Calvano, EmilioPossnig, ClemensTolvanen, Juha2026-06-102026-06-102026-02-12https://hdl.handle.net/10012/23582We analyze strategic communication when advice is generated by a reinforcement-learning algorithm rather than by a fully rational sender. Building on the cheap-talk framework of Crawford and Sobel (1982), an advisor adapts its messages based on payoff feedback, while a decision maker best-responds. We provide a theoretical analysis of the long-run communication outcomes induced by such reward-driven adaptation. With aligned preferences, we establish that learning robustly leads to informative communication even from uninformative initial policies. With misaligned preferences, no stable outcome exists; instead, learning generates cycles that sustain highly informative communication and payoffs exceeding those of any static equilibrium.enPreprint