What is the moat of your AI products?

It seems that more and more AI products are entering the market this year, but they appear to be so similar, with almost no difference. As AI product design becomes increasingly homogenized, how can we build our own moat? I believe data is crucial. And product design should dig deep into real user needs and scenarios. Does anyone have any ideas?