It’s not useful to me because it’s not accurate. I already knew everything it told me about myself and recognized the parts that weren’t accurate, but what can I do with that information? I can’t expect that its conclusions about anyone else or how to relate to them are any more accurate.
I use them as Bayesian priors in my model of a person's behavior.
In my experience, when someone tells you they are x personality type, that's generally a starting point for you to refine your understanding of them as opposed to starting from zero. Unless they had meant to throw me off altogether, I generally find that information helps my mental model of them to converge fairly quickly.
Example: you ask a girl out and she tells you she self-identifies as INTJ, so you think to yourself -- she gets her energy by being alone or with a few people, she's probably a little cerebral and probably tends toward orderliness. You then have a basis of testing your hypothesis to see what her actual preferences are (vs stated), and also you might not want to plan your dates around meeting lots of strangers and doing crazy stuff. Remember, the utility of the MBTI isn't so much whether it is actually accurate or not -- it is more what it communicates. When someone gives you their type, it's them communicating to you a compressed signal that says, "I think I kinda fit this description" -- and that's a useful signal.
You may later learn that your model of her is wrong and needs refinement, but you'd be making iterations in small steps as opposed to throwing the entire model out. Once you know her, you no longer need MBTI or whatever, but during the initial stages, it definitely helps guide the exploration.
How is that not useful?
Sure, there are more complex instruments like Big5 and such that are more "accurate" but outside of research circles, no one uses them. That right there is an example of a more accurate, but less (day-to-day) useful signal.
(though I would say Big5 is good for understanding where you are on certain dimensions with respect to the general population. And it is undeniably useful in psychology.)
I’ve personally found it to be not that accurate but still tremendously useful.