The Content Optimizer selects a Rank Factor class using a two-stage process that first exponentially deprioritises classes that have already been heavily explored and then draws random samples from their Beta posteriors to pick the leading hypothesis.