⁂

That's where the evil part came in. We were pretty sure we could extract a list of the 100,000 most commonly searched terms from Google, and then we could use our web-crawler to capture the top 100 results for each. We could feed these to our Bayesian machine-learning tool to create statistical models of the semantic structure of these results, and then we could generate thousands of pages of word-salad for each of those keywords that matched those statistical models.

10/