Overcoming common‐pool resource dilemmas requires learning across different sectors of society. However, policy actors frequently entrench themselves in so‐called echo chambers by preferring to rely on information from those whose policy beliefs resemble their own. Policy forums can reduce the limiting effects of echo chambers by encouraging actors with diverse knowledge bases to exchange information and learn from one another. This paper applies exponential random graph models to network data from the South African tree plantation policy domain to investigate how belief homophily, reputational influence, and forum co‐participation shape information exchange behavior. Results show that echo chambers are important determinants of information exchange ties and that reputational influence is likely to "deepen" the echo. Results also show that the more forums that a pair of actors co‐participate in, the more likely they are to exchange information. This applies to information exchange generally, as well as information exchange with trusted partners. Findings indicate that forums enable both cognitive learning (as knowledge gains) and relational learning (as improved relations). Nonetheless, when echo chambers are strong, and many forums are polarized, then forum co‐participation may not break up echo chambers.