Generally, cause and effect can be only ascertain in the absence of residual confounding (and other criterias, such as temporality). Indeed, there are cohort structures which are more robust to residual confounding (fixed-effects, mendelian randomization, negative-control, hybrid-designs). Yet, this is rarely discussed when comparing study designs. Further, I would argue that a surprisingly large proportion of published RCT's are so flawed that they are unable to make any cause and effect inference.
In conclusion, I think the comparison between study designs are a simplification taught to undergraduate students. Arguably, we must assess every unique case, especially as causal assumptions may be more or less violated depending on the research field.