Most studies use self-reported behaviors for how people behave. More invasive studies (as some people have mentioned in this comment section) show that self reporting is inaccurate and, for example, people eat many more calories than they self report. If you have some system of automatic recording of behavior, which smart-phones might be able to do, and individual DNA results, then these large scale studies might be able to tease out what is healthy behavior for people with specific DNA.
Googling "problems with self-reporting in diet studies" gives a good listing of scientific papers for me. Of course such a search is screaming out "confirmation bias", but it is still useful I think. This[1] is one study that seems to cover what I was talking about, although I have not read it in full.
If you want to take a deep dive into a subject, search for a recent paper on the subject and as you read it the paper will reference other papers, especially in the introduction. Here is where you can find possibly a more relevant paper on the subject. Read that paper and repeat. Soon you will have a decent idea of what is going on in the field. Hard to do if you don't know the jargon and the mathematical methods though.
