An Insider’s Reflection from Trust & Safety
“My post didn’t break any rules. Why was it removed?”
“My post didn’t break any rules. Why was it removed?”
If you’ve spent any time on social media, you’ve probably seen this question.
Sometimes it’s a friend venting on Facebook. Sometimes it’s a creator posting screenshots of a sudden account suspension. Sometimes it’s trending across X, Reddit, or YouTube with thousands of people sharing the same frustration.
And the conclusion almost always arrives fast:
“The internet is being over-moderated.”
After more than a decade working inside Trust & Safety operations — reviewing thousands of decisions, coaching reviewers, running quality audits, and investigating policy edge cases — I’ve heard this concern more times than I can count.
And here’s my honest answer after all of it:
It’s not a yes or no question. It never was.
🌐 The Internet Isn’t the Same Place It Was Ten Years Ago
When most people picture content moderation, they imagine teams removing the obvious stuff.
Spam. Graphic violence. Illegal content. Pornography.
Ten years ago, that covered the majority of the work.
Today the landscape looks completely different — and from inside the industry, the shift has been staggering.
Modern Trust & Safety teams now actively deal with:
- Coordinated harassment campaigns designed to target individuals
- Deepfakes and AI-generated synthetic media used for manipulation
- Financial scams engineered to steal from vulnerable users
- Child safety threats that evolve across platforms faster than policies can
- Terrorist propaganda and online radicalization pipelines
- Election manipulation through coordinated inauthentic behavior
- Medical misinformation with documented real-world harm
Every year, bad actors find smarter ways to abuse online platforms.
Policies expand because the threats expand.
From outside, it can feel like platforms are endlessly creating new rules.
From inside? It often feels like we’re barely keeping pace with threats that didn’t exist three years ago.
🔍 A Real Scenario That Permanently Changed How I See This Work
Early in my career, I was assigned a batch of reports that looked almost identical at first glance.
The posts themselves? Completely harmless on the surface. No offensive words. No explicit threats. Nothing a casual user scrolling past would even pause on.
Reviewed individually, every single post would have been allowed. I would have cleared the queue and moved on.
But something felt off about the pattern.
When our team investigated further, what emerged was impossible to ignore.
The same cluster of accounts was repeatedly targeting one individual — different wording, different images, different times of day — in a way that made each post look isolated and innocent.
Together, they formed a coordinated harassment campaign designed specifically to evade detection by looking harmless one post at a time.
One isolated post — not harmful.
One hundred coordinated posts targeting the same person — absolutely devastating.
If our moderation had only examined individual content pieces, the victim would have had zero protection and no recourse.
That experience taught me something I carry into every shift since:
Sometimes moderation isn’t responding to a single piece of content. It’s responding to a pattern — and patterns are invisible unless you’re trained to find them.
💔 Why It Always Feels Personal When It Happens to You
Here’s a perspective most users never consider.
Every single day, billions of posts go live across major platforms.
Only a tiny fraction ever receive any enforcement action.
But when your post gets removed, your personal experience becomes 100%.
It feels targeted. It feels personal. It feels political.
I’ve handled countless appeals from users who were genuinely convinced that a moderator had specifically disagreed with their opinion and acted on it.
In most of those cases, the actual reason was far more straightforward:
- A policy keyword triggered an automated flag
- A behavioral pattern matched a known risk signal
- A reviewer applied existing guidelines consistently
Does that mean every removal decision was correct?
Absolutely not.
Moderators make mistakes. Automation makes mistakes. Policies sometimes lag behind reality.
But here’s the part that rarely makes it into the public conversation:
The vast majority of these decisions aren’t ideological. They’re procedural.
The intent is rarely censorship. The execution is sometimes genuinely imperfect. Those are two very different problems.
🤖 Where Automation Helps — And Where It Hurts
Let me be direct about something: modern platforms cannot exist without AI-assisted moderation.
Millions of pieces of content are uploaded every single hour. No human workforce on earth could review everything in real time.
Automation genuinely helps — detecting known harmful content at scale, identifying coordinated spam networks, prioritizing the highest-risk reports, and significantly reducing direct human exposure to deeply disturbing material.
That’s the part that works.
The part that causes real damage to real users is that AI doesn’t understand context. It calculates probability.
From my own experience reviewing escalated cases, I’ve personally seen:
- Educational journalism about extremism flagged because it referenced extremist symbols
- Legitimate historical discussions removed because keyword patterns matched hate speech policies
- Obvious satire actioned as genuine harmful content because the system couldn’t read tone
These are false positives — and they’re one of the most significant unsolved problems in Trust & Safety.
For the user who experiences one, it feels like censorship.
For the teams inside the industry, it’s a constant signal that the systems require better training, better context awareness, and better human oversight.
🔄 The Side of This Conversation Nobody Wants to Have
Every time I hear “platforms remove too much,” I find myself asking a quieter question.
Too much — compared to what?
Because from the same desk where I’ve reviewed wrongful removals, I’ve also reviewed:
- Children being actively groomed and targeted across platforms
- Individuals receiving thousands of abusive coordinated messages every day
- Scam networks systematically draining life savings from elderly victims
- Non-consensual intimate images being spread across multiple platforms simultaneously
- Credible violent threats that required immediate law enforcement coordination
The people on the receiving end of those situations never ask whether platforms moderate too much.
They ask: “Why didn’t this get removed sooner?”
This is the reality that makes Trust & Safety genuinely difficult:
One group experiences moderation as too aggressive. Another experiences it as dangerously insufficient. Both are telling the truth from where they’re standing.
⚖️ The Impossible Balance I Watch Play Out Every Single Day
Picture this scenario.
You’re responsible for moderating one billion posts daily.
Make your AI stricter:
✅ More harmful content gets caught
❌ More innocent content gets wrongly removed
Make your AI more lenient:
✅ Fewer innocent users are affected
❌ More harmful content reaches more people for longer
There is no perfect configuration. Every platform is permanently adjusting this dial. Every adjustment creates new trade-offs. And every trade-off has a human being on the other end of it.
What operations work taught me — slowly and sometimes painfully — is this:
Moderation isn’t about achieving perfection. It’s about managing an imperfect system as responsibly as possible.
✅ What Good Moderation Actually Looks Like From the Inside
A lot of people assume good moderation simply means removing more content.
After 11 years, I believe the opposite.
Good moderation means making better decisions — and building systems that support that quality consistently.
That means:
Clear, accessible policies — Rules users can understand before they ever post, not after enforcement.
Consistent enforcement — The same standard applied regardless of who posted it or how large their following is.
Fast, fair appeals — Mistakes are inevitable. What matters is how quickly they get corrected.
Human oversight at scale — Automation that amplifies human judgment rather than replacing it entirely.
Transparent communication — Users who understand why action was taken are far less likely to feel targeted or silenced.
The goal was never maximum enforcement.
The goal is appropriate enforcement. The distance between those two things is where most of the public frustration lives.
💭 So — Are We Over-Moderating the Internet?
Sometimes. Genuinely, yes.
Automation overreacts. Policies sometimes expand too aggressively following high-profile events. Enforcement mistakes happen and real people pay the cost.
None of that should be dismissed or minimized.
But there are also thousands of situations every single day where moderation quietly prevents serious harm before most users ever know a threat existed.
The biggest misconception I encounter — from users, from journalists, from people outside the industry — is the belief that content moderation exists to control conversations or suppress particular viewpoints.
From everything I’ve experienced across more than a decade inside this field, the actual mission is far simpler and far less dramatic:
Protect users. Reduce harm. Keep healthy conversation possible.
That balance is harder than it looks from the outside.
And it has to be rebuilt, recalibrated, and re-earned every single day.
🏁 Final Thoughts
The internet is now the world’s largest public conversation — billions of participants, thousands of cultures, and threats that evolve faster than any policy document can fully capture.
Moderation will never satisfy everyone. Some people will always experience it as too strict. Others will always see it as dangerously weak. Both will find genuine evidence to support their position.
After years inside this work, I’ve stopped debating whether moderation should exist.
Instead, I ask better questions:
- Are the rules clear and genuinely accessible to users?
- Are they applied consistently regardless of who you are?
- Can people meaningfully appeal decisions that affect them?
- Are platforms actually learning and improving from their mistakes?
Because this work was never about removing as much content as possible.
It’s about protecting the space where billions of people connect, communicate, and build communities every day.
And that remains one of the hardest — and most important — balancing acts in the modern digital world.
The future of Trust & Safety won’t be defined by how much content we remove. It will be defined by how fairly, transparently, and responsibly we make those decisions.
💬 Over to You
Have you ever had content removed that felt completely unfair? Or wished a platform had acted faster on something harmful you reported? Drop your experience in the comments — I read every one.
📚 You Might Also Like
- SLA Pressure: The Invisible Force Behind Every Trust & Safety Decision
- Feedback Delays Are More Dangerous Than Negative Feedback
- What Nobody Tells You About Leading a Content Moderation Team
Categories: Trust & Safety | Content Moderation | Digital Policy | Operations Leadership
Tags: Trust and Safety Content Moderation Over-Moderation Digital Safety AI Moderation Platform Policy False Positives T&S Leadership Internet Safety Social Media Moderation