> SSRIs have an NNT around 7, depending on the study you look at [...] Which sounds terrible if you know nothing about NNT. But when you learn that Tylenol has an NNT of almost 5 and even a powerful drug like Xanax has an NNT of 4, you realize that NNT is a difficult measure of drug efficacy

This kind of comparison is utterly meaningless (like saying a Cohen's d of 0.3 is large for phenomena X, so if we see 0.4 in phenomena Y, it is large in Y), and is just one part of why NNT as usually reported is basically deceptive for antidepressants.

What qualifies as an effective "treatment" has to be anchored to a minimal important difference, and this is what antidepressants really don't clearly have, on average. I.e. saying the NNT of SSRIs is around 7 is not really practically interpretable, because they tend not to base NNTs here on the amount of patients that actually experienced a clinically meaningful change, they just count the number that exceed some arbitrary (usually purely statistical) threshold.

If you reformulate NNT competently to count number of people needed to be treated to ensure one has (on expectation) a minimally important difference in the depression scales, then the NNT gets even larger than is typically reported.