How Social Media Could Be Redesigned To Reduce Shame And Increase Grace

To understand why social media produces what it produces, you need to understand the optimization target. Every major platform's algorithmic feed is optimized for a single primary metric: engagement. Time on platform. Clicks, shares, comments, reactions. The algorithm learns, through continuous feedback, what produces those behaviors — and then shows you more of it.

The problem is not that the algorithm is evil. The problem is that the most reliable producers of engagement are the most emotionally activating content types. And the most activating emotions — not the most pleasant, but the most activating — are fear, anger, and disgust.

This was demonstrated empirically by Jonah Berger's research on viral content: content that produces high arousal negative emotions (anger, anxiety) spreads faster than content that produces low arousal positive emotions (contentment, gratitude). The algorithm, optimized for spread, learns to prefer content that produces high-arousal negative states. Not because any engineer programmed it to prefer that. Because the data teaches it that.

What does this look like at scale?

It looks like outrage being the primary currency of political communication. It looks like nuanced positions being algorithmically suppressed because they don't generate the emotional activation that simple, infuriating takes do. It looks like public shaming being one of the most reliably engaging social activities available — more engaging than celebrating, more engaging than discussing, more engaging than learning.

And it looks like shame becoming a tool of social regulation so effective and so low-cost that it's deployed constantly — not for genuine accountability, but because it works, and it's entertaining, and the platform rewards it.

Social media shame functions differently from interpersonal shame in ways that amplify its damage:

Scale without relationship: In a community where people know each other, social disapproval is calibrated by relationship. The person who shames you knows your history, cares (however imperfectly) about your wellbeing, and bears social costs for going too far. On social media, the shaming crowd has no relationship with the target and bears no social cost for excess. This removes every natural brake on escalation.

Permanence: A public shaming in a physical community fades. Memory is imperfect. People move on. Online, the pile-on is documented, searchable, and permanent. The person who said something offensive at 22 can have it follow them at 42. The lack of temporal decay removes the possibility of genuine recovery.

Decontextualization: Social media's format — the screenshot, the clip, the quote without context — is perfectly designed for misrepresentation. A nuanced, three-minute speech becomes a five-second clip. A joke becomes evidence of malice. An error becomes proof of character. The platform mechanics reward sharing the decontextualized version because it's more activating.

Contagion dynamics: Pile-ons spread not because thousands of people independently decided that the target deserves condemnation, but because social media's mechanics make joining a pile-on feel like participating in community. The like, the share, the piling-on comment are social bonding behaviors in a context that provides the social reward of belonging without the relational cost of consideration. You can bond with your tribe by destroying someone you've never met.

Monica Lewinsky has spoken and written extensively about being, before the term existed, the first major victim of internet pile-on culture. What she describes — the experience of thousands of strangers coordinating contempt at a global scale — was not possible before social media. It's now routine. It happens hundreds of times a day across platforms.

Jon Ronson's So You've Been Publicly Shamed documented the early years of this phenomenon and the lasting psychological damage it produces — not just to high-profile targets but to ordinary people who had a bad day, said something stupid, and found themselves the target of a coordinated international humiliation campaign. Most never recovered fully. Some never recovered at all.

The good news is that design matters. The platform is not neutral — it shapes behavior through its architecture — and different architectural choices produce different behaviors. We know this not just theoretically but empirically.

Friction before sharing

Twitter's 2020 experiment: when users tried to share a news article without clicking through to read it, the platform prompted: "Want to read this before sharing?" This simple friction reduced the sharing of external links without reading by 40%. Information that people hadn't read — and therefore couldn't have evaluated — spread less.

If a similar friction were applied to content generating outrage signals (high engagement, negative sentiment, rapid spread), the viral cascade would slow. The cascade is the harm mechanism. Slowing it is the intervention.

Cooling periods for high-engagement targets

When a piece of content is receiving an unusual volume of responses (a signal of pile-on dynamics), the platform could introduce friction: slow down notification delivery to the target (reducing the experience of being buried alive), prompt responders to consider whether they're adding something new or simply joining a mob, slow the amplification of the content generating the pile-on.

None of this removes speech. It adds temporal friction to the mechanics that currently turn normal social conflict into catastrophe.

Removing public metrics

The psychology of social comparison is well-documented: visible public metrics (likes, followers, view counts) drive comparison, status anxiety, and performative behavior. People optimize for the metric rather than for genuine expression.

Instagram's 2019-2020 experiments with hiding like counts showed measurable reductions in reported anxiety among users. The company rolled back the experiment, then made it optional. Making it default-off would produce the benefit at scale. The reason it hasn't happened: likes drive behavior that drives engagement that drives ad revenue. The anxiety is a feature of the business model, not an unfortunate byproduct.

Algorithmic opt-out

Currently, users of major platforms have no meaningful control over what their algorithm optimizes for. You get what the algorithm thinks will keep you engaged — which, as described, tends to mean high-arousal negative content.

A grace-based alternative: let users select their optimization target. A "nuance feed" that deprioritizes content generating outrage and prioritizes content generating thoughtful engagement. A "slow feed" that shows content from people you've directly connected with rather than viral content. A "chronological feed" that removes algorithmic curation entirely.

Some platforms have added chronological options under regulatory pressure (Instagram, after EU pressure). None have made grace-oriented optimization a first-class option. The regulatory mechanism exists to require this; the political will to use it is the variable.

Restorative features

What if platforms built in mechanics for repair, not just for attack?

Some smaller platforms have experimented with "context addition" — mechanisms for people who were misrepresented to add context to shared content about them. Community notes on Twitter/X is a partial, imperfect version of this. More robust versions could allow targets of pile-ons to add narrative, require that context appear alongside shared content, and algorithmically reduce the spread of content where significant context exists.

What if there were a "resolution" mechanism — a way for two parties who have had a public conflict to indicate that it's resolved, and for that resolution to be amplified as prominently as the original conflict? Currently, the conflict spreads; the repair is private and invisible. This asymmetry is architecturally chosen, not inevitable.

Several companies have built, at least partially, with different values:

Mastodon and the Fediverse: Decentralized architecture removes the profit incentive that drives shame-maximization. No venture capital, no ad revenue, no engagement-optimization. The result is slower, less viral, and — by most user reports — less anxiety-producing. The tradeoff is smaller scale and less content discovery. The model demonstrates that different architectures produce different cultures.

Substack's community features: Built primarily for long-form, subscription-based writing, with comments systems designed for slower, considered engagement. The economics (subscription rather than advertising) remove the engagement-maximization incentive.

Nextdoor's redesign: After research showing that the platform was amplifying neighborhood conflict and racial profiling (neighbors reporting "suspicious" people who were just black people existing in their neighborhood), Nextdoor redesigned several features. They added friction to crime reports, requiring more specific information before posting. They removed the ability to tag specific people in crime reports. The incidence of racial profiling posts dropped significantly. A design change produced a behavioral change.

BeReal's premise: The app's design — a notification at a random time each day, requiring a photo of what you're doing right now, front and back camera simultaneously — was explicitly an anti-performance architecture. No editing, no choosing the right moment, no optimization. Its limitations are obvious, but its design principle — build friction into the performance layer — is transferable.

The pattern is consistent: platforms will make design changes that improve user wellbeing if and only if those changes don't reduce engagement or revenue, or if they face sufficient regulatory or reputational pressure to make the calculation change.

Facebook's internal research, leaked by Frances Haugen in 2021, documented that the company knew its platform was damaging to teenage girls' mental health and chose not to act, because the engagement produced by the harmful features was too valuable. This isn't corporate villainy in the movie-villain sense. It's the predictable behavior of a company optimizing for shareholder returns in the absence of external constraints.

External constraints — regulation — are the mechanism.

The EU's Digital Services Act (DSA), which took effect in 2024, requires large platforms to assess and mitigate "systemic risks" including effects on mental health, particularly of minors. Platforms must now provide non-personalized options (no algorithmic curation) on request. Algorithmic audits are required. This is a beginning.

The US has no equivalent legislation, though the Kids Online Safety Act and similar proposals have moved through Congress. The lobbying opposition from platforms is substantial. The political will to override it has, so far, been insufficient.

This matters beyond individual mental health outcomes.

Democratic deliberation requires that citizens be capable of engaging with complexity, holding nuance, updating their views based on evidence, and tolerating disagreement without it feeling existential. The current social media architecture systematically degrades all of these capacities.

When the dominant communication infrastructure of a civilization optimizes for outrage, it shapes the political imagination of that civilization. People who spend significant time in outrage-optimized environments become more extreme, more certain, less capable of the kind of complexity-holding that democratic governance requires. This is not a moral judgment — it's what the research shows.

Public health — particularly the mental health of young people — is being measurably damaged by the current architecture. The surgeon general's reports, the academic research, the clinical data all point in the same direction: social media as currently designed is producing epidemics of anxiety, depression, and social isolation, particularly in adolescents.

The question is whether a civilization can build communication infrastructure that is optimized for something other than the most profitable form of human activation. That requires being willing to name what the current architecture is doing — shame amplification, outrage maximization, public humiliation as entertainment — and to decide that those are not acceptable features of how we talk to each other.

It's a design choice. Made by people. It can be remade.