That's an interesting point, but I only think it's true under the correct assumptions. If we assume that all bodies are more or less interchangeable, then yes. You'd want to look at large scale, repeatable studies that tie diet into health outcomes, and just follow those.

If, however, there is a large amount of variation among human bodies, large scale objective studies may not hold much value for any particular individual. In that case, the proper bounds of the experiment might be one person, or a small group of similar people. In that case, trying various experimental diets and observing the effects is quite scientific, even if not universal.

No, that is LITERALLY science. You come up with hypotheses and test them.

Hypothesis: [Specific diet] will cause no effects on myself, given my unique biochemicalpsycho makeup.

Hypothesis: [Specific diet] will cause [list of benefits/mali] for myself, given my unique biochemicalpsycho makeup.

etc.

Just because it doesn't automatically apply to the population at large does not make it "not science".

Well I imagine what OP meant was that the testing phase in this "science" is not verifiable/repeatable/etc. Perception is an awful way to perform tests, and is very frequently wrong.