In short
A confounding variable is a factor that is associated with both the exposure or treatment being studied and the outcome being measured, and that can create or distort an apparent association between the two even when no true causal relationship exists between them.
For a variable to confound a relationship, it must satisfy two conditions: it has to be associated with the exposure being studied, and it has to independently influence the outcome, without being purely a downstream consequence of the exposure itself. A classic textbook example is that people who carry a lighter tend to have higher lung cancer rates — but carrying a lighter doesn't cause cancer; smoking is the confounder, since smokers are more likely to carry lighters and smoking independently causes lung cancer.
In longevity and supplement research specifically, socioeconomic status, overall health-consciousness, and baseline health status are particularly common and powerful confounders: people who can afford and choose to take a premium longevity supplement, seek out a longevity clinic, or adopt an intensive wellness protocol also tend to have better healthcare access, lower stress, higher baseline fitness, and other advantages that independently predict better health outcomes, regardless of whether the specific intervention being studied does anything at all.
Well-designed studies address confounding through methods including randomization (which, done properly, distributes both known and unknown confounders evenly between groups — the core strength of an RCT), statistical adjustment for measured confounders in observational studies, and matching study groups on key characteristics — but statistical adjustment can only account for confounders the researchers thought to measure, which is why observational research, however carefully adjusted, cannot fully rule out confounding the way randomization can.