What is prosocial media?

A proposal for building social media that does the opposite of platforms that profit by spreading posts designed to provoke anger or shock. It shows which groups accept a post and where opinion splits, gives priority to posts that can be shared across divides, and has communities pay for the value of those connections.

Open a social media app and you may see nothing but posts that make you angry or startle you. The longer people stay glued to the screen, the more the platform earns from advertising, so the system is built to spread such posts. Prosocial media is a way of building social media that aims for the opposite. "Prosocial" means tending to improve relationships between people and society.

In short

Many of today's social networks spread posts that provoke anger or shock to keep people on the screen, and make money from advertising. Prosocial media is a proposal for the opposite [1][2]:

It shows which groups broadly accept a post, and where opinion is split.
It gives priority to posts that can be shared across divides.
Communities (local governments, companies, religious groups and others) pay for this value of strengthening connections.

Glen Weyl, Audrey Tang and others sketched its prototype in chapter 6-3 of the book ⿻ Plurality, and turned it into a paper in 2025 [2][3].

For example

Taiwan's Sunflower Movement (2014). In an essay in the magazine Noema in May 2025, Weyl, Tang and Jacob Mchangama cite this movement as an example of prosocial design that worked. When students occupied Taiwan's Legislative Yuan (its parliament), Tang, then part of the civic technologists' community g0v, used digital tools to connect the opposing sides and build a coherent set of demands [1]. Tang later described this as follows: "We built prosocial media to replace antisocial media" [4]. → Sunflower Movement

Community Notes on X. The same essay treats Community Notes as a model for citizens raising up content that brings people together. Volunteers attach background explanations (notes) to misleading posts, and only notes rated "helpful" by people with different political positions alike are made prominent. The authors also write, however, that Community Notes only adds context after outrage-provoking posts have already spread, and does not touch the root problem [1].

We the People 250 (United States). In a December 2025 interview with Nikkei, Tang described a platform with about 2,000 participants, chosen as a representative miniature of the whole population, five from each US congressional district. She explained: "It replaced the algorithm that optimized for outrage with one that optimized for overlap" [5]. An "algorithm" here means the computational procedure that decides what is shown and in what order.

How it works

The problem is less bad posts than missing context

According to the Noema essay, the core of the problem is not harmful posts themselves. It is that social media hides the social context: who has seen a post, and who agrees with it [1].

This causes two kinds of misperception [1]:

False consensus. Taking a post that is only popular in a narrow group to be what everyone thinks.
Hidden consensus. Views that are in fact widely shared stay buried, out of sight.

So the answer, the essay says, is not removing posts from above, but more speech and more information [1].

The model: a 1947 commission

Chapter 6-3 of the book writes that the era of "press responsibility" in the United States dates only to the 1940s, and the "Hutchins Commission on Freedom of the Press". The commission held that a central role of a free press is to clarify to all citizens both the points of consensus and facts (news everyone sees) and the points on which opinion divides (balanced coverage) [3].

The book asks how social media, which has brought in a diversity of voices, can remain "pro-social media in the sense of the Hutchins report" [3]. The Noema essay proposed recreating this principle computationally, for many overlapping audiences [1].

Three measures

The essay names three concrete measures [1]:

Show the social context. Show in which communities a post is broadly accepted and where it divides people, and attach labels such as "common ground" or "diverse perspectives". The same post can differ by community: divisive in your church, say, but agreed on within the party you support. Recommendation systems already have this information, the essay says, so they need only show it to users.
Rank common ground higher. Bring out what the communities involved share, and rank posts so that attention is distributed fairly across every community. This is the same family of computation as Community Notes. → Bridging
Grow gatherings that cross communities, and make them pay. Create business models in which social media companies profit from gatherings that cross divides.

The paper's version

The paper "Community by Design" (the first version was titled "Prosocial Media", February 2025) treats the social fabric not only as "material" that social media uses, but also as an "outcome" that it deliberately produces [2].

In the paper's view, people belong to several communities at once, and every community contains differences within it. This follows the ideas of the sociologist Georg Simmel (→ intersectional identity). Each community values content that wins agreement across its internal differences, and content that presents those differences fairly. Posts carry labels showing for which communities they serve as a bridge [2].

The flow of money is also described. Citizens and communities can pay a fee so that more of the content they value is shown. Communities that cannot yet pay are subsidized, and people who create or select good content are rewarded [2].

Who pays?

Chapter 6-3 of the book criticizes both subscriptions and advertising for appealing "to consumers rather than citizens of diverse communities". It writes that social media should be funded by organizations with an interest in connecting communities: "churches, civic associations, governments at many levels, charities, universities, corporations etc." [3].

The book also says that makers of productivity software for businesses could be a place for new prosocial media to grow, since they do not want opinions among employees to split into conflict [3]. The paper, too, considers workplace social networks and collaborative software-development platforms a natural place to start [2].

What is debated

Not yet built or tested. The paper itself admits that much of its content is speculative and has not yet been tried on real social networks. It lists as open questions whether communities can be identified at large scale, whether the business model can work commercially, and whether it is realistic to ask users for more active participation than today [2].
Human connection is not always good. In the same paper, the authors cite research finding that in Germany during the rise of the Nazis, the Nazi party consolidated power faster in areas with stronger social ties. They acknowledge that mechanisms that strengthen communities could produce the opposite of what is intended. They also note a tension: the more attention goes to bridging content, the less is left for opposing views [2].
The line with advertising blurs. Because content that communities pay to promote is shown more, the paper cautions that payments need to be transparent, given laws requiring advertising to be distinguished from ordinary posts. It also says legal review is needed for handling communities defined by age, gender, ethnicity, religion and the like [2].
This is not about more censorship. The proposers reject restricting speech from above, as damaging to freedom and concentrating power in technical experts. Instead of removing or hiding content, they call for adding context [1].

Where it fits in Plurality

The starting point is chapter 6-3 of the book, "Media". Acknowledging the harms of today's "anti-social" media, the book applies the approach of chapter 5-4, "Augmented Deliberation". In this design, social media systems identify communities, and show both the common content that crosses the divides within each and the important points of division. This, the book writes, would avoid "false consensus" and "pluralistic ignorance", and redirect creators' incentives "away from divisive content" toward stories that bring people together [3][6].

Chapter 5-4 sketches a future in which the design of Community Notes is used not only to rank notes but to decide what attention goes to in the first place, making prominent in the feed the content that bridges the several communities a person belongs to, such as religious, local and political ones [7]. → ⿻ publics

Developments have continued since the book. In March 2025, in a roundtable held for the release of a documentary, Tang said that she calls a design that brings people together by adding bridging voices, instead of silencing voices, "prosocial media", as opposed to antisocial media [8]. The paper is referred to as the technical backbone of Tang and others' 6-Pack of Care [2].

Sources

Weyl, Tang and Mchangama, "Building A Prosocial Media Ecosystem" (Noema, 2025-05-01) — https://www.noemamag.com/building-a-prosocial-media-ecosystem/
Weyl, Thorburn, de Keulenaar, Mchangama, Siddarth and Tang, "Community by Design" (arXiv 2502.10834) — https://arxiv.org/abs/2502.10834
Plurality, 6-3 "Media" — https://www.plurality.net/v/chapters/6-3/eng/ ［CC0］
Audrey Tang, "Communal AI for a Plural Future" (2025-11-06), the paragraph looking back on Taiwan's civic movement ten years earlier — https://archive.tw/2025-11-06-communal-ai-for-a-plural-world ［CC0］
Audrey Tang, "Interview with Nikkei" (2025-12-08) — https://archive.tw/2025-12-08-interview-with-nikkei ［CC0］
Plurality, Japanese edition, 6-3 (in Japanese) — https://www.plurality.net/v/chapters/6-3/jpn/ ［CC0］
Plurality, 5-4 "Augmented Deliberation" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
Combinations, roundtable on the release of "Good Enough Ancestor" (2025-03-09) — https://www.combinationsmag.com/good-enough-ancestor/

We welcome error reports and questions ("What is…?"). Submissions become public issues.

Report an error in this article ・ Ask a question