I feel like I've been thinking about how to test the emoji aspect of this wrong. I was ignoring a key part of all of the companies' explanation for emoji reactions, which is that they're fun.
Really I should be taking inspiration from experiments that create a compelling experience that provides some value (entertainment, maybe) and still capture data that can be used to provide insight. One great example is the Columbia Music Lab study, which was a website where people could download tracks of music. Some visitors saw the number of downloads each track had, some did not. Resulted in an n of over 14,000, with no cash spent.