Last week, Substack CEO Chris Best was dressed down on the Verge’s Decoder podcast for refusing to tell host Nilay Patel whether users would be allowed to remain on Substack’s new Twitter-like Notes service for making racist statements such as (this was the host’s example) “all brown people are animals and they shouldn’t be allowed in America.” Best said he wasn’t going to speculate on hypothetical cases, which caused Patel to sanctimoniously lecture the CEO about how he’d flunked an important moral litmus test: “You know this is a very bad response to this question, right? You’re aware that you’ve blundered into this. You should just say no. And I’m wondering what’s keeping you from just saying no.”

No one likes racist trolls on social media. But Best was perfectly correct to resist Patel’s line of questioning. Once you set hard and fast rules for what you’re allowed to say or not allowed to say on a platform, the first thing people do is figure out how (a) to communicate the same idea in veiled form; and (b) to get their enemies suspended by reporting them, in bad faith, for communicating the forbidden words as a joke or thought experiment (e.g., reporting Patel for saying the words, “all brown people are animals”). I’m not a free-speech absolutist. Some safeguards do need to be in place. But setting down granular, content-specific rules is a formula for false positives and false negatives alike.

In any case, the best kind of content moderation is the kind you never have to do, since the users on your network don’t want to say awful things—either because they’re the sort of people who aren’t apt to say awful things to begin with, or because they recognize that they’re in a space where that kind of content is stigmatized.