In Bayes' theorem, the prior is multiplied by the likelihood and renormalised to give the posterior.
Why it matters in an interview: a p-value is P(data | null), and converting it into P(null | data) - which is what you actually want - requires a prior. Testing an a priori implausible hypothesis and getting p = 0.05 leaves it still probably false.
Informative versus uninformative. A flat prior seems neutral but is not invariant to reparameterisation - flat in one scale is informative in another. Worth knowing if a discussion turns Bayesian.