Alert message (AM) dissemination is a fundamental yet challenging issue in vehicular networks, as it relies on wireless transmissions in a highly mobile, potentially dense, and changing environment. Emerging network infrastructure-based vehicular networks are being considered as an alternative to Vehicular Ad-hoc NETworks (VANETs) for alert message dissemination. Indeed, assuming that Vehicle to Infrastructure (V2I) communication links are nominally available, with some transient and time-limited connectivity losses, recent alert message dissemination schemes primarily rely on V2I links to widely broadcast AMs. Vehicle to Vehicle (V2V) rebroadcasts performed by some selected relay vehicles located within pre-computed rebroadcast zones are then used to ensure the full dissemination within an area of interest. This paper focuses on rebroadcast zones placement. It proposes a Q-learning-based method that computes the minimum number and optimal locations of rebroadcast zones. From these computed zones, the combination of V2I broadcasts with V2V rebroadcasts allows the delivery of AMs in a whole area, even in the presence of locations with poor wireless connectivity. The performance results show that high information coverage and low delivery delays are achieved with our proposed Q-learning based placement. Useless duplicate rebroadcasts and collisions are also avoided, saving network resources.