Just train a model to break the returned text into a set of claims, evaluate each claim for likelihood of being an ad, and then excise those claims that are lik

Just train a model to break the returned text into a set of claims, evaluate each claim for likelihood of being an ad, and then excise those claims that are likely ads.