The issue may be with the threshold you put on it. How you define whether it's accurate or not?
You could have a hypothetical typology that randomly assigns people to X and Y type (two of them) and still get at least 50%. If you describe the types vaguely, you can get it to even 90% of people reporting to be accurately typed.
Some people are satisfied with their horoscopes. Don't forget that.
I'd never consider it scientific and I wouldn't like psychologist examining me to use this thing. Nah, it's far from good enough for such use. Too much variety within types and too much uncertainty between them.
The goal is the key here. For ordinary people for fun or casual use it can be good enough. Certainly better than horoscopes because it takes some meaningful input to throw output based on it.