[gruez]

That's where the "randomly choose 50" comes in. Even if your initial list of 100 topics was biased, randomly picking from that list allows you to do a comparison between topics that got randomly picked vs ones that didn't. In other words, a randomized controlled trial, the gold standard in experiments. If the group that got randomly picked got a 40% hit rate (ie. that corresponded to ads), but group that didn't randomly get picked got 41%, then you can probably conclude they're not listening to you, even if a 40% hit rate seems spooky. On the other hand if there were significant differences (eg. 40% vs 20%), then there might be other stuff going on worth investigating.