One day you publish a careful, balanced explanation of a complicated issue. It receives twenty thousand views.

The next day you publish something sharper. Someone is wrong. Someone is dangerous. Someone is destroying something. There is a clear villain, a clear victim and very little room between them.

Two million views.

Thousands of comments.

People attack you. Others defend you. Some send the video to friends because they agree with you. Others send it because they cannot believe how wrong you are.

Your follower count rises.

So does your visibility.

Perhaps your income does too.

What did you just learn?

Maybe nothing at first. One successful post proves very little.

But what if it happens again?

And again?

At what point does an emotional reaction become a lesson?

At what point does a lesson become a strategy?

And at what point does a strategy become a business model?

The problem is not anger.

Anger can be justified. Anger has confronted corruption, abuse, discrimination, exploitation and violence. A society incapable of anger would not necessarily be a peaceful society. It might simply be a society incapable of reacting when something is wrong.

The problem begins somewhere else.

It begins when anger stops being a signal that something should be resolved and becomes a resource that must continue to be produced.

Because if anger creates attention, attention creates engagement, engagement creates visibility, and visibility can create money, then anger has entered an economic system.

And once a human emotion becomes economically useful, we should ask who has an incentive to keep producing it.

The modern social-media creator does not need everyone to love them.

This is one of the strangest features of the system.

A person who loves you may watch your video, comment on it and share it.

A person who hates you may do exactly the same thing.

They may watch every second in disbelief. They may return to see what you said next. They may write an angry response. They may send your video to five friends with the message, "Look at this idiot."

From the perspective of human emotion, love and hatred are opposites.

From the perspective of an engagement counter, they can sometimes look remarkably similar.

Both can produce attention.

Both can produce activity.

Both can keep the screen alive.

This is part of the attention economy (an economic environment in which human attention is a scarce resource that platforms, advertisers and creators compete to capture).

And it creates a disturbing possibility.

If someone who hates you is still economically useful to you, do you really need their hatred to disappear?

Research gives us reasons to take this mechanism seriously.

A 2023 study in Nature Human Behaviour examined tens of thousands of randomized headline experiments involving hundreds of millions of impressions. For an average-length headline, each additional negative word increased click-through rates by about 2.3 percent.

That does not mean negativity always wins. It does not mean every angry video will outperform every calm one. Human behaviour is more complicated than that.

But it shows something important.

Negative information can have measurable attention value.

Psychologists often discuss negativity bias (the tendency for negative or threatening information to attract disproportionate attention or psychological weight).

Social media did not invent that tendency.

It industrialized the environment around it.

Now imagine what happens to a creator inside that environment.

A 2021 study examining 7,331 Twitter users and 12.7 million tweets found that positive social feedback for expressions of moral outrage predicted more outrage expression in the future, and experiments supported a causal role for social feedback. The researchers also found that users adapted their expressions to the norms of the networks around them.

This resembles reinforcement learning (a process in which behaviours followed by rewards become more likely to be repeated).

You become angry.

You receive likes.

You repeat the behaviour.

You receive more attention.

Soon the platform may not merely contain your anger.

It may be teaching you which version of yourself performs best.

But the process works in the other direction too.

Creators train audiences.

Audiences train creators.

Platforms observe both.

Algorithms learn what keeps people engaged.

Creators learn what algorithms reward.

Users learn what kinds of expression seem normal.

Then everyone responds to the behaviour everyone else has helped produce.

Who is manipulating whom?

Perhaps that question is already too simple.

The system may not require a single manipulator.

A feedback loop can become powerful even when nobody controls the whole loop.

This is where individual psychology becomes sociology.

Suppose the creator stops saying, "This person is wrong."

Instead, the language slowly changes.

"These people are always like this."

"They hate people like us."

"They are destroying our society."

"They want to silence us."

Now disagreement is no longer between arguments.

It is between identities.

Social identity theory (the idea that part of our sense of self comes from the groups to which we believe we belong) helps explain why this change matters.

If you attack my argument, I may defend an idea.

If I believe you are attacking my group, I may feel that I am defending myself.

Now there is an in-group.

And an out-group.

Us.

Them.

A 2021 PNAS study examined more than 2.7 million posts from US news organizations and members of Congress on Facebook and Twitter. In that specific political dataset, posts about the political out-group were shared or retweeted about twice as often as posts about the in-group. Each additional reference to the political out-group was associated with substantially greater odds of sharing.

That does not prove that every social-media culture behaves this way.

It does reveal an incentive worth examining.

The enemy can be engaging.

And once the enemy becomes engaging, the enemy can become useful.

A creator who builds a community around shared curiosity can survive without an enemy.

A creator who builds a community around shared hatred may face a different problem.

What happens if the enemy disappears?

What happens to the content calendar?

What happens to the audience's shared identity?

What happens to the creator who has become the person who tells the group whom to fear, mock or despise?

This is where a political communication pattern enters the picture even when the creator is not formally political.

Populism has many definitions and cannot be reduced to one communication trick. But research on populist communication repeatedly finds versions of a familiar structure: "the people" are presented as a morally meaningful in-group, while elites, outsiders or another blamed group are positioned as responsible for decline, threat or betrayal.

A 2026 systematic review of 51 peer-reviewed studies on emotions in populist communication found anger to be especially prominent, alongside fear, resentment, nostalgia and other emotions. The review also identified threat-and-blame framing, identity appeals and emotionally charged narratives as recurring communication strategies.

This does not mean every angry influencer is a populist.

It means the grammar can travel.

You do not need to run for office to say:

We are the real people.

They are the problem.

You are being cheated.

They are laughing at you.

Nobody else will tell you the truth.

I will.

Once that grammar works, complexity becomes inconvenient.

A complicated housing problem may involve zoning, interest rates, construction costs, demographics, taxation, wages, financing and decades of policy.

That is difficult content.

"This group caused it" is easier.

This is scapegoating (the reduction of a complex problem to the blame assigned to a convenient person or group).

It offers something psychologically powerful.

A face.

Complex systems are frustrating because they often contain no single person to hate.

A villain solves that problem.

Suddenly uncertainty becomes certainty.

Structural complexity becomes moral simplicity.

And anger gains a direction.

Does that make the explanation true?

Not necessarily.

Does it make the explanation emotionally efficient?

Very possibly.

The more uncertain the world becomes, the more attractive a person can become who offers absolute certainty, a named enemy and a simple cause.

This is one reason the relationship between social-media outrage and historical propaganda deserves examination.

But the comparison must be made carefully.

A modern influencer is not Joseph Goebbels.

A creator chasing engagement is not equivalent to the propaganda ministry of Nazi Germany.

The Nazi propaganda apparatus operated with state power, censorship, centralized media control, systematic persecution and a political project that contributed to mass murder.

Flattening that history into an internet analogy would itself be irresponsible.

But history can still teach us about mechanisms.

The United States Holocaust Memorial Museum describes propaganda techniques that include selective omission, simplifying complex issues, playing on emotions and attacking opponents. Its historical material on Nazi propaganda documents the construction of a national in-group, the definition of enemies and outsiders, simple emotionally resonant messages, and the use of the most powerful media technologies available at the time.

Goebbels understood radio.

He understood film.

He understood repetition.

He understood that mass communication does not merely transmit information.

It can organize perception.

Today the technological structure is radically different.

There may be no propaganda ministry deciding every message.

There may be no central command.

There may simply be thousands of independent actors discovering that the same psychological buttons work.

Fear works.

Humiliation works.

Enemy construction works.

Repetition works.

Certainty works.

Simple explanations travel.

The mechanism can become decentralized.

That possibility should make us uncomfortable.

Does propaganda disappear when there is no propaganda minister?

Or can some propagandistic patterns emerge from millions of people independently discovering that the same emotional techniques receive attention?

Repetition deserves special attention.

Psychologists call one relevant phenomenon the illusory truth effect (the tendency for repeated statements to feel more truthful than new statements simply because they have become familiar).

A 2024 review of the research found that repetition can increase belief even in misinformation, including false headlines and implausible claims.

This does not mean repetition can force anyone to believe anything.

It means familiarity itself can quietly enter our judgment.

Now add algorithmic distribution.

You hear one creator say that a group is dangerous.

Then another.

Then another.

Then a clip.

Then a meme.

Then a reaction video discussing the original claim.

You may not experience this as repetition from a single source.

It feels like independent confirmation.

But what if all of those messages are circulating because the same engagement incentives selected them?

How many independent voices does it take before repetition begins to feel like evidence?

And what if the crowd you think is confirming reality is partly the product of the same machine that showed the crowd to you?

This leads to another distortion.

A 2023 Nature Human Behaviour study found that people systematically overestimated the moral outrage expressed by others on social media. That overperception increased beliefs that groups were more hostile toward one another and that hostile communication norms were stronger than they actually were.

Consider what that means.

Perhaps ninety-five people are relatively calm.

Five are furious.

But the five furious people post constantly.

Their messages create stronger engagement.

Their content travels farther.

Creators react to them.

Other users screenshot them.

Journalists discuss them.

Soon the five appear everywhere.

The ninety-five are mostly invisible.

What does the observer conclude?

"Everyone has gone insane."

This resembles pluralistic ignorance (a situation in which people misperceive what others actually believe or accept and then behave according to the false perception).

The quiet majority can begin adapting to the visible minority.

One side sees hostility and becomes more defensive.

The other side observes that defensiveness and interprets it as proof of hostility.

Now an inaccurate perception begins creating behaviour consistent with the inaccurate perception.

A self-fulfilling prophecy (a belief that helps produce conditions that make the belief appear increasingly true) may begin to form.

What if society is not initially as divided as the screen makes it appear?

And what if seeing a distorted image of division eventually helps make society more divided?

This is where the platform can no longer be treated as an invisible pipe.

Platforms do not merely decide whether speech may exist.

They also decide what is recommended, ranked, repeated and amplified.

Those are different powers.

Freedom of expression is not identical to a right to algorithmic amplification.

A person may have the right to stand in a public square and speak.

That does not automatically imply a right to have an invisible machine place a loudspeaker in front of them and carry their voice into ten million homes.

This distinction matters because platform design can alter exposure.

In 2026, a randomized Nature study assigned 2,000 people to different social-media feed-ranking systems for eight weeks around the 2024 US presidential election. Engagement-based feeds amplified intergroup, moralized, emotional and toxic content compared with reverse-chronological feeds, with some of the largest increases appearing in moral outrage and political content. They also increased perceived partisan animosity. Importantly, the study did not find that engagement-based feeds significantly changed participants' own engagement behaviour, so the result should not be exaggerated into "algorithms automatically turn people hateful."

But it demonstrated something narrower and important.

Ranking design changes what people see.

And what people repeatedly see can influence what they believe the social environment looks like.

Algorithmic amplification (the disproportionate increase in visibility produced by a recommendation or ranking system) is therefore not the same thing as neutral hosting.

The platforms themselves already recognize some responsibility.

YouTube's current harassment policy explicitly allows penalties when creators persistently incite hostility between creators for personal financial gain. Its advertiser rules can restrict monetization of hateful, inflammatory, humiliating and demeaning material.

That matters.

It means the question is no longer whether platforms acknowledge the problem at all.

They do.

The harder question is whether moderation at the edges can solve a problem that may also exist in the architecture at the center.

What happens if you ban explicit hate while still rewarding the emotional patterns that stop just short of it?

What happens if "I hate these people" violates the rules, but "Look what these people are doing to you" performs extremely well?

The European Union's Digital Services Act reflects this broader concern. Very large platforms are required to assess and mitigate systemic risks connected to recommender systems, including risks affecting civic discourse, elections, harmful content and well-being.

The idea is significant.

Responsibility may begin before a post becomes illegal.

It may begin with the design of the machine deciding which post deserves another million impressions.

But platform responsibility does not erase creator responsibility.

"The algorithm made me do it" is an explanation.

It is not automatically an excuse.

The creator still chooses the title.

The thumbnail.

The cut.

The target.

The missing context.

Whether the most extreme member of another group is presented as an individual or as representative of millions.

Whether a correction is posted.

Whether followers are encouraged to calm down or attack.

Whether a misunderstanding is resolved or converted into three more videos.

Whether the person on the other side remains a human being or becomes a recurring character in a profitable drama.

This is moral agency (the capacity to choose among meaningful alternatives and therefore carry responsibility for those choices).

If a creator knows that conflict performs well and deliberately increases the conflict because it performs well, the ethical situation has changed.

The algorithm may have presented the reward.

Who decided to reach for it?

This does not require us to read the creator's mind.

We do not need to declare someone evil.

We can examine behaviour.

Does the creator correct false claims even when the correction weakens the story?

Do they show strong examples from the opposing side, or only the most ridiculous ones?

Do they distinguish an individual from a group?

Do they ever de-escalate a conflict that could generate another million views?

What happens when reconciliation becomes possible?

Do they help it happen?

Or does a new enemy appear?

These are not accusations.

They are questions about incentives and choices.

And if creators are permitted to question society, society is permitted to question creators.

There is also a philosophical layer beneath all of this.

When another person's fear becomes useful because it keeps them watching, that fear has become an instrument.

When another person's hatred becomes useful because it generates comments, that hatred has become an instrument.

When an opposing group becomes useful because it supplies endless villains, human beings themselves can become instruments.

This is instrumentalization (treating people primarily as tools for another goal rather than as persons with value beyond their usefulness).

The opposing side becomes content.

Your own audience becomes engagement.

Their anxiety becomes retention.

Their anger becomes distribution.

Their loyalty becomes revenue.

And the creator may gradually discover that solving the conflict is economically less useful than maintaining it.

That produces a conflict of incentives even if the creator sincerely believes in the cause.

A person can genuinely want a better society while operating inside a business model that rewards a worse conversation.

Those two things can coexist.

That is precisely why incentives deserve examination.

Imagine a creator whose largest videos always depend on social conflict.

If tomorrow everyone became calmer, more nuanced and more willing to understand one another, what would happen to the channel?

Would it grow?

Would it shrink?

Would the creator be happy about the social improvement even if the analytics collapsed?

There may be no way to answer that from outside.

But why should we be afraid to ask?

This is also where economics offers a useful concept.

A negative externality is a cost created by an activity that is borne partly by people who did not receive the benefit.

A factory can keep the profit from a product while its smoke enters everybody's air.

The digital version may look different.

A creator receives the views.

The platform receives the engagement.

Advertisers receive attention.

But who receives the social distrust?

Who absorbs the suspicion between groups?

Who lives inside the environment where humiliation becomes normal conversation?

Who pays when increasingly extreme language becomes the price of being heard?

The profit can be private while the cost of the outrage becomes social.

Perhaps the digital factory has a chimney too.

It just does not emit smoke.

It emits atmosphere.

But there is still one participant missing.

Us.

It would be comfortable to finish the story with greedy platforms and irresponsible creators.

That would also be incomplete.

We click.

We watch.

We comment.

We share.

We send the video to someone because we agree with it.

We also send it because we hate it.

We reward the person we claim to want gone.

We teach recommendation systems what holds our attention.

We sometimes search for the most foolish person on the opposing side because defeating the strongest argument is harder.

We sometimes enjoy moral certainty more than understanding.

We sometimes want the villain.

The user has moral agency too.

But responsibility should not be distributed equally simply because everyone participates.

A useful principle may be this:

Responsibility should grow with knowledge, control, scale and benefit.

A platform has enormous behavioural data, massive distribution power and the ability to redesign recommendation systems.

A creator controls framing, language, editing and repeated content choices and may benefit directly from the resulting attention.

An ordinary user has far less power, but not zero power.

We cannot redesign the recommendation system.

But we can refuse to become free labour for an outrage campaign.

We can stop hate-watching.

We can avoid sharing something merely because it enraged us.

We can ask whether the most extreme example is representative.

We can deliberately search for the strongest version of the opposing argument.

We can notice when a creator needs us angry again tomorrow.

Because that may be the most revealing question of all.

If calming you down would cost someone money, how strong is their incentive to calm you down?

If helping you understand the other side reduces engagement, how strong is their incentive to help you understand it?

If a social wound produces content every day, what happens to the interests of the person whose career depends on documenting, enlarging or reopening that wound?

This does not mean every creator who discusses conflict is exploiting it.

It does not mean every political movement is propaganda.

It does not mean every emotional message is manipulation.

It does not mean every algorithm wants division.

Systems do not need intentions to produce incentives.

And people do not need to be villains to respond to incentives.

That may be the most uncomfortable part.

A creator can begin by sincerely defending a cause.

An audience can begin by sincerely opposing an injustice.

A platform can begin by sincerely trying to show people content they enjoy.

An advertiser can simply want attention.

No single participant needs to wake up and decide:

Today I will divide society.

And yet the combined machine can still reward division.

That is why searching for one guilty person may prevent us from seeing the phenomenon.

The deeper question is not:

Who is evil?

It is:

What behaviour does the system reward?

Who learns from those rewards?

What happens when everyone learns the lesson?

A creator discovers that anger performs.

The platform discovers that angry content retains attention.

The audience discovers that outrage receives social approval.

Political actors discover that enemies mobilize groups.

Advertisers discover where the attention is.

Then the outputs of one layer become the inputs of the next.

Anger becomes engagement.

Engagement becomes visibility.

Visibility becomes influence.

Influence becomes income.

Income rewards the production process.

And the process begins again.

Somewhere inside that loop, a real human grievance may still exist.

That matters.

We should not dismiss it.

But a grievance and an industry built around a grievance are not the same thing.

A wound can be real.

So can the business that grows around keeping everyone looking at it.

Perhaps that is the distinction we should learn to see.

When someone makes you angry online, ask whether the anger points toward understanding, action or resolution.

Or whether it simply points toward the next video.

Ask whether the enemy is necessary to explain reality.

Or necessary to maintain the audience.

Ask whether repetition is adding evidence.

Or merely familiarity.

Ask whether the platform is showing you society.

Or showing you the portion of society that best holds your attention.

Ask whether you are witnessing public opinion.

Or an engagement-selected sample of public emotion.

And before pressing share because something made you furious, ask one more question.

Who benefits from the next click?

The creator may have responsibility.

The platform may have responsibility.

The political communicator may have responsibility.

The advertiser may have responsibility.

The user may have responsibility.

Different power.

Different knowledge.

Different control.

Different responsibility.

But no participant becomes invisible simply because another participant has more power.

The outrage economy is not built by one hand.

That may be why it is so difficult to stop.

Its raw material is human emotion.

Its machinery is social.

Its distribution is algorithmic.

Its incentives are economic.

Its language can become political.

Its techniques can resemble propaganda.

And its costs do not remain inside the screen.

A society repeatedly shown an exaggerated image of its own hatred may eventually begin to resemble the image.

So perhaps the final question is not whether anger belongs on social media.

Of course it does.

The question is what happens when too many people acquire an interest in keeping it alive.

If your income rises when society becomes angrier, what happens when society begins to heal?

If your audience stays together because it shares an enemy, what happens when the enemy disappears?

If your platform earns more when people cannot look away, what responsibility does it have for what it teaches people to look at?

And if you keep returning to the person you claim to hate, are you their opponent?

Or have you quietly become their customer?

A society should be able to become angry when something is wrong.

The danger begins when someone needs it to remain angry after the reason for the anger should have been resolved.

Because at that point, anger is no longer only a reaction.

It has become inventory.