For a communication system to be stable, senders should convey honest information. Providing dishonest information, however, can be advantageous to senders, which imposes a constraint on the evolution of communication systems. Beyond single populations and bitrophic systems, one may ask whether stable communication systems can evolve in multitrophic systems. Consider cross-species signalling where herbivore-induced plant volatiles (HIPVs) attract predators to reduce the damage from arthropod herbivores. Such plant signals may be honest and help predators to identify profitable prey/plant types via HIPV composition and to assess prey density via the amount of HIPVs. There could be selection for dishonest signals that attract predators for protection from possible future herbivory. Recently, we described a case in which plants release a fixed, high amount of HIPVs independent of herbivore load, adopting what we labelled a ‘cry-wolf’ strategy. To understand when such signals evolve, we model coevolutionary interactions between plants, herbivores and predators, and show that both ‘honest’ and ‘cry-wolf’ types can emerge, depending on the assumed plant–herbivore encounter rates and herbivore population density. It is suggested that the ‘cry-wolf’ strategy may have evolved to reduce the risk of heavy damage in the future. Our model suggests that eco-evolutionary feedback loops involving a third species may have important consequences for the stability of this outcome.