The difference between observational vs randomised trials is that one watches people and the other changes something and compares groups. Observational work can never establish cause. Randomised trials can, and they are also the design you are least likely to see quoted in a headline, because the results are usually smaller and considerably duller.
What each design can actually deliver
An observational study tracks a large group over years and reports that people with higher intake of something had a different rate of some outcome. It is cheap relative to a trial, it runs in real populations, and it can follow 50,000 people for 10 years, which no randomised trial will ever do. What it cannot do is rule out the thing that actually explains the difference.
People who take supplements daily differ from people who don't in dozens of ways. They tend to be older, wealthier, more health conscious, and more likely to see a doctor. Any of those could drive the outcome on its own. Randomisation breaks that link by assigning the intervention by chance, which is why a trial of 200 participants can outrank an observational cohort of 40,000.
Statistical adjustment helps and never fully solves it. You can only adjust for the confounders you thought to measure, and the ones nobody measured are the ones that hurt.
The headline maths favours the weaker design
Here is the mechanical reason headlines prefer observational work. Associations get reported as ratios that sound dramatic, and the language stays loose enough to imply causation. A cohort finding that a group had a 20 percent lower rate of something converts easily into linked to lower risk. A randomised trial typically reports a modest effect with a confidence interval that crosses into unimpressive territory, and the honest write-up has to include the words no significant difference.
The weaker design produces the better sentence almost every time. That is the whole mechanism, and it is not a conspiracy. It is an editor choosing the paper with the punchier result from a stack of 20.
Where observational work still earns its place
It generates hypotheses, and it does the job when randomisation would be unethical. Nobody is going to randomly assign 10,000 people to smoke for 20 years. Nobody is going to randomise people into a nutrient deficiency. For rare outcomes, long horizons, and questions where you cannot intervene, observational data is the only evidence that will ever exist.
Garlic is a decent example of the pattern in practice. Much of the widely quoted work on garlic and blood lipid markers is observational or comes from small trials using inconsistent preparations, and the honest reading is that effect sizes are modest and study quality varies a lot. Format matters enormously too. Aged extract, raw clove and garlic powder are not interchangeable inputs, and many trials do not fully describe what was used.
Duration in these studies is often short as well. A 4 week trial measures a shift in a marker, not a change in anything you would notice, and the supplements literature is full of 12 week windows being read as long term evidence.
Two questions that sort any claim
Ask whether the study intervened or observed. Then ask how many people and for how long. If the answer is a cohort of 30,000 followed for 12 years, you're reading an association, which is a hypothesis with good manners. If the answer is a randomised trial of 180 people over 12 weeks, you're reading a causal test at one dose in one defined group.
Neither is worthless. They answer different questions, and only one of them justifies the word causes. When you read a claim about a supplement, the design tells you what you're allowed to conclude before you read a single result.
Check the design first. The rest of the article gets easier to judge once you have.
These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.