Points listed are not too surprising. However, I disagree with #2. Essentially, the author is concluding that you can increase people's attention if you have a picture with a face looking at the target text. I think the "study" lacks the proper control to warrant such conclusion.
It could be that a frontward face draws attention away from the text. In other words, the frontward face has a negative affect on attention as opposed to directional face having a positive affect. A simple control stimulus with a face looking in the opposite direction of the text would have revealed whether a directional face actually increases attention. There should have also been a control stimulus of no face or a non-face image to establish a baseline.
The author is also concluding that glances directly translates into attention. The peer-reviewed eye tracking literature have shown that while this is generally true, there are many levels of attention that must be considered. The level of attention desired by content providers is one where the user is reading and comprehending the text, in other words, a high level of attentional processing. It could be the case that a significant number of the glances are accidental or reflexive glances as a result of being steered by the directional face. One way to tease out these accidental glances is to subtract the data from the first few glances across all the images. The glances remaining would be the more intentional ones, the ones that reflect higher level processing.
This is a demonstration of the lack of rigor in non-peer reviewed studies.