I’m not sure a 5 year old could understand it, but this document by the Stan team is really great for how to choose priors.
Likelihoods are more going to be driven by the nature of your data. Is it count data → Poisson. Is it binary → Bernoulli. Do you have a best guess that you know isn’t perfect → Normal, etc. It’s harder to give general advice on this front, since it’s really deeply tied to your specific model/data/research question.