Adding - triple check / proofread / rephrase the ai output. Assume the words may be used against you. If your manager is close with whomever reads the feedback, they could ask for “evidence” of any claims. You either need strong evidence, or to avoid any concrete claims. More vague more better / more defensible.