Generalized linear models (GLMs) are commonly used to investigate relationships between a response variable and covariates, offering straightforward interpretability.However, concerns about model misspecification impacting inferential outcomes arise.An established frequentist technique, the quasi-likelihood, enhances robustness by necessitating specification of only the first two moments.We leverage on quasi-likelihoods in developing a robust approach for Bayesian inference in GLMs.Quasi-posteriors follow a coherent generalized Bayes approach and have desiderable large sample properties.In this paper we use the quasi-posterior for modeling the abundance of Eurasian chaffinch (Fringilla coelebs) in Finland in year 2014.