Gossip Learning enables the automatic learning of a model without using a central server. This distributed control prompts the study of the possibility of malicious attacks by the participants themselves. While poisoning attacks on other forms of machine learning have been extensively studied, their effects on collective learning via gossip did not receive a similar level of attention yet. Our work is the first to propose a methodology for evaluating poisoning attacks in this context in scenarios with or without churn. Additionally, to assess our methodology on a representative case study, we extended an existing simulator with a poisoning attack injection module.