The Content Optimizer selects hypotheses using a two-stage acquisition policy that first weights Rank Factor Classes by inverse testing frequency and then evaluates individual claims via Thompson Sampling or Upper Confidence Bound.