Should AI ever give mental health “advice”?

As someone building AI for emotional support, I struggle with the ethical lines. Should we design bots to just reflect or also to guide users emotionally? Curious what devs and ethicists here think.