Moderation Removes Bad Comments. It Doesn't Create Good Ones.

TL;DR: Content moderation tools only subtract — they detect bad comments and delete them. But deletion doesn’t produce good comments, doesn’t teach commenters anything, and selects for the trolls most motivated to evade filters. Bad comments are a skill problem, not a character problem, and the fix is real-time feedback before posting, not removal after. That’s the approach Respectify takes.

There’s an old joke about the sculptor who was asked how he carved such a beautiful elephant. “Easy,” he said. “I just chip away everything that doesn’t look like an elephant.”

That joke is, more or less, the business plan of the entire content moderation industry. Take your comment section, chip away everything that doesn’t look like civil discourse, and what remains will — the theory goes — be civil discourse.

Except it doesn’t work that way. I’ve spent a lot of time looking at comment sections (an occupational hazard, these days), and I can tell you what’s left after aggressive moderation chips away all the bad stuff. It isn’t an elephant. Most of the time, it’s nothing at all.

Subtraction is not a strategy

Every moderation tool on the market today does fundamentally the same thing. It detects toxicity, spam, harassment, or profanity, and then it removes it — or removes the person who posted it. The tools differ in speed and accuracy, and the newer ones use very impressive AI to do the detecting. But strip away the marketing and they’re all subtraction machines. Find bad thing. Delete bad thing.

And look, subtraction has its place. Spam should be deleted. Genuine harassment should be deleted, and the harasser shown the door. I’m not arguing for a moderation-free internet. (I’ve seen the moderation-free internet. No thank you.)

But here’s the question the whole industry has quietly skipped over: after you’ve deleted everything bad, what makes anything good appear?

Nothing does. That’s the problem. Deletion doesn’t produce a single thoughtful comment. It doesn’t teach a single commenter to make their point without the insult attached. It just clears the floor and hopes something better shows up. Twenty-five years of evidence suggests it won’t.

What deletion actually teaches people

Worse than teaching nothing, deletion teaches the wrong things.

Think about what a commenter actually learns when their post silently disappears. They don’t learn what was wrong with it — nobody told them. They learn that the site is against them. Maybe they learn to be sneakier: swap a letter in the offending word, split the insult across two sentences, find the phrasing that slips past the filter. Every veteran moderator knows this arms race by heart. The filters get smarter, the evasions get smarter, and around and around we go.

Meanwhile, the reasonable people — the ones you actually want in your comment section — watch this circus and quietly conclude that commenting isn’t worth it. So the comment section slides toward the very people most motivated to fight the filter, which is precisely the population you were trying to remove. Congratulations: your subtraction machine is now selecting for trolls.

I spent years managing software developers, and we had a saying about testing: you can’t test quality into a product. Testing finds defects, but the quality has to be built in from the start, by people who know how to build it. Comment sections have the exact same problem. You cannot delete quality into a conversation. The quality has to come from the commenters themselves — and most of them were never shown how.

Bad comments are a skill problem, not a character problem

Here’s the assumption buried inside every moderation tool: bad comments come from bad people, so remove the people and you remove the problem.

I don’t buy it. Read a hundred hostile comments — really read them — and you’ll find that most aren’t written by monsters. They’re written by ordinary people who have a genuine opinion, feel strongly about it, and have never in their lives been taught how to disagree with a stranger in writing. That’s not a character flaw. That’s a skill gap. Shoot, professional writers spend years learning this skill, with editors correcting them the whole way. We hand random people a text box and a “post” button and act surprised when it goes badly.

And skill gaps have a known solution, and it isn’t punishment. It’s feedback — the kind an editor gives, not a bouncer. Specific, immediate, actionable feedback: here’s what your comment is doing, here’s why readers will tune it out, here’s how to make your actual point land harder. Give a person that feedback at the moment they’re about to hit “post” — before the comment goes live, before there’s anything to delete — and something interesting happens. Most of them revise. Not because they were forced to, but because nobody actually wants to be ignored, and a comment that reads as a tantrum gets ignored.

Do that a hundred times across a community and you’re no longer moderating a comment section. You’re training one.

Addition, not subtraction

This is the idea Respectify is built on, and it’s the dividing line between us and everyone else in this space. The industry optimizes for removing bad comments. We optimize for producing good ones — coaching commenters to sharpen their point before it posts, the way an editor would. Deletion still exists for the true garbage, but it’s the backstop, not the strategy.

Sure, this is a harder problem than detection. Detecting a slur is easy; understanding what someone is trying to say and helping them say it better is genuinely difficult. But it’s the right problem, because it’s the only approach that leaves your community better than it found it. Subtraction, at its theoretical best, gets you to zero. Only addition gets you above it.

The sculptor’s joke is funny because it’s absurd — chipping away can’t be the whole method, or every block of marble would contain an elephant. Every comment section owner who’s bought another, better chisel has learned the same lesson the hard way.

At some point you have to stop chipping and start teaching people to carve.

Common questions

Why doesn’t content moderation improve comment quality? Because moderation is purely subtractive. Deleting a bad comment removes it, but produces nothing in its place and teaches the commenter nothing about what went wrong. Comment quality only improves when commenters get feedback that helps them express their point better.

What’s the alternative to deleting comments? Pre-post coaching: analyzing a comment before it’s published and, when it has problems — hostility, logical fallacies, off-topic drift — showing the commenter specific feedback and letting them revise. Deletion remains the backstop for spam and genuine harassment.

Does coaching commenters actually work? Most hostile comments come from ordinary people with genuine opinions and no training in written disagreement — a skill gap, not a character flaw. Skill gaps respond to specific, immediate feedback, and most people revise when shown why their comment will be dismissed by readers.

How is Respectify different from Perspective API, Hive, or CleanSpeak? Those tools are detection systems: they score or flag toxic content so it can be removed. Respectify analyzes comments before posting and coaches the commenter to revise, treating moderation as a communication-quality problem rather than a removal problem. It also offers a near drop-in replacement for the retiring Perspective API.