[jimrandomh]

Never start from a news article, unless your goal is to help others by debunking it. News articles rarely reflect the research they claim to report on, and even if they did, they'd be reporting on a highly skewed subset of the research.

Read sources that are intended for people who genuinely care about getting the right answer. For exercise, that means writing where the target audience is professional athletes and their coaches. For nutrition, try to start from Wikipedia (if you don't have a specific question) or PubMed (if you do). Carefully avoid anything that's focused on weight loss, because that's where the crackpots, laymen and officially-respectable conmen are.

Japan has the longest life expectancy and the lowest obesity rate. Every time you see a claim about what a good diet is, check it against the Japanese diet; if the Japanese are doing the supposedly-bad thing, reject both the advice and its source.

[perl4ever]

I'm not sure whether your reasoning about Japan is a good way to evaluate things. That's because Japan could be different because of a particular thing out of all the possible factors. The reason I say that is because I am thinking of the effect of lead exposure on crime rates as an analogy. There are a million things that could and probably do affect crime, but lead in the environment seems to dominate. If the Japanese are particularly healthy, it could be to one particular factor, say fish oil consumption (just to be arbitrary) and everything else might conceivably be irrelevant or harmful. Assuming the "goodness" is distributed equally among the various characteristics of the typical diet could lead you astray when you focus on something that doesn't matter or is harmful. It's the same issue as if you observe some people living a very long time while smoking - is smoking contributing to their lifespan or working against it?

[jimrandomh]

It's not a completely reliable heuristic, but it's easy to check, and the things it's flagged that I've checked have turned out to be bullshit. (Sometimes highly respectable, promoted-by-mainstream-institutions bullshit, but not supported by quality studies if you read carefully.)