... so instead, you end with a bunch of observational studies (which cannot show causation), or studies of placeholders assuming it maps to something people actually care about (eg. "cholesterol levels in the blood stream after X hours" for "Y years of life lost to heart disease per hundred thousand"), or bench studies (some chemical component in broccoli can kill cancer cells in a petri dish becomes "broccoli fights cancer" ignoring the fact that eg. fire kills cancer cells in a petri dish... that doesn't mean you should eat it).
... which are subject to all the usual difficulties for associated with subjects that are hard to study: small samples, inadequate surveys, bad statistics, HARKing, p-hacking.
... the resulting papers are then reported on uncritically, often generalizing the claims well beyond what the underlying science would support... because people read articles about things they actually do (eat meat) and outcomes they care about (live longer/avoid heart disease)... not about bench results.