There are several reasons for sampling with 96kHz: this allows for more natural results when doing non-linear processing (all kinds of distortion, modulation, pitch shifting) since that causes frequencies outside of the hearing range to have an effect into the hearing range.
It also makes digital IIR filters more comparable to their analog counterparts by having much less "frequency warping" in the hearable range.
And aliasing/sampling artifacts are much easier to keep out of the hearing range. Which is particularly important if you haven't decided on one of 48kHz or 44.1kHz for reproduction yet.
For reproduction, 44.1kHz is certainly enough. For sampling, you have comparatively cheap devices like handheld recorders where the 96kHz (and 24bit) sampling is more or less filling in fantasy data, and you have very expensive devices where the 48kHz variant is actually representing the 0-20kHz range of the input very accurately. It's the in-between area of quite-good equipment where doing the final sample rate reduction offline and with high-quality digital filters might also improve the results even when you are not doing any non-linear processing while editing.