Social Media

Social media are internet-based services that let users create, share and react to content, and maintain networks of social ties with other users. Unlike broadcast media, which send a few producers' messages to many passive audiences, social media are organized around user-generated content, profiles and the relationships between accounts. They build on the internet and the World Wide Web, and their development since the mid-2000s has changed how people communicate, how news circulates, how politics is organized, and how businesses reach customers. Because the platforms are run by private companies whose choices about ranking, data and moderation shape what users see, the subject also sits at the junction of computer science, economics, sociology and law.

This entry first sets out what counts as social media and how the field developed, from early online communities to the mobile, short-video and AI-driven platforms of the present. It then explains the technical and economic structure of the platforms, including the algorithms that rank content, the advertising and data systems that fund them, and the network effects that concentrate users on a few services. The central part of the entry covers social consequences: civic participation and politics, news and misinformation, polarization, mental health and well-being, culture and language, privacy and data protection, and content governance. For each, it distinguishes well-established findings from claims that remain contested.

The final sections address regulation, the methodological difficulties of studying platforms that control their own data, and open questions about decentralization, synthetic content and long-term effects. Throughout, the entry aims to separate what the platforms are, what research has measured, and where the evidence remains divided.

Defining Features and Scope

Social media refers to a family of online services in which much of the content comes from users rather than from professional publishers, and in which each participant has a persistent identity and visible links to others. The content side of this is usually described as user-generated content, meaning media published by individual online contributors rather than traditional publishers. The social side is the set of connections between accounts. Most platforms combine user-generated content, a social graph of friends, followers or connections, and interactive tools such as comments, likes, shares and direct messages. Each of these three elements can be found in other kinds of online service, so the category is defined by how they are combined, not by any one of them.

Boyd and Ellison's Definition

The most-cited academic definition comes from danah boyd and Nicole Ellison's 2007 article. They described social network sites as services that let people build a public or semi-public profile inside a bounded system, list the other users with whom they share a connection, and browse their own list and the lists of others. The definition is structural. It says nothing about the quality of the content or the size of the audience, and it puts relationships and visibility at the centre. The authors also deliberately preferred the term "social network sites" to "social networking sites". Most sites support the maintenance of pre-existing social networks, but others help strangers connect based on shared interests, political views, or activities. The narrower term would have implied that the defining activity is meeting new people, which the authors did not regard as essential.

Applied strictly, the definition excludes a personal blog or a news website without account-level connections, because neither has a traversable network of users. It also says little about services whose value lies in one-to-many publishing, such as video channels with subscriber lists. Later scholars have therefore used it as a reference point rather than a settled boundary.

Boundaries with Other Media

Social media is best understood by contrast with neighbouring forms. Broadcast media runs on a division of labour in which a small number of producers serve a large audience. Social media collapses that division: a platform's end-users are both contributors and consumers. The open web differs in that a page can be published without any account system or audience feedback, while social platforms make the audience and its reactions visible within the service.

Messaging apps and online forums are harder to separate. Forums and bulletin boards are often the direct ancestors of social media, and messaging apps now include public broadcast channels, group feeds and profile-like features. Some analysts list YouTube, Wikipedia, Yelp, WhatsApp, and Twitter together as social media, even though their designs differ sharply. Research on these services also draws on the idea of persistence, searchability, replicability, and invisible audiences, which can apply to any archived public conversation. The boundaries are blurry because the same features are reassembled differently by each service, and because platforms keep adding features from their neighbours.

Platform Architecture and User Behaviour

Feeds, likes, shares and follower relations are design choices, and they shape how people act. The profile, friends list, comments and private messaging were most prominent in early services, although features varied from site to site. Later platforms shifted emphasis: the field moved from simple chronological updates to algorithmic feeds, personalized recommendations, short-form video, and messaging. Visible counts of likes and followers make social standing legible and encourage users to present themselves in particular ways. Follower relations are often asymmetric, so a user can read a public figure's posts without a reciprocal link, which changes the norms of attention and reply. These architectural decisions are the main reason the same underlying technology can produce very different social practices on different platforms.

Precursors and Early Online Communities

Bulletin boards and Usenet

Before the internet became a mass medium, most online sociability took place on bulletin board systems (BBSs). A BBS was typically a single computer that users reached by dial-up modem, where they posted messages, exchanged files and joined discussions. Many were local operations run by hobbyists, and some could accept only one caller at a time. Commercial online services such as CompuServe and GEnie offered wider reach, but for much of the 1980s BBSs and these services were the main online communities outside academic, government and corporate networks.

Usenet, conceived in 1979 and launched in 1980, extended the BBS model beyond any single machine. Its designers, Tom Truscott and Jim Ellis, built a distributed system in which messages propagated between servers and were organized into a hierarchy of newsgroups by topic. Threaded discussion, persistent archives and topic-based community were all features that later forums and social platforms inherited. Usenet's organizing unit was the subject rather than the person: users found others by shared interests, not by declared ties. That distinction matters later, when platforms began to make a user's own connections the visible structure of the service. Usenet declined in the mid-1990s, when an influx of newcomers from commercial services, the first widespread spam and competition from the Web weakened it.

Six Degrees and the first commercial services

Six Degrees, founded by Andrew Weinreich in 1996 and launched in 1997, is generally treated as the first commercial service to combine member profiles, explicit friend lists and affiliations such as schools. Users could message people within their first-, second- and third-degree connections and see the chain linking them to any other member. The company reported millions of registered users, but the network itself was thin: few of a member's friends were online, so the connection graph offered little value. This is a classic case of a network-effect failing to take hold, because a social service is only useful once enough of a user's contacts have joined.

Six Degrees also had a weak business. It was sold to YouthStream Media Networks, a marketing company, for a reported $125 million, and the site was closed around 2000–2001. Sources differ on the exact year of the sale and closure. The lesson its founders drew, that the concept was ahead of its audience, became a recurring theme in the history of social network services.

Weblogs and LiveJournal

Personal publishing developed on a parallel track. The term "weblog" appeared in the late 1990s, but the practice drew on earlier forms of sharing such as Usenet, e-mail lists and BBSs. LiveJournal, launched in March 1999 by Brad Fitzpatrick, was the platform that most clearly joined publishing and social networking. Users kept diary-like journals, but each account could maintain a list of "friends" whose posts appeared together. The relationship was not symmetrical: two users could have no tie, a mutual tie, or a one-way tie in either direction. Comments were threaded, users could restrict who might comment, and "communities" allowed groups to publish collectively under maintainers who moderated membership. Paid accounts appeared in September 2000, and the server code was released as open source in March 2001.

LiveJournal showed that a blog could function as a social space, in which readership was built through reciprocal links and commenting rather than search or syndication alone. Many of its design choices, including one-way follows, friend-restricted visibility and group spaces, reappeared in later services.

Friendster and MySpace

Friendster, launched in March 2002 by Jonathan Abrams and Peter Chin, built its appeal on the six-degrees idea. Its interface showed how a user was linked to strangers through mutual friends, which made introductions feel less awkward and encouraged dating and friendship uses. Demand grew faster than the company could handle. Its servers struggled with latency and outages, and frustrated users moved to MySpace. Friendster abandoned user profiles in 2011 and closed in 2015.

MySpace, founded in August 2003 by employees of the Internet marketing firm eUniverse, offered customizable profile pages with photographs, music and personal styling. It attracted teenagers and musicians, and it became one of the first social networks with a substantial audience outside the United States. News Corp acquired its parent company in 2005. Its decline came from a different direction. Facebook, which opened in 2004 first to university students, offered a more uniform and less cluttered design, and users gradually moved to it. Commentators have also pointed out that loosely held connections gave users less reason to stay: many accounts on these early networks had few ties, so the social bonds that sustained use were weak. MySpace was sold in 2011 for a fraction of its former valuation.

Platform Milestones and Mainstream Adoption

The mid-2000s wave: Facebook, YouTube and Twitter

Facebook was founded in 2004 by Mark Zuckerberg, along with his Harvard College roommates and fellow students, and it began as a closed network. It spread to Stanford, Yale and Columbia by March of that year. Growth was quick: the site reached 1 million users in December 2004, and by May 2005 it had expanded to over 800 college campuses. Later that year it opened to high school networks and became simply "Facebook". Its feature set also changed rapidly, with News Feed launched in 2006. By 2026, Facebook had more than 3.07 billion monthly active users, according to Wikipedia's figure, though other sources cite smaller numbers from earlier dates.

YouTube followed in 2005. It was founded by Chad Hurley, Jawed Karim, and Steve Chen. The three were early employees at PayPal and had become wealthy after eBay's acquisition of the company. Its first upload, shared by Jawed Karim, was titled "Me at the zoo." Its design made hosting and embedding video straightforward, and Google became its parent company in 2006.

Twitter emerged from a different project. It came out of the podcasting venture Odeo, founded in 2004 by Evan Williams, Biz Stone, and Noah Glass. It launched in 2006 as a service for short, real-time updates, and it was initially meant to be an SMS-based platform for people to share short updates with their friends.

Photo sharing, video and the smartphone shift

Photo sharing was an earlier and separate strand. Flickr's developers began building the photo-sharing service as part of an online role-playing game they were working on called Game Neverending. Flickr officially launched in February 2004. Its design emphasised community: tags, photostreams, comments, faves, and groups became features that later spread across social media. Yahoo bought Flickr in March 2005, and its founders left the company in 2008. Yahoo only launched an official Flickr app in late 2009, which left room for mobile-first competitors.

Instagram's public launch came in October 2010, initially on Apple's iOS platform. It combined capabilities that had previously required separate tools: the integration of camera, editing software, internet connection and social distribution in a single portable device removed several stages from image publishing. More broadly, the move to smartphones changed how often and where people posted, as the smartphone and mobile phone became the main way to access the internet. Flickr's later decline is often linked to this shift; it was acquired by SmugMug, a family-owned company focused on photography, in 2018.

Video sharing also moved onto mobile devices. Flickr itself introduced video sharing in 2008, but the larger change was in short clips filmed and edited on phones.

Short-form video and the rise of Chinese platforms

Musical.ly was founded in 2014 as a short-form app where users could create and share 15-second lip-sync music videos. Its Chinese counterpart was Douyin. Douyin was launched by ByteDance as A.me in September 2016 and was rebranded as Douyin (抖音) in December 2016. Its interface was a notable design choice: Douyin presents a video in full screen and played it automatically as soon as the app opens. Douyin's popularity skyrocketed, and within a year it had 100 million users.

TikTok started out as two distinct apps that ultimately merged: Musical.ly and Douyin. The international version was launched in 2017. In September 2017, they expanded outside of China to select international markets as TikTok. In February 2019, TikTok, together with Douyin, hit one billion downloads globally, excluding Android installs in China. Today the two apps share a similar appearance and logo, but they currently operate independently. A marketing-industry source reports that Douyin passed one billion monthly active users in March 2026, the first short video app in China to cross that line. This expansion is part of the wider globalization of platforms that started outside the United States.

Web 2.0 and participatory platforms

The platforms above are often grouped under the label Web 2.0, a term popularised in 2004–2005 by O'Reilly Media and associated with the World Wide Web becoming a read-write medium. The central idea was that users would generate much of the content, rather than merely reading pages published by companies. Features such as tagging, RSS syndication, embeddable widgets and open programming interfaces made this possible. Flickr illustrates the pattern: its photos could be embedded in a "weblog", and its tools for friends and groups anticipated the social features of later services. These design choices helped turn individual publishing into the networked, feed-based participation that defined the next decade of platforms.

Technical Architecture and Data Infrastructure

The social graph and its data model

Most social media services are built on a social graph, in which users, posts, comments and reactions are stored as entities joined by typed relationships. Engineers often treat this as a directed graph, a structure studied in graph-theory. Facebook's TAO system, described in a 2013 USENIX paper, is a well-documented example. It models data as typed objects and typed, directed associations. Each association has a time field — no, see below.

At most one association of a given type can exist between any two objects. This constraint keeps the schema simple. Each association also carries a timestamp that lets the system return results in recency order, such as the newest comments on a post. Every association has a time field, which is critical for ordering query results by recency.

Storage is usually layered. TAO's persistent layer remains MySQL, while a graph-aware cache sits in front of it. It replaces application-managed look-aside caching for suitable data types while retaining MySQL as the persistent storage layer. Partitioning follows the same logic: objects are assigned permanently to shards, and associations are stored according to their source object.

Scale: caching, replication and consistency

The scale of these systems is much larger than a conventional web application. TAO was reported to handle a billion reads and millions of writes per second, and it runs on thousands of machines spread across regions. Its design is read-heavy. Within each region, leader caches coordinate access to databases and serialize writes, while follower tiers serve clients and scale read throughput. Regions replicate data asynchronously, so reads stay local while writes go to the region that masters each shard.

These choices involve trade-offs. TAO favors availability, read performance, and machine efficiency over strong consistency. Its usual model is eventual consistency, and critical reads can be sent to the master region when fresher data is needed. Beyond the database layer, large platforms also depend on in-memory caches, message queues for asynchronous work, and content delivery networks that store images and video close to users. Real-time features such as live counts and notifications generally rely on streaming connections or queued event processing.

Feed generation: push, pull and hybrid models

Feed generation is one of the most consequential architectural decisions. A post by an account with many followers must reach each follower's timeline, and this can be handled at write time or read time. In the write-time, or fan-out-on-write, model, the post is copied into precomputed timelines. In the read-time, or fan-out-on-read, model, the timeline is assembled when requested. The costs differ sharply by audience size. One systems-design analysis estimates that a user with 30 million followers triggers 30 million cache writes per tweet. The pull model avoids that spike, because the Timeline Service fetches their followee list, retrieves recent tweets from each followee's tweet store, merge-sorts by timestamp, and returns the top N.

Most large services therefore combine the two. For regular users (fewer than ~10,000 followers): use fanout-on-write. For celebrities (more than ~10,000 followers): do NOT fanout. When a follower loads their timeline, the system merges the pre-computed timeline (from regular users, fanout-on-write) with a real-time fetch of celebrity tweets (fanout-on-read). Such descriptions are based on external analyses rather than official documentation, and the exact thresholds are not public.

APIs and third-party ecosystems

An API lets separate software systems exchange requests and data. An API serves as a software "middleman" allowing two applications to talk to each other. Early social platforms opened their APIs to outside developers, which enabled third-party clients, analytics tools and scholarly research. Since 2006, Twitter's APIs have been rich sources of data for researchers studying social phenomena such as misinformation and crisis response.

That openness has narrowed. In early 2023 Twitter (later X) ended its free API tier. These three tiers include a bare-bone free level mostly meant for content posting bots, a $100 per month basic level and a costly enterprise level. Access to larger data volumes was reportedly priced at $42,000 a month at the enterprise level. A mid-level option followed in May 2023, when a $5,000-per-month tier was introduced for startups. Researchers reported serious consequences. One 2024 study describes how the company dismantled its academic access program, leaving many academics unable to continue data-intensive projects. Other platforms have also tightened developer access, particularly after the 2018 Cambridge Analytica revelations.

Mobile clients, push notifications and app stores

Most users reach these services through smartphone apps rather than browsers. Mobile clients keep local caches so the feed can render quickly, synchronise changes in the background, and queue actions when the network is unavailable. Push notifications are delivered through operating-system services, such as Apple Push Notification service on iOS and Firebase Cloud Messaging on Android. Because the operating system controls delivery, background activity can be throttled by battery-saving policies, and platforms must design around these limits.

App stores act as distribution gatekeepers. Apple's App Store and Google Play review submissions, control installation and updates, and take a share of in-app purchases, typically in the range of 15 to 30 percent under standard terms. Those terms have been challenged by developers and regulators, and the competitive implications are discussed in competition law and market power contexts.

Recommendation and Ranking Algorithms

Ranking signals and predictive models

Most content on large platforms reaches users through ranking rather than in chronological order. Rather than one fixed formula, the feed is described by its operator as layered machine-learning models that estimate which items each person will find relevant. Early versions leaned on visible popularity: the company once used how many people reacted to, commented on or shared a post to decide how high it appeared in News Feed. In January 2018 Facebook changed this so that the system would predict which posts you might want to interact with your friends about, and show these posts higher in feed. Internal analyses later described the change as weighting specific reactions and comments more heavily than simple clicks, with increased weights attached to specific types of engagement (comments, shares, reactions etc.). A 2020 adjustment reportedly took away the boost given to reshares, while keeping it high for comments.

The clearest public description of a modern recommender comes from YouTube. Its 2016 system is built in two stages. A candidate-generation model selects a small number (hundreds) of candidate videos from the overall corpus, and then a ranking model scores these candidates. The ranking inputs include a viewer's earlier interactions with similar videos and the channels that produced them. In the ranking stage, the training objective was adjusted so that positive examples were weighted by watch time, which allowed the model to learn the odds of a view more accurately. The authors also reported that the models were trained at very large scale, with approximately one billion parameters and trained on hundreds of billions of examples.

Machine-learning methods

Two families of methods dominate. Collaborative filtering predicts a user's interest from patterns in the behaviour of other users, for example by finding users with similar histories or items that are frequently consumed together, often through matrix factorization that represents users and items as vectors. It works well when interaction data is abundant but struggles with new users and new items, the so-called cold-start problem, which is why platforms usually combine it with content features such as text, audio, images and creator identity. Deep-learning recommenders replace hand-built features with learned embeddings and sequence models that read a user's recent history as an ordered series. The YouTube authors attributed dramatic performance improvements to this approach. Modern systems typically optimize several objectives at once, such as the probability of a click, expected watch time, likes and shares, and then combine these predictions into a single ranking score. Platforms also must handle freshness and noise, which the YouTube team listed alongside scale as one of the three key challenges, along with the other two.

Experimentation and product development

Ranking changes are rarely launched on the strength of offline accuracy alone. Platforms run A/B tests in which randomly assigned groups see different ranking variants, and the outcomes are compared on engagement metrics and on guardrail measures such as reports, complaints and user surveys. Long-term holdouts, in which a small group keeps an older ranking for months, are used to detect effects that appear only later. Facebook documents show this practice in its own work: two documents describe an internal experiment conducted after the company tweaked its internal MSI metric in mid-2018, specifically looking at user's overall response. Because these tests are designed and interpreted by the platform, they measure what the company chooses to optimize, and they seldom test broader social outcomes unless those are added as metrics. This is one reason experimentation is discussed in the causal inference and randomized experiment literatures.

Criticisms and the difficulty of auditing

The most persistent criticism is that optimizing for engagement rewards content that provokes strong reactions. Leaked internal documents from 2018 to 2019 suggested that the MSI change may have pushed the feed toward divisive material; by early 2019 publishers and political parties were raising questions about whether MSI was incentivizing negativity. Academic work points in the same direction. One study found that the engagement gains from the 2018 change came with increased ideological segregation of news consumed by liberals and conservatives and a decrease in information quality. A recent preprint measuring political exposure reported that the share of political content surged by 15.5% in the first year and continued to climb, reaching 22.1% by the fourth year. Meta has disputed such interpretations. It has argued that a ranking change is not the source of the world's divisions, and it has described one researcher's conclusions as implausible.

Auditing these claims is hard because outside researchers rarely see the ranking models, their training data or their experiment logs. Researchers studying a 2018 shift in local party engagement acknowledged that it's "difficult" to know for sure what caused the shift, because they lacked the underlying algorithm. Regulation has begun to address this gap. The EU's Digital Services Act requires platforms to disclose, in "plain and intelligible language," the main parameters of their recommender systems, and very large platforms must provide at least one option for each of their recommender systems which is not based on profiling. Critics argue that disclosure of general parameters remains thin in practice, since details about how their recommendations work are still scarce.

Business Models and the Attention Economy

Most major social media services are free to use at the point of access, so their revenue depends on other sources. The most important is the sale of user attention to advertisers, although subscriptions, premium features and, more recently, the licensing of user-generated content to AI developers have become secondary streams.

Advertising and targeting

Advertising is by far the largest revenue source for the largest platforms. Meta's advertising income reached about $196 billion in 2025, against roughly $201 billion in total revenue. The company attributes its advertising growth to two levers: the number of ad impressions it delivers and the average price per ad, which rose about 9% over the full year. Targeting is what makes those prices defensible. Platforms assemble profiles from declared information, follows and likes, clicks, time spent on content, and inferences drawn from those signals, and then sell delivery to audiences defined by these profiles. The value of an advertising product to a buyer therefore depends on how accurately the platform can predict which user will respond. This is the main reason why data collection and large-scale data analysis sit at the centre of the business model.

Advertisers typically pay according to one of several units: impressions (cost per thousand), clicks, or completed actions such as purchases or app installs. Auction systems determine which ad is shown in a given slot, and that auction logic is closely tied to the ranking models described elsewhere in this entry.

Attention as the traded resource

The idea that attention is the resource being sold draws on an argument made by the economist Herbert A. Simon in 1971: as information becomes abundant, the scarce factor is the capacity of people to absorb and act on it. The attention economy applies this observation to digital markets, where platforms, advertisers and users compete for the same finite stock of time. In practice, each minute a user spends scrolling becomes inventory that can be measured, priced and resold. Users provide attention and personal data in exchange for a free service, platforms package that attention for advertisers, and advertisers pay for access to it. Commentators have noted that brands can still purchase reach, but that attention increasingly has to be earned through content that people choose to watch or share, which has pushed advertisers toward formats such as creator content and memes.

Subscriptions, premium features and data licensing

Alternative revenues remain small relative to advertising but are growing in some cases. Meta's "other revenue" line rose by 50% in 2025, driven mainly by paid business messaging on WhatsApp and by Meta Verified subscriptions. Premium tiers that sell extra features, verification or reduced advertising are also common across the sector, though the revenue they generate varies widely by service.

Data licensing is a newer stream. Reddit reportedly agreed in 2024 to let Google train its AI models on Reddit content for about $60 million a year, and a second deal with OpenAI is estimated at around $70 million a year. By the second quarter of 2026, Reddit's data licensing revenue had reached about $43 million in a single quarter, roughly 5% of its total revenue. The Google agreement was approaching renewal, and Reddit reportedly considered restricting Google's access, citing concern that AI-generated search answers were reducing traffic to the original sites. Reddit has also discussed usage-based "dynamic pricing" for future deals, which would tie payments to how often content is used in AI responses.

Growth, retention and competing goals

Platform operators face several goals that do not always align. Public companies are judged on revenue growth, and because advertising revenue depends on the number of impressions served, growth requires users to spend more time on the service. Retention is therefore central: a platform that loses users loses inventory, and its targeting data becomes less useful. Retention and growth in turn compete with other objectives, including user well-being, content quality, the protection of expression, legal compliance and the cost of moderation. The scale of investment also adds pressure. Meta has guided to capital expenditure of $130 to $145 billion for 2026, which requires continued growth in advertising income. Decisions about ranking, notification design or the introduction of new formats are often evaluated against these commercial metrics, which is one reason why critics argue that engagement-maximizing design can conflict with the interests of users and society.

Types of Platforms and Their Functions

Social media services are usually grouped by the function that dominates their design, although the categories overlap. The main types are social networking, microblogging, media sharing, messaging, forums and professional networks. Each has a typical unit of content (a profile, a short post, a photograph, a video, a thread) and a typical pattern of connection.

Principal categories

General social networking services such as Facebook organize activity around personal profiles, reciprocal friendships and a central news feed. Microblogging services, such as X (formerly Twitter) and Weibo, restrict posts to short text with attached media and rely on public, directed follows and reposting, which lets content move quickly across follower networks; the form is described in more detail in microblogging. Media-sharing services are built around one medium: Flickr and Instagram for photographs, YouTube for long-form video, and Pinterest for images saved into user-curated collections. Messaging services such as WhatsApp and Telegram prioritize private or small-group conversation, although many have added broadcast-style channels and status updates that blur the line with broadcasting. Forums and discussion boards organize content into topic-based threads ranked by votes or replies. Reddit is the largest contemporary example, with its topic communities and vote-based ordering. Professional networks such as LinkedIn center on career profiles, endorsements and recruitment, and their feeds are oriented toward work rather than personal life.

Live streaming and short-form video

Live streaming and short-form video are distinct formats, each with its own norms. Live streaming is synchronous: viewers interact in real time through chat, virtual gifts and paid subscriptions, and the format rewards immediacy and the streamer's ability to respond to the audience. Game streaming on Twitch and live rooms on Chinese platforms such as Douyin illustrate this model, and live commerce, in which products are sold during a broadcast, extends it into e-commerce. Short-form video, by contrast, is typically vertical, full-screen and looping, and discovery depends on recommendation rather than on an existing follower base. This allows new creators to reach audiences quickly. Its conventions include reusable sounds, duets, stitched responses and remix formats, which encourage imitation and make trends spread rapidly.

Hybrid services and the expansion of scope

Most large platforms are hybrids. Facebook combines profiles, groups, messaging, marketplace and video; X combines microblogging with direct messages and video; Reddit combines forum structure with image and video hosting; WeChat combines messaging with payments and third-party mini-programs. Platforms tend to expand their scope for several reasons. Each added function gives users another reason to return, which raises time spent on the service. Functions also make it easier for a platform to keep its existing users and to capture data and advertising inventory from new activities. Where network effects are strong, adding a feature to an existing user base is cheaper than attracting users to a new, single-purpose service. Expansion has a cost, however: a platform with many functions can have conflicting design goals, and users often adopt features in ways the designers did not intend.

Design differences by medium

Text-centred services such as microblogs and forums privilege quotation, reply threads, searchability and argument, so their interfaces emphasize chains of responses. Image-centred services such as Instagram and Pinterest privilege aesthetic curation, grid layouts and the presentation of the self through profiles, so cropping, filters and visual consistency become important. Video-centred services privilege watch time, autoplay and automated recommendation, because content is consumed serially and selected by the system rather than chosen by the user. These design choices shape self-presentation and cultural practice as well as the kinds of content that spread. Character limits, aspect ratios and autoplay are not neutral features; they influence how people write, look and communicate on each service.

Global Adoption, Demographics and the Digital Divide

Measuring who uses social media

Most detailed demographic knowledge comes from national survey programmes, and the best-known series is the one run by the Pew Research Center in the United States. Its survey data are weighted to represent the U.S. adult population by gender, race and ethnicity, education and other categories. Surveys of this kind have limits. Many of them do not distinguish between kinds of use, such as browsing, commenting, or posting original content, and because their samples are modest, they say little about regional differences within countries.

Global totals rely on a different method. DataReportal, which draws on platforms' own advertising-reach figures, put the number of social media user identities at 5.79 billion at the start of April 2026, equal to 69.9% of the world's total population. These are accounts rather than people: the same publisher warns that user figures may not represent unique individuals, and because of duplicate accounts they may exceed the figures published for internet users, or even the total population. Counts of this kind therefore indicate reach, not the number of distinct participants.

Patterns by age, gender, education and income

Age is the sharpest dividing line in the American data. In Pew's 2025 survey, eight in ten adults aged 18 to 29 say they use Instagram, a share higher than in older groups, especially among adults aged 65 and older (19%). YouTube and Facebook are the exceptions: they are the only platforms that a majority of every age group uses. Gender differences are narrower but consistent. Women outnumber men in their use of Facebook (78% versus 63%), Instagram (55% to 44%) and TikTok (42% to 30%), while men are more likely to use X (25% versus 16%) and Reddit (29% to 23%).

Education and income also shape which services people use. About four in ten adults with at least a college degree say they use Reddit, compared with 28% of those with some college and 15% of those with a high school diploma or less. Earlier Pew releases found that X use was higher among people in households earning $100,000 or more. Such differences are descriptive: they show who reports using a service, not why.

Regional variation and China's separate ecosystem

Platform dominance varies sharply by region because access to Western services is not universal. Many countries restrict foreign platforms, and local services often lead within their borders. The clearest example is China, where a separate set of domestic services has developed under state regulation. WeChat, operated by Tencent, combines messaging with payments, business accounts and small applications within a single interface. Weibo, a microblogging service, serves as the country's public-facing platform for news and commentary. Douyin, the Chinese version of ByteDance's short-video app, has become one of the largest video platforms domestically. Facebook, YouTube and X are generally not available to ordinary users in mainland China. This separation means that global user statistics, which are dominated by Western and Indian audiences, can understate the scale of the Chinese market while leaving it out of comparisons of platform reach.

Access inequality and its effects on participation

Adoption depends first on being connected. The International Telecommunication Union (ITU) estimated that about 6 billion people, roughly three-quarters of the world's population, were using the internet in 2025, but 2.2 billion people remained offline. 96 percent of those offline live in low- and middle-income countries. The gaps are large: nearly 94 percent of people in high-income countries are online, compared with just 23 percent in low-income countries. Within countries the divide is also marked. 85 percent of urban residents are online, against 58 percent in rural areas, and in rural Africa the share is lower still, at 21%. The ITU also reports a gender gap, with 77% of men online compared with 71% of women.

These gaps shape who takes part. Because platforms depend on connectivity, people without reliable or affordable access are underrepresented in both user counts and the data used in research. Price matters as well as coverage. ITU analysis shows that affordability gaps exist not only between high- and low-income economies but also within countries, and the cost of a mobile broadband plan can take a far larger share of the income of the poorest population groups.

Adoption trends and the role of mobile internet

Global internet use has grown steadily. The share of people online rose from 60% in 2020 to around three-quarters by 2025, and earlier estimates have been revised as measurement has improved. Growth has been fastest where mobile devices are the main route online. In many low- and middle-income countries, a smartphone is the only computing device a household owns, so mobile broadband, rather than fixed-line service, defines how people reach social platforms. Expansion of mobile networks has therefore been a major driver of new users in developing regions. At the same time, ITU notes that a typical user in a high-income country generates nearly eight times more mobile data than one in a low-income country, which suggests that the intensity and quality of use still differ greatly even where access exists.

Network Effects and Market Structure

Direct and indirect network effects

A social network's value to each user generally rises with the number of other people on it. In a network effect, the benefit of joining comes from who else is present. A messaging app is useful mainly when one's contacts use it, which is a direct effect. A platform that attracts many users also attracts advertisers, creators and third-party developers, and their presence makes the platform more attractive to ordinary users. This is an indirect effect, and it links the two sides of the market. Because of these feedback loops, a service that gets ahead can keep gaining users, while a rival with fewer users offers less value and finds it harder to grow. The result in many categories is a small number of dominant services, often one in each function. Economists describe this as a tendency toward market concentration rather than a stable competitive equilibrium.

Acquisitions and their competitive effects

One way dominant firms have expanded is by buying companies that could have become rivals. Facebook acquired the photo-sharing app Instagram in 2012 and the messaging service WhatsApp in 2014. In December 2020 the US Federal Trade Commission (FTC) sued Meta, alleging that the company had viewed both apps as emerging threats and bought them in a "buy or bury" strategy. The FTC had cleared both deals when they were announced, which made the case legally difficult. On 18 November 2025, after a six-week bench trial, Judge James Boasberg of the US District Court for the District of Columbia ruled that the FTC had not shown Meta currently held a monopoly in personal social networking. The judge held that the relevant question was present market power, not whether Meta had dominance when the deals were made. He also found that the FTC had not defined the market convincingly, because the company competes with services such as TikTok and YouTube. Critics of acquisitions argue that buying nascent rivals removes competition before it can develop, and that such deals are hard to unwind once they close. Defenders argue that the acquisitions brought scale and investment that benefited users, and that the boundaries of a fast-changing market are difficult to fix in court.

Switching costs, data lock-in and interoperability

Even when users are free to leave, several factors discourage them. Switching costs include the loss of a personal contact graph, past content and audience. A user may also be locked into a platform's messaging groups or into a creator's follower base. Data accumulated over years can give an incumbent advantages that new entrants cannot easily replicate, a problem often described as data lock-in. Interoperability, meaning the ability of users on one service to communicate with or move their data to another, would reduce these barriers, but services have seldom offered it voluntarily. The EU Digital Markets Act (DMA) takes a regulatory approach to this problem. It requires designated gatekeepers to allow interoperability and data access in defined circumstances, and to avoid combining personal data across services without consent.

Antitrust cases and policy responses

Antitrust enforcement has followed two tracks. In the United States, the courts apply monopolization law after the fact. In the case against Google, the Department of Justice and state attorneys general argued that payments to be the default search engine on phones and browsers maintained a monopoly. Judge Amit Mehta ruled against Google in August 2024. In September 2025 he ordered behavioural remedies rather than a break-up, including an end to exclusive default deals and the sharing of certain search data with rivals. Google appealed in January 2026, and the DOJ and states cross-appealed, so the appeal process was still ongoing as of late 2026. The European Union has taken a different approach. The DMA, which became applicable on 2 May 2023, sets objective thresholds for designating "gatekeepers" and imposes ex ante obligations, such as prohibitions on self-preferencing and on certain uses of combined data. In September 2023 the European Commission designated six gatekeepers, including Meta and ByteDance (TikTok), covering 22 core platform services. Fines for non-compliance can reach 10 per cent of worldwide turnover. Debates continue over whether these measures can change market structure or mainly manage the behaviour of incumbents.

Content Moderation and Platform Governance

Content moderation is the set of rules, tools and staff that decide which posts, accounts and media a platform removes, labels, demotes or restricts. Most services publish community guidelines and enforce them through a mix of user reports, automated classifiers and human reviewers. Automation can process very large volumes of content, but it handles context-dependent material such as satire, reclaimed slurs, news footage and political speech poorly, so contested cases are usually routed to people.

Automation, human review and their limits

The Santa Clara Principles, drafted in 2018 by academics and advocacy groups, remain the most widely cited baseline for this area. They say that automated tools should be used to identify or remove content only when their accuracy has been shown to be sufficiently high, whether or not a human reviews the result. The same document asks platforms to disclose the extent of human oversight over automated decisions and to let users request human review. In practice, the difficulty is that classifiers trained on majority languages and common formats often miss local idioms, coded hostility and images whose meaning depends on context, while also removing lawful speech by mistake. Error rates are rarely published in a form that outsiders can check.

Appeals, transparency reports and independent oversight

Appeals are the main channel through which users contest decisions. Meta's Oversight Board, which began hearing cases in October 2020, is the best-known example of an external appellate body. The board has issued upwards of 200 decisions and 320 recommendations, according to its five-year assessment published in December 2025. Meta has implemented all of its binding decisions and around 75% of its recommendations. Critics note the limited reach of this model. A 2022 academic review counted 20 decisions in 2021 out of 130 shortlisted cases, against roughly a million cases submitted. The Board also has to choose cases, and it prioritises cases with the potential to affect many users around the world, that are of critical importance to public discourse, or that raise important questions about Meta's policies. Reversing a decision does not automatically take a case out of its review, according to the Board's own interpretation of its charter. Summary decisions are a transparency tool, publishing information about cases where Meta reversed its original decision after receiving notice of the appeal.

Transparency reporting is the second pillar. The Santa Clara Principles call for three things: published numbers of removals and suspensions, clear notice to affected users, and meaningful appeals. The European Union's Digital Services Act, which came fully into effect on Nov. 16, 2022, turns part of this into law. Platforms with at least 45 million monthly active users in the EU are designated as very large platforms, and they must publish their transparency reports at least every six months. These reports must also disclose the human resources dedicated to content moderation, broken down by each applicable official language of the Member States, along with appeal notices and out-of-court dispute settlements. Analysts have found, however, that there is no standardised categorisation of illegal content across platforms, sometimes not even within the individual reports, which makes cross-platform comparison difficult.

Language, region and political context

Moderation quality varies considerably across languages and political settings. The Santa Clara Principles stress cultural competence, meaning that those who make moderation and appeal decisions should understand the language, culture and political and social context of the posts they review, and they ask that appeal mechanisms be available in the languages users actually write in. The Oversight Board's caseload illustrates the point. In one case it addressed a Facebook post in Thai urging others not to support shops or companies employing Cambodians. In another group of cases involving threats against authorities in Ethiopia, Pakistan, Ukraine and Italy, Meta reversed its takedown decisions after the board's intervention, admitting that the comments were rhetorical expressions of criticism rather than credible threats. Such cases show how the same wording can be read differently depending on the local situation, and how difficult it is for a single global rulebook to capture that variation. Government requests add a further layer, since the Santa Clara Principles ask companies to recognise the particular risks to users' rights when states shape moderation rules or enforcement.

Competing objectives

Governance decisions are shaped by competing goals. Freedom of expression, user safety, legal compliance and commercial interests do not always point in the same direction. Removing a deepfake that targets a politician or a private individual may protect dignity and reduce harassment, but it also requires a judgement about the boundaries of permissible speech; the Oversight Board has ruled both ways on such material, for example ordering removal of a deepfake video of a UK local politician that contained hate speech against refugees. Advertising revenue, growth targets and the cost of human review all affect how much capacity a platform devotes to appeals and local-language enforcement. The Board's own assessments have noted that its calls for demystification of how certain algorithms work have only been partially addressed, which illustrates how hard it is to make governance commitments that touch core product design. Whether these structures produce accountability, or mainly provide a procedural veneer, remains an open debate among researchers, regulators and civil society, and it is closely linked to questions of freedom of expression and human rights standards.

Privacy, Data Collection and Data Protection

Categories of collected data

Social media services gather far more than the material users choose to publish. Explicit data such as profiles, posts and messages is only the visible layer. Behavioural data is generated by how people use the service. One teaching tool that simulates platform tracking records engagement metrics from the user's browsing session, such as time spent viewing an image, periods of inactivity, reactions, follows and whether content is shared privately or publicly. Location data is a second major category. Many social media platforms have built-in check-in features, but location can also come from Bluetooth or WiFi signals detected by nearby devices, and some platforms use third-party location data providers that collect information from GPS, IP addresses and cell tower data. A 2024-era analysis by a privacy software vendor, Surfshark, reported that even when users disable precise location sharing, companies can still infer approximate location, as outlined in their privacy policies.

Inferred data is the most consequential category because users rarely provide it. Advertising and analytics firms combine tracking cookies, device and browser fingerprinting, social media data, location data, and the behaviour of people in a user's location and social circle to estimate interests and preferences. Tracking also crosses service boundaries: tracking libraries follow app usage and web browsing histories, along with data from partnering tracking companies.

Consent and privacy settings

Consent is the most common legal and practical basis for this processing, but it has structural limits. Privacy settings typically govern what a user can see or share with other users; they do not necessarily control what the platform infers, what it purchases from brokers, or what it has already collected. Opt-outs also operate unevenly. Even when a user restricts location sharing, their location may still be inferred from the device's IP address or from metadata embedded in photos. Default settings, long policies and interface design shape choices in ways that many users do not examine, and researchers have debated whether the notices people accept can meaningfully be called informed consent. In the European Union, the GDPR sets a higher standard, under which consent must be freely given, specific, informed and unambiguous, but compliance in practice remains contested.

The Cambridge Analytica episode

The episode illustrates how platform data-access rules, rather than a conventional hack, can expose personal data at scale. According to reports from the New York Times and the Observer, the firm began obtaining data from over 50 million Facebook users in 2014 through an app developed by a British researcher, Aleksandr Kogan. The app's permissions reached beyond its own users: the developer was able to gather information not only on the users of the app, but also on all of their Facebook friends. Later estimates put the number of affected users at about 87 million. The scandal became public in March 2018 through whistleblower accounts and investigative reporting. Facebook disputed the characterisation of the event as a breach, arguing that Kogan obtained access to information from users who chose to sign up to his app, and everyone involved gave their consent. Critics countered that what made such activities possible was largely due to Facebook's privacy policies at the time. The firm itself announced it was declaring bankruptcy and shutting down in May 2018, and governments around the world launched investigations. The affair shifted regulatory attention towards third-party access and the enforcement of platform data policies. In 2019 the US Federal Trade Commission imposed a penalty of about $5 billion on Facebook, and the platform subsequently restricted developer access to friend-related data.

Data protection regimes

The EU General Data Protection Regulation, in force since 2018, is the most extensive regime applied to platforms. It covers the processing of personal data of people in the EU regardless of where the company is based, and it sets out lawful bases for processing, rights of access and erasure, and substantial fines. Enforcement has been concentrated in Ireland's Data Protection Commission, which acts as lead regulator for many large platforms. Its largest penalty to date followed a dispute over transatlantic data flows: in May 2023 the Irish Data Protection Commission fined Meta Platforms Ireland €1.2 billion for unlawfully transferring personal data to the United States. The DPC found that Meta's reliance on standard contractual clauses and supplementary measures did not satisfy the requirements arising from the Court of Justice's Schrems II judgment. Meta said it would appeal. Earlier, the previous largest fine was a 746 million euro charge against Amazon in 2021. Other penalties against Meta concern transparency, children's data and security design. Taken together, Meta's GDPR fines were reported as roughly EUR 2.83 billion in finalised fines since September 2021, although the scale of fines has not by itself resolved questions about how platforms collect and infer data. Outside the EU, protection is more fragmented: the United States relies on sector-specific federal rules and a growing number of state laws.

Politics, Civic Participation and Social Movements

Campaigns and political communication

Social media has become a standard channel for election campaigns, alongside broadcast and print media. Campaigns use platforms to publish messages directly to supporters, to recruit volunteers and donors, and to buy targeted advertising. Targeting depends on the data described in the privacy and advertising sections, which allows different audiences to receive different messages, sometimes with little public visibility. The 2016 US election showed how state-linked influence operations could use fake accounts, pages and paid ads to amplify divisive content, and this led to sustained scrutiny of political advertising and of platform transparency.

Regulators have approached political advertising in different ways. The European Union's regulation on the transparency and targeting of political advertising began to apply in October 2025, and several large platforms stopped accepting political ads in the EU in response. Policy in other jurisdictions remains uneven, and the reach of organic (unpaid) political content, which is harder to regulate, is a continuing concern.

Mobilization: case studies and their interpretation

The Arab Spring, beginning in late 2010, is the most studied case. A 2016 survey of 88 peer-reviewed articles found that the bulk of the research argued that social media enabled or facilitated protest by giving voice to people in societies where traditional media were state-controlled. Survey evidence from the period is often cited: most Egyptian and Tunisian respondents said they used Facebook to spread information or organize actions in early 2011. Data from the University of Washington indicated that political posting on Twitter about Egypt rose sharply in the week before President Hosni Mubarak resigned. Researchers also stress the uneven effects. Platform use was lower in Syria and Yemen, and offline networks, grievances and repression mattered as much as online tools. Egypt and other states responded with monitoring, blocking and, in Egypt's case, a near-total shutdown of internet access.

Occupy Wall Street, which began in September 2011, used online forums, social platforms and tumblr-style image sharing to coordinate encampments and spread its "99%" framing. Scholars have debated whether online networks gave the movement wider reach or whether they produced loose affiliations that were hard to sustain without physical organization.

#MeToo, which spread as a hashtag in October 2017 after allegations against Harvey Weinstein, allowed large numbers of people to share personal accounts of sexual harassment and assault across countries. Its origins predate the hashtag, and studies focus on how online sharing shifted public agendas and institutional responses. Black Lives Matter, which began in 2013 after the acquittal of George Zimmerman in the killing of Trayvon Martin, used hashtags and live video of police violence to challenge mainstream accounts of events. Research on hashtag activism, such as analyses of the Breonna Taylor case, finds that online campaigns combine news-sharing, protest organization and tagging of influential accounts, and that a small set of key connectors often shapes which messages travel furthest.

Government use, surveillance and shutdowns

Governments and political actors use platforms for public communication, crisis information and policing, and they also monitor social media for surveillance and intelligence purposes. Internet shutdowns are a related form of control. Access Now and the #KeepItOn coalition recorded a record 313 shutdowns in 52 countries in 2025, and they report that both democratic and autocratic governments impose them, often during protests, elections or conflict. Platform-specific blocks have also grown: the coalition counted 94 shutdowns targeting services such as WhatsApp, Facebook, Instagram, TikTok and Telegram in 40 countries, up from 77 in 2024. Governments have also targeted circumvention tools such as VPNs. These measures affect organizers and journalists directly, and they often limit the documentation of abuses.

The debate over democratic participation

Scholars disagree on whether social media strengthens or weakens democracy. Optimists argue that it lowers the cost of information and participation, lets dispersed groups find one another, gives marginalized voices a route into public debate, and allows citizens to monitor officials. Sceptics point to several problems: "slacktivism", in which low-effort online actions substitute for sustained engagement; the use of platforms by authoritarian governments for surveillance and censorship, a concern Evgeny Morozov developed in his critique of "cyber-utopianism"; the spread of disinformation and foreign influence operations; and the effects of engagement-optimized ranking on polarization and public deliberation. Empirical evidence is mixed. Effects vary by country, by movement and by the offline organizations that exist alongside online activity, and much of the relevant data is controlled by platforms, which makes independent assessment difficult.

News Consumption, Information Ecosystems and Misinformation

From broadcast and print to platform feeds

News increasingly reaches audiences through platforms rather than through publishers' own channels. The 2026 Reuters Institute Digital News Report, based on nearly 100,000 respondents in 48 markets, found that 54 percent of respondents said that they use social media and video networks to access online news, compared to 51 percent who use the news organizations' own websites and apps. The report's authors call this trend the intensifying 'platformisation' of news consumption. Television has also lost ground: across the markets covered, social media and video networks are, on average, more popular than both TV and owned news websites and apps as sources of news.

In the United States the pattern is similar. About half of U.S. adults (53%) say that they get at least some news from social media. Growth is uneven across platforms. A fifth of U.S. adults now regularly get news on TikTok, up from just 3% in 2020. Among adults under 30, 43% say they regularly get news from TikTok, up from 9% in 2020. Reach and news use also diverge: although 7 out of 10 respondents reported using Facebook, only 38% said they got news from the platform.

Gatekeepers and trust

The traditional role of editors as gatekeepers has been partly taken over by ranking systems, recommendation feeds and individual creators. About one-in-five U.S. adults say they regularly get news from news influencers on social media, and this is especially common among younger adults. Trust has not kept pace with reach. Trust in news has fallen to its lowest level since measurement began in 2015, concerns about misinformation have increased, and interest in news continues to decline in many countries. Researchers generally link this partly to the fragmentation of the information environment, although the causal weight of platforms relative to other factors is contested.

How false content spreads

The most cited study of diffusion is Vosoughi, Roy and Aral (2018), which analysed approximately 126,000 cascades of news stories spreading on Twitter, tweeted by about 3 million people over 4.5 million times. Stories were classified as true or false by cross-checking with independent fact-checkers, and controlling for many factors, false news was 70% more likely to be retweeted than the truth. The most widely spread 1% of false cascades reached between 1,000 and 100,000 people, while the truth rarely diffused to more than 1,000 people. The authors attribute part of this to novelty: false items were more unfamiliar to readers, and people were more likely to share novel information.

The same study examined the role of automated accounts. Its authors found that the spread of false information is essentially not due to bots programmed to disseminate inaccurate stories; instead, false news speeds faster around Twitter due to people retweeting inaccurate news items. This finding concerns organic sharing on one platform before 2017 and does not rule out coordinated campaigns. Coordinated inauthentic behaviour, in which networks of accounts act together to deceive audiences, is a separate phenomenon that platforms detect and remove as a matter of policy, and its measured reach varies by country and event.

Fact-checking and labelling

Platforms have used several responses: third-party fact-checking partnerships, warning labels, reduced distribution of flagged posts and, more recently, crowd-sourced annotation. Meta, for example, replaced third-party fact-checking in the United States with Community Notes in 2025. Some analysts have linked a recent rise in Facebook's news use to that change, though this remains a hypothesis: Facebook seems to have regained its value for news in the last two years, which could also be related to its switch to Community Notes. Evaluations of labels generally find that warnings can reduce belief in or sharing of flagged items, but effects on overall exposure and on unlabelled material are less clear, and measurement is constrained by platform-controlled data.

Challenges for journalism

Publishers depend heavily on platform referrals, and that dependence has become a financial risk. Data from the analytics provider Chartbeat, cited by the Reuters Institute, show that referral traffic to news sites from Facebook (-43%) and X, formerly Twitter (-46%) fell substantially over the last three years. Search referrals are also weakening: aggregate traffic to hundreds of news sites from Google search has already started to dip, with publishers that rely on lifestyle content saying they have been particularly affected by the roll out of Google's AI overviews. The 2026 report warns that these changes are adding pressure to traditional publishers by weakening direct relationships with audiences. For newsrooms, this combination of algorithmic distribution and declining direct reach makes revenue planning difficult, and it pushes editorial decisions toward formats and topics that perform well on platforms, a tension that intersects with the attention-driven logic discussed elsewhere in this article.

Polarization, Echo Chambers and Filter Bubbles

Two metaphors dominate public debate about social media and political division. An echo chamber is a social epistemic structure from which other relevant voices have been actively excluded and discredited, according to Nguyen (2020). A filter bubble is different: it describes personalization that can narrow the range of viewpoints a user meets, sometimes unintentionally. Researchers do not always use these terms consistently, and there is not always a clear consensus on exact definitions of these concepts, even among academics.

The underlying hypothesis combines two mechanisms. Recommendation systems may favour content a user already agrees with, while people tend to seek out like-minded contacts. Together, the argument goes, they limit exposure to diverse views and can push opinions toward extremes. Sociologists describe the second mechanism as homophily, the tendency of people to form ties with those who resemble them. The first mechanism is the one most often attributed to platform design, but the empirical question is whether the two actually combine to produce isolation at scale.

Observational studies have not generally supported the strong version of the hypothesis. A review of social science work concluded that echo chambers are much less widespread than is commonly assumed, finds no support for the filter bubble hypothesis, and offers a mixed picture on polarisation. Another review found that most users encounter heterogeneous content, and algorithmic filtering appears to account for only a limited share of observed polarisation. Data from Twitter point the same way. In a survey linked to public accounts, a Barberá et al. analysis (2015) found that for about 75% of Twitter users from Germany, Spain, and the United States, at least 25% of their following are accounts with a different ideology. Eady and colleagues (2019) reported a substantial amount of overlap in the ideological distributions of accounts followed by users on opposite ends of the political spectrum. Some studies, however, point the other way. Mønsted and Lehmann (2022) found that Twitter's recommender system helped create polarised factions around vaccine scepticism during the COVID-19 pandemic, which suggests that effects depend on topic and context.

Experiments on Facebook and Instagram during the 2020 US presidential election produced some of the most-cited null results. In one field experiment with 23,377 adult users, reducing exposure to like-minded political sources had no measurable effects on attitudinal measures such as affective polarization, ideological extremity, candidate evaluations and belief in false claims. A separate study found that replacing the ranked feed with a chronological one did not reduce political polarization, which the researchers measured by surveying users before, during, and after the experiment. The same chronological feed, however, increased the amount of misinformation people saw, so changing the ranking altered what people encountered even where attitudes did not move. Researchers therefore tend to describe the results as showing that exposure can be changed substantially while attitudes shift little over a short period.

Measuring exposure is itself difficult. Researchers have to separate what the platform's architecture does from what users choose, a distinction that is hard to make in observational data. One analysis of the Meta experiments argues that structural exposure mechanisms shape information environments independently of self-selection, which is why experiments that manipulate feeds are useful. Platform-controlled data raise further problems. Meta's involvement gave the 2020 research team unprecedented access to user data with no substantive restrictions, but critics noted that during the study, Facebook instituted a series of emergency changes to its algorithm to reduce the spread of misinformation. Commentators have also argued that the experiments ran during a period of special safeguards, which may limit how far their findings generalize to ordinary operation. Survey-based measures of polarization, and the assumption that following an account equals seeing its posts, add further uncertainty.

Cross-national evidence is thin. The review cited above observed that research outside the United States is limited, which means that most findings about echo chambers come from one political system with a two-party structure. Few studies compare platforms, languages or media systems directly, so claims about how exposure differs between countries rest largely on extrapolation. The role of offline social networks is similarly under-studied in comparison with online data. Since people's offline ties often determine whom they follow online, offline homophily may account for much of what looks like online segregation, but separating the two requires linked data that remain rare. A more recent review of Twitter/X research reached a nuanced conclusion: the platform may not foster sustained cross-ideological engagement, but it does facilitate regular encounters with political diversity.

Mental Health, Well-being and Adolescent Use

Concern about social media and young people's psychological health rests mainly on correlational research. Across many cross-sectional surveys, heavier use has been associated with self-reported anxiety, depression, loneliness, poor sleep and low self-esteem. Some commentators have linked a rise in adolescent distress in the early 2010s, a period that coincided with the spread of smartphones and image-centred platforms, to these patterns. Cross-sectional associations of this kind cannot show direction. Distressed teenagers may also turn to their phones more, and a third factor such as family conflict or offline isolation may explain both.

Size of the observed associations

Meta-analyses generally find small average associations rather than large ones. One meta-analysis of 45 studies covering more than 150,000 adolescents reported pooled correlations of roughly 0.12 with depressive symptoms, 0.10 with anxiety and 0.15 with loneliness, alongside a weak negative link with self-esteem. An umbrella review of earlier reviews found that associations with adolescent ill-being ranged from very small to moderate, and that the individual estimates varied considerably; in one meta-analysis, estimates for depressive symptoms ran from slightly negative to about 0.33. Another meta-analysis of studies on adolescents aged 11 to 18 reported a correlation of 0.11 but very high heterogeneity, which the authors took to mean that overall use measures obscure important differences between types of use.

The most contested work is by Amy Orben and Andrew Przybylski. Their 2019 analysis of large UK and US datasets, using specification curve analysis, concluded that the link between screen time and well-being was too small to justify policy change. A companion PNAS study that used longitudinal panel data found that effects in both directions were trivial in size and possibly dependent on gender. Critics argued that six analytical choices in the original work shrank the estimated effects, and that a reanalysis of the same datasets showed practically meaningful differences, particularly for girls who use social media.

Cyberbullying, social comparison and body image

Research on mechanisms has focused on three processes. Cyberbullying can be persistent, visible and difficult to escape, and victimization is consistently associated with depression and anxiety, although the strength of the effect and the direction of causation are hard to establish. Social comparison is a second mechanism: experimental and experience-sampling studies have found that passive browsing, meaning scrolling through others' content without interacting, is followed by lower affective well-being, and that envy partly explains this. A third concern is body image, since image-based platforms emphasize appearance. In 2021, internal research leaked from a major platform reported that some teenage girls said Instagram worsened their body image; the company disputed how the findings were characterized.

Effects are not uniform across types of use. A review covering the COVID-19 period found that one-to-one communication and self-disclosure within mutual friendships were associated with less loneliness and stress, while addictive patterns of use, social comparison, fear of missing out and exposure to negative content were associated with worse outcomes. Meta-analytic work also reports that associations are often stronger for girls and younger adolescents, though these differences are not consistent across studies.

Methodological debate

Several methodological problems explain why the literature remains disputed. Most studies rely on self-reported time spent online, a measure that bundles very different activities together. Heterogeneity is often very high; in some meta-analyses, more than 90 percent of the variance between studies reflects differences rather than sampling error. Results also depend on analytical choices, such as whether non-linear relationships are tested, which control variables are included and whether individual items or validated scales are used. Reverse causation is a further problem, because a teenager who is already anxious may check social media more often.

Research directions

Several designs aim to address these limits. Experience-sampling studies send participants repeated prompts during daily life, which allows researchers to link momentary use with subsequent mood. One early study texted young adults five times a day for two weeks and found that Facebook use predicted later declines in affective well-being. Longitudinal panels with repeated measures can separate short-term from lasting effects, and experiments that vary how people use platforms can test causal claims more directly. Many researchers argue that progress requires platform-collected data on what users actually see and do, which companies have been reluctant or restricted in sharing, and that studies should move from overall screen time toward specific features and types of use. These approaches are discussed further in the sections on causal inference, natural experiments and reproducibility.

Culture, Language and Everyday Practices

A large share of social media activity consists of making, passing on and altering shared cultural material. The most visible form is the internet meme. Richard Dawkins used the word in his 1976 book The Selfish Gene for small cultural units that spread from person to person by copying or imitation, an analogy with genes. Limor Shifman later defined internet memes as digital content units that share common characteristics, are created with awareness of each other, and are circulated, imitated and transformed by many users. Her framework also separates such clusters from single items that simply spread widely, which is the older sense of "going viral."

Remixing is central to this activity. Meme genres range from amateurish lip-synch videos, in which a person matches their lip movements to a song, to reaction images that edit an existing photograph to express an attitude. Alongside these genres sits a visual vernacular of recurring framings, cropped screenshots, fixed image templates, stickers and emoji. These conventions let audiences recognise a joke at a glance, and they can often be understood across language barriers more readily than running text.

Hashtags show how a convention moves from one community to another. Chris Messina borrowed the idea from the way Internet Relay Chat (IRC) used the pound sign to name chat rooms. He first suggested using a hashtag on Twitter on 23 August 2007, aiming simply to help people group topics together. Twitter reportedly rejected the idea, telling him the tags were for nerds and would never catch on. The practice took hold during the October 2007 San Diego wildfires, when a user was asked to add #sandiegofire to tweets about the fire, and others quickly began using the same tag. Twitter built official support for hashtags in July 2009, and the American Dialect Society later named the hashtag its word of the year for 2012. By the end of the decade the hashtag could be seen on most major platforms, and Instagram eventually set a limit on the number of hashtags per post to curb abuse.

Platform-specific slang follows a similar pattern. Each service develops its own abbreviations, in-jokes and named formats, such as duets and stitches on short-video apps, and users mostly learn these conventions by participating rather than through instruction. Norms of self-presentation are shaped by the audience a user imagines, which is often wider and more mixed than the people who actually see a post. Researchers describe this mismatch as context collapse, a term widely associated with work by Alice Marwick and danah boyd on Twitter in 2011.

Creator culture draws on an older psychological pattern. Donald Horton and Richard Wohl introduced the concept of parasocial interaction in 1956 to describe audience responses to media performers, particularly on television. They argued that viewers act as if they are involved in an actual interpersonal interaction. A parasocial relationship is a one-sided relationship that media users form as a result of exposure to media personas. Social media makes this pattern more interactive: it has enabled individuals to feel closer to public figures through direct communication and personal storytelling. Creators reply to comments, broadcast live and narrate ordinary routines, which can intensify the sense of closeness. Because the bond is asymmetrical, audiences may also expect personal access that creators cannot fully provide, and the resulting pressure is a recurring theme in discussions of creator labour and wellbeing.

Ranking systems strongly influence which sounds, dance formats, hashtags and memes spread. Recommenders that reward early engagement tend to favour content that is quick to understand, easy to reuse and built on existing audio or templates. Creators watch what is promoted and imitate it, so a trend becomes a loop of observation, imitation and amplification, and a sound can travel through many communities with its origin lost. This concentrates visibility on a small number of formats and makes it harder to say who started a cultural movement. Because platforms rarely publish their ranking criteria, researchers rely on case studies, platform-released data and creators' own accounts. See algorithm and attention-mechanism for the underlying mechanisms, and culture, language and socialization for the wider frameworks.

Commerce, Creator Economy and Labor

Influencer marketing and affiliate commerce

Sponsored posts, paid partnerships and affiliate links now form a large share of advertising on social platforms. In the United States, the Federal Trade Commission treats any financial tie between an endorser and a marketer as something that must be disclosed. Any material connection between an endorser and the marketer of a product must be disclosed clearly and conspicuously, including compensation such as commissions, free products, discounts, or flat fees. The Endorsement Guides were revised in 2023 to cover these relationships in more detail.

Disclosure rules govern how a label is presented, not only whether it exists. Disclosures should be visible without a click, and platform tools are not a complete substitute: FTC guidance makes clear that platform-provided disclosure tools, such as a paid-partnership label, can supplement a disclosure but do not replace it. Vague terms such as "collab" or "ambassador" are generally considered insufficient on their own. Where a message is seen in an image and heard in a video, the disclosure must match that format. If an endorsement is made through visual means, the disclosure must also be visual; if it is audible, the disclosure must also be audible. Affiliate content follows the same logic, with the disclosure placed near the link itself.

Enforcement actions have typically involved omitted disclosures, disclosures hidden in captions or descriptions, and reliance on hashtags alone. Social commerce features, such as in-app product tags, shoppable video and live-selling sessions, extend these obligations into direct transactions. Their rapid growth has made it harder to separate editorial recommendation from advertising, a tension that disclosure rules attempt to manage rather than resolve.

Creator monetization and income precarity

Creators earn through several channels: shares of advertising revenue, subscriptions and memberships, tips and virtual gifts during live streams, brand deals, and sales of their own products. Platforms set the split for each channel. A payments-industry analysis reports that platforms take 20% to 45% of gross revenue, and payouts arrive weeks or months after content is published. Such figures come from vendor analyses and vary by program, so they should be treated as indicative.

Income distribution is highly uneven. A large-scale study of about 153,000 YouTube channels with at least 100,000 subscribers found that monetization and networking strategies have become more important over time, but differ substantially between channel sizes, content categories, and geographic locations. The authors argue that low barriers to entry, performance-based pay and cumulative advantage are bound to produce a long-tailed income distribution where the vast majority of participants will not be able to find economic success. Estimates of typical earnings are less reliable. Some compilations put the median creator below a few thousand dollars a year, but these figures draw on multiple survey-based reports, and the methodology varies across sources.

Dependence on a single platform adds another layer of risk. Payment programs are revised or discontinued at the platform's discretion. TikTok's Creator Fund, launched in 2021, disbursed hundreds of millions to select creators, though payouts were often criticized as insufficient. It was later replaced by other programs.

Platform labor: moderators, gig workers and users

Content moderation is a major form of paid platform labor, and it is often outsourced. In Kenya, moderators working for Meta through outsourcing firms have been central to litigation. Daniel Motaung, a content moderator for Facebook in Nairobi, filed a lawsuit against Meta and its subcontractor Sama in May 2022. Moderators have alleged that the work caused post-traumatic stress and that they were dismissed for organizing. In 2024, the Court of Appeal found that the company could be sued in Kenyan courts. A 2025 investigation reported that Meta moved its regional content moderation to Ghana, recruiting workers through a French multinational outsourcing firm. These are allegations and reported findings, not adjudicated conclusions about all moderation work. In the United States, a California judge approved an $85 million settlement in 2021 between Facebook and more than 10,000 content moderators.

Platform-based work more broadly is marked by insecurity. A qualitative study of workers on several gig platforms found that those who depend on the platform to cover basic living expenses express more dissatisfaction and experience more precarity. Creators are not employees in the same sense, but their situation shares this pattern of dependence on opaque platform rules.

Users also contribute unpaid labor when they produce posts, comments, reviews and data that make platforms valuable. Scholars have described this as "free labor," a concept that has been applied to the engagement and content that feed advertising systems. The claim that users are underpaid for this work is contested, since many users see it as voluntary participation.

Regulatory responses

Regulators have responded on several fronts. Consumer protection rules on sponsored content are the most established, as described above. In the European Union, the Digital Services Act imposes transparency duties on online advertising, and national consumer authorities have pursued enforcement against influencers. Labor protection is more fragmented. In Kenya, the government proposed a Business Laws (Amendment) Bill in 2024 that would prevent the clients of outsourcing companies from being sued by Kenyan contractors. Such proposals show that where legal responsibility for platform labor sits, between platforms, contractors and national courts, remains a central policy question. See also e-commerce, economic regulation, legislation and human rights.

Law, Regulation and Litigation

Intermediary liability

Legal systems differ on whether a platform should answer for what its users publish. In the United States, Section 230 of the Communications Decency Act (1996) has long shielded online services from being treated as the publisher of third-party content, and it became the default defence in nearly every lawsuit against a platform. That shield is now being narrowed by courts. In April 2026 the Massachusetts Supreme Judicial Court allowed the state's claims that Meta built an addictive product and misled the public to proceed, reasoning that the alleged harm came from the company's own design rather than from user posts. On 10 August 2026 the Ninth Circuit held that Section 230 is a defence to be raised against liability, not a bar to being sued, which meant that Meta's and TikTok's immediate appeals could not halt the claims.

The European Union takes a different structure. The Digital Services Act (DSA) keeps conditional exemptions for hosting services but attaches duties to them: notice-and-action procedures, statements of reasons for moderation decisions, transparency reporting, and, for very large platforms, annual risk assessments and independent audits. Fines can reach 6% of global annual turnover.

Competition and digital markets

The Digital Markets Act (DMA) applies ex ante obligations to firms designated as "gatekeepers" rather than relying only on case-by-case competition law. Seven companies have been designated, including Meta and ByteDance, the owner of TikTok. In April 2025 the European Commission fined Meta €200 million over its "pay or consent" advertising model, finding that users were not offered a genuinely equivalent option with less personal data. Meta appealed, and later agreed a revised model that the Commission said it would monitor. In June 2026 the EU General Court partly sided with Meta, ruling that the Commission had not properly reasoned its designation of Facebook Marketplace while upholding the designation for Messenger. These cases show how gatekeeper rules interact with broader concerns about market power and personal data use.

Age rules and child online safety

Age limits have moved from platform terms of service into statute. Australia became the first country to enforce a nationwide ban on accounts for under-16s, which took effect on 10 December 2025. Platforms must take "reasonable steps", such as age-assurance tools, to block such accounts. The government announced in June 2026 that it would raise the maximum penalty to A$99 million and give the eSafety regulator power to compel evidence of compliance. The regulator is investigating Facebook, Instagram, Snapchat, TikTok and YouTube. Compliance remains contested: a study published in The BMJ found that most under-16 participants were still using social media three months after the restrictions began, and Reddit has challenged the law in the High Court. In the United States, several states have passed age-verification and parental-consent laws, many of which have been litigated, while federal child-safety bills remain under debate. Critics raise concerns about privacy, free expression and the practical effectiveness of such measures, and these debates connect to wider questions of freedom of expression and legislative design.

Litigation over design, harm and data

Private and public lawsuits have become a major route for regulating platform design. In March 2026 a California jury, hearing a bellwether case, found Meta and YouTube liable for negligently designing features that harmed a young plaintiff's mental health, and awarded $3 million in damages, with a further $3 million in punitive damages recommended. TikTok and Snap settled before that trial began. Courts have generally separated claims about product features, such as infinite scroll, autoplay and recommendation systems, from claims about content, allowing the former to proceed. Thousands of cases are now consolidated in federal and state proceedings, including claims by school districts and state attorneys general. Judicial reasoning in this area is still developing, and appeals and later trials will determine how far the design-based approach extends. Their outcomes are likely to shape how governments and courts balance platform responsibility against the protection of users.

Research Methods and Data Access

Research on social media increasingly treats platforms as sources of large-scale behavioural traces. Posts, reshares, reply threads, follower graphs and timestamps are analysed with graph-theory measures, text classifiers, topic models and machine-learning tools. This work, often called computational social science, depends on data that someone must first collect, through an API, a vendor tool, scraping, or a direct data-sharing arrangement. Industry uses similar big-data and data-mining methods, but outside researchers usually see only the slice of a platform that the operator chooses to expose.

Access to that slice has become less stable. Meta bought the monitoring tool CrowdTangle in 2016, and it was widely used by journalists and academics to track public Facebook and Instagram content. Meta confirmed in 2024 that CrowdTangle would close on August 14. Its successor, the Meta Content Library, is restricted to researchers at "qualified academic or nonprofit institutions", and applications are handled by an external consortium. Critics argued that the replacement was less capable, and a survey of independent researchers found that most expected the shutdown to hinder their work. Twitter followed a similar path. In 2023 the company dismantled its academic access program and set the price of its Enterprise API at $42,000 per month. A coalition of public-interest researchers reported more than 250 projects that would be jeopardised by the loss of free or low-cost access. The effect on scholarship has been to shift work toward smaller samples, toward other platforms, or toward groups that can afford commercial data.

Regulators have started to create legal routes to data. The EU's Digital Services Act includes Article 40, under which vetted researchers can request data from very large online platforms and search engines to study systemic risks in the EU. A delegated act on the procedures was adopted in 2025 and created a data access portal, but researchers have reported that significant barriers remain in practice. Public data under Article 40(12) falls outside that delegated act, so the rules for public and non-public data are handled separately.

Causal inference is the central methodological problem. Observational data can show that heavy users differ from light users, but not whether the platform caused the difference, because users select into platforms and content. Platforms run randomized experiments continuously in the form of A/B tests, and some firms have allowed outside scientists to run experiments on their own products. A 2020 collaboration between academic teams and Meta, for example, studied how feed settings affected users' political attitudes during the US election campaign, and another study paid users to deactivate Facebook for several weeks. Such designs are powerful, but the platform controls the randomisation, the outcome measures, and the timing, and can end or alter the study. Outside researchers therefore often rely on natural experiments: sudden algorithm changes, service outages, staggered national rollouts or regulatory bans that create comparison groups without researcher intervention. These designs need careful assumptions, such as parallel trends before the change, and they cannot easily isolate the effect of a single feature.

Reproducibility is hard when the underlying data cannot be shared. Platform terms often restrict redistribution of content, so replication usually requires collecting the data again, and the collection may no longer be possible if an API has closed or if posts have been deleted. Sampling is another problem. Some platform samples are not documented, so the population from which a sample was drawn is unclear. Published work on Twitter data has been difficult to compare because each team used different collection rules. Better documentation of data provenance, preregistration of analyses, and shared analysis code have been proposed as remedies, but they do not solve access restrictions.

Research on user populations raises ethical questions that differ from those of survey research. Many datasets include identifiable personal information, and aggregation does not always prevent re-identification. Informed consent is often absent when researchers analyse public posts, yet users may not expect their posts to be studied. Experiments are more sensitive still: a 2014 study that manipulated the emotional content of users' news feeds drew strong criticism over the absence of consent, and it helped prompt wider debate about institutional review and platform experimentation. These questions connect to broader concerns in ethics, data-privacy and reproducibility.

Open Questions and Future Directions

Decentralized and federated protocols are the most visible challenge to the centralized model. Mastodon operates as a federated network of independently managed servers that communicate using the ActivityPub protocol, and each instance sets its own moderation policies and content guidelines. Its user base has experienced multiple surges in adoption, most notably following Twitter's acquisition by Elon Musk in 2022. Bluesky takes a different route. It is built on the AT Protocol, an open protocol designed to give users control over their identity, data, feeds, and overall social media experience. Meta's Threads also uses ActivityPub, and Meta launched Threads with plans to support Mastodon in the future. Whether these designs can shift platform power is still open. The two protocol families remain largely separate, since AT Protocol and ActivityPub are separate ecosystems with no direct interoperability. User figures also vary by source: one guide reports that the fediverse has surpassed 15 million users, while another cites 35 million users for Bluesky, so such numbers should be treated with caution.

Regulation is another route toward interoperability, but its reach is uncertain. The EU's Digital Markets Act only included messaging in its final interoperability obligation, even though the European Commission had evaluated social networking as well. Advocates argue that interoperability requirements should be expanded to cover social networks, enabling entry by overcoming the single-homing tendencies of strong network effects. The Commission's first review of the Act took a more cautious line, concluding that it is too early to extend the interoperability obligations currently imposed on messaging services to online social networking services, because interoperability of messaging services has yet to see wide adoption. Critics of such mandates stress that interoperability carries high compliance costs and security and privacy risks, while proponents see it as a way to lower switching barriers. The debate therefore links to competition-law, market-power and network-effect, and to the broader question of how far economic-regulation should reach into social platforms.

Generative AI adds a different problem: whether users can tell authentic content from synthetic content. One review of 40 platforms found that more than two-thirds explicitly addressed AI-generated content in their governance policies. Labelling has been the main response. Meta's 2024 scheme initially used a single "Made with AI" tag that appeared on any content that had been touched by generative AI tools, even minor edits like background changes or filter applications. Evidence on whether labels work is mixed. One study found that labels such as "AI-generated" were perceived as useful for content made with AI, but they did not push users to consider whether the content was truthful or misleading. Technical provenance schemes are also being tried. Coalitions such as C2PA provide Content Credentials standards, and YouTube requires creators to disclose when realistic content is made with altered or synthetic media, including generative AI. Beyond authenticity, researchers note that introducing GenAI into online social spaces can create new concerns around privacy, authenticity, manipulation, bias, and the provenance of synthetic content. Uses such as AI summarizing or filtering feeds remain considerably less investigated by researchers and practitioners alike. These issues connect to artificial-intelligence, freedom-of-expression and journalism.

The long-term social and psychological effects of platform use remain largely unresolved. Much of the existing evidence is correlational, and short-term experiments can capture only part of what happens over years of use, as people's habits, relationships and attention change. Answering these questions requires longitudinal designs, data collected by platforms, and independent access to that data, which is itself contested (see the section on research methods). Decentralized networks may offer a partial testbed. Bluesky's researchers have invited the research community to use Bluesky as a dataset and testing ground for new approaches in social media moderation. Yet decentralization is not a cure: the same authors note that decentralization also introduces new challenges. Future governance will likely have to balance user autonomy, data-privacy, mental-health and public-interest goals, and the choice between causal-inference designs and platform-controlled evidence will shape what can be known about these effects.

References

1. User-generated contentebsco.com
2. History of Social Media Statsbroadbandsearch.net
3. Social Network Sites: Definition, History, and Scholarship - boyd - 2007 - Journal of Computer-Mediated Communication - Wiley Online Libraryonlinelibrary.wiley.com
4. User-Generated Content and Social Media - ScienceDirectsciencedirect.com
5. Social Network Sites: Public, Private, or What?researchgate.net
6. Social Network Sites: Definition & Historyscribd.com
7. Form Without Function: Agent Social Behavior in the Moltbook Networkarxiv.org
8. Then and now: a history of social networking sitescbsnews.com
9. The Unabridged History of Social Mediabuffer.com
10. SixDegrees.combritannica.com
11. SixDegrees.comen.wikipedia.org
12. The history of social media: A timeline of key momentsblog.hootsuite.com
13. Timeline: The evolution of online communitiescomputerworld.com
14. Timeline of LiveJournalen.wikipedia.org
15. LiveJournalen.wikipedia.org
16. History of bloggingen.wikipedia.org
17. From Bulletin Boards to Billionaires: Tracing Social Media‘s Winding Journey - History Toolshistorytools.org
18. Bebo, MySpace, Friendster: The First Social Media Sites Looked Like Thisinterhacktives.com
19. Facebooken.wikipedia.org
20. History of Facebook: Facts and Latest Developments - TheStreetthestreet.com
21. History of Facebook timelinelucensoftware.com
22. YouTubeen.wikipedia.org
23. The History of Social Media Since 2003blog.hubspot.com
24. X (formerly Twitter)britannica.com
25. Flickrebsco.com
26. 20 Years of Significant Moments in Flickr’s Developmentblog.flickr.net
27. What Ever Happened to Flickr?techspot.com
28. Instagram Marketing History: How A Photo-Sharing App Transformed Social Media Marketing, Influencer Culture And Social Commerce - Marketing Museummarketing.museum
29. Instagram vs Flickr in 2025 - Through the Lens Blog - Marco Secchimarcosecchi.com
30. History of Flickr: from rise to fall - nss magazinenssmag.com
31. THE TIKTOK PHENOMENON AND THE RISE OF SHORT FORM VIDEOprime-journal.com
32. The Evolution of Douyin (Chinese TikTok) for Content Creatorstianranli.medium.com
33. The birth of Douyin, TikTok's elder sister - by Zichen Wangpekingnology.com
34. TikToken.wikipedia.org
35. Douyin SEO: Ranking Videos On The Chinese TikTok - Marketing Chinamarketingtochina.com
36. Social Media Marketing. A Brief Historyopen.library.okstate.edu
37. TAO: Facebook’s Distributed Data Store for the Social Graphpaperswelove.org
38. TAO: Facebook’s Distributed Data Store for the Social Graphblog.acolyer.org
39. TAO: Facebook's Distributed Data Store for the Social Graphresearchgate.net
40. Design Twitter System Design: A Complete Guide 2026systemdesignhandbook.com
41. System Design: Twitter/X News Feed — Timeline Generation, Fanout-on-Write vs Read, Celebrity Problem, Rankingtechinterview.org
42. Twitter’s new data fees leave scientists scrambling for fundingtheconversation.com
43. [2404.07340] RIP Twitter API: A eulogy to its vast research contributionsarxiv.org
44. Twitter announces new API with only free, basic and enterprise levelstechcrunch.com
45. Twitter introduces a new $5,000-per-month API tiertechcrunch.com
46. Codemiacodemia.io
47. Insights from Paper-TAO: Facebook’s Distributed Data Store for the Social Graphhemantkgupta.medium.com
48. Bringing People Closer Together - About Meta - Facebookabout.fb.com
49. Analyzing News Engagement on Facebook: Tracking Ideological Segregation and News Quality in the Facebook URL Datasetarxiv.org
50. Deep neural networks for YouTube recommendationsblog.acolyer.org
51. [PDF] Deep Neural Networks for YouTube Recommendationssemanticscholar.org
52. Here's What the Facebook Papers Say About the Ranking Algorithms That Control Your News Feedgizmodo.com
53. Facebook's algorithm change in 2018 benefitted Republican groups, researchers sayengadget.com
54. Demographic Divides in Political Content Exposure on Facebookarxiv.org
55. Facebook defends 2018 algorithm changes, blames 'partisan divisions' on historyfoxbusiness.com
56. Action Recommended: How the Digital Services Act Addresses Platform Recommender Systems - Mozilla Foundationmozillafoundation.org
57. A Right to Constructive Optimization: A Public Interest Approach to Recommender Systems in the Digital Services Act - PMCpmc.ncbi.nlm.nih.gov
58. Deep Neural Networks for YouTube Recommendationsdl.acm.org
59. Using Deep Neural Networks to make YouTube Recommendationsmedium.com
60. 2023 Facebook Algorithm Guide: Overview & Best Practicestinuiti.com
61. Meta Platforms, Inc. - Form 10-K - FY2025sec.gov
62. Meta - Meta Reports Fourth Quarter and Full Year 2025 Resultsinvestor.atmeta.com
63. What is the attention economy?ubs.com
64. Attention can’t be bought, so brands turn to memestheconversation.com
65. Reddit stock drops 9% as Google AI content deal nears expirationqz.com
66. Reddit's data licensing revenue grows 24% to $43M as OpenAI and Google lead the buyer listcryptobriefing.com
67. Reddit Stock Drops 6% as Google AI Licensing Deal Renewal Talks Hit Turbulencekucoin.com
68. Reddit Demands Better AI Deal: More Money, Users from ...techbuzz.ai
69. Meta Ads Statistics 2026: 60+ Facebook Ad Stats With Sourcesadsuploader.com
70. FOR RELEASE JANUARY 31, 2024 BY Jeffrey Gottfriedpewresearch.org
71. Urban Social Media Inequality: Definition, Measurements, and Applicationarxiv.org
72. Social Media Statistics 2026: Global Users • SQ Magazinesqmagazine.co.uk
73. Global Social Media Statistics — DataReportaldatareportal.com
74. 1 PEW RESEARCH CENTER www.pewresearch.org FOR RELEASE NOVEMBER 20, 2025pewresearch.org
75. Pew's 2025 Social Media Report Shows Endurance of YouTube, Facebook - JCKjckonline.com
76. ITU: 6 Billion Now Online Globallytvtechnology.com
77. Progress and Gaps: Key Findings from ITU’s Facts and Figures 2025social.desa.un.org
78. Global internet use reaches 6 Billion, but ITU warns connectivity gaps are deepening - CADEcadeproject.org
79. ITU Measuring Digital Development: Facts and Figures 2025 —...libertify.com
80. ITU Statistics Update - December 2025itu.int
81. Global Internet usageen.wikipedia.org
82. Digital 2026: Global Overview Report — DataReportaldatareportal.com
83. Latest social media statistics and facts 2026wix.com
84. Meta wins FTC antitrust trial over Instagram and WhatsApp deals - The Japan Timesjapantimes.co.jp
85. Meta wins landmark antitrust case over Instagram, WhatsApp acquisitionsaxios.com
86. Meta Prevails in FTC’s Antitrust Case Challenging Instagram and WhatsApp Acquisitionssullcrom.com
87. Federal District Court Rejects the FTC’s Monopolization Case Against Meta Platformscongress.gov
88. Meta wins FTC antitrust suit, will keep Instagram and WhatsAppfinance.yahoo.com
89. Alphabet Inc. - Form 10-Q - FY2026sec.gov
90. Google Search Antitrust Remedies Take Effect as Appeals Court Battle Beginslitigationlogic.io
91. Google's Search Antitrust Remedies Are Now in Effectsigmalawgroup.com
92. US v. Google Antitrust Case: Ruling & Remedies Explainedthelawlion.com
93. Compliance principles for the Digital Markets Actbruegel.org
94. Digital Markets Act - Overview - Lexologylexology.com
95. Digital Markets Acthsfkramer.com
96. Digital Markets Acttechpolicy.press
97. About the Digital Markets Act - European Uniondigital-markets-act.ec.europa.eu
98. Meta’s Oversight Board races to govern the AI surge - Rest of Worldrestofworld.org
99. Meta’s Oversight Board: A Review and Critical Assessmentlink.springer.com
100. Oversight Board Announces New Cases and Review of Meta’s COVID-19 Misinformation Policiesoversightboard.com
101. Oversight Board Case of Hateful Memes Video Montageglobalfreedomofexpression.columbia.edu
102. Tracking the First Digital Services Act Transparency Reportstechpolicy.press
103. How the Digital Services Act enhances transparency onlinedigital-strategy.ec.europa.eu
104. Article 42, the Digital Services Act (DSA)eu-digital-services-act.com
105. The DSA’s transparency reportshiig.de
106. The Santa Clara Principles on Transparency and Accountability in Content Moderationsantaclaraprinciples.org
107. Oversight Board - Transparency Center - Metatransparency.meta.com
108. Decisions & Policy Advisory Opinionsoversightboard.com
109. The Santa Clara Principles on Transparency and Accountability in Content Moderationbenton.org
110. An Educational Tool for Learning about Social Media Tracking, Profiling, and Recommendationarxiv.org
111. How Social Media Platforms Gather and Use Location Informationpeople.utm.my
112. Social media apps may know your exact locationsurfshark.com
113. Dual Protection Ring: User Profiling Via Differential Privacy and Service Dissemination Through Private Information Retrievalarxiv.org
114. cambridge analytica facebook data scandalglobalnews.ca
115. Cambridge Analyticadig.watch
116. “It wouldn't happen to me”: Privacy concerns and perspectives following the Cambridge Analytica scandal - ScienceDirectsciencedirect.com
117. The Cambridge Analytica and Facebook data scandal: how to tell if your data was sharedtechradar.com
118. globalnews.caglobalnews.ca
119. Record-breaking GDPR fine: Meta faces €1.2 billion penalty for data transfers to the USharneys.com
120. meta fined record 13 billion eu user data transfers us rcna85515nbcnews.com
121. Meta's GDPR Fines: EUR 2.83 Billion Running Total in 2026privacyterms.io
122. Facebook Cambridge Scandal Data Breach: What Happened, Impact, and Lessonshuntress.com
123. (PDF) Social Media and Social Mobilization in the Middle East: A Survey of Research on the Arab Springresearchgate.net
124. Social media's role in the Arab Springen.wikipedia.org
125. THE ROLE OF SOCIAL MEDIA IN THE ARAB SPRINGsites.stedwards.edu
126. Social media and political protest mobilization: unveiling the dynamics of the Arab Spring 2.0link.springer.com
127. Online social cohesion reflects real-world group action in Syria during the Arab Springncbi.nlm.nih.gov
128. Calling for justice with #JusticeforBreonnaTaylor: a case study of hashtag activism in the evolution of the black lives matter movementncbi.nlm.nih.gov
129. An Examination of Social Activism and the Role of Social ...research.library.fordham.edu
130. Lowest common denominator messaging: issue prioritization in #BlackLivesMatter and #MeToo digital coalitionslink.springer.com
131. When repression meets resistance: internet shutdowns in 2025 - Access Nowaccessnow.org
132. 2025 was the worst year on record for internet shutdowns as censors move to more targeted blockstechradar.com
133. The Reuters Institute Digital News Report 2026edelman.com
134. Reuters Institute Digital News Report 2026benton.org
135. ‘Unsettled Time’: 2026 Reuters Institute Digital News Report Finds Rise of Social Media, Video Creators, and AI Chatbots Amid Falling News Interest and Trustgijn.org
136. Report Finds Facebook and TikTok Are Key News Sourcessocialmediatoday.com
137. 1 in 5 Americans regularly get news on TikTok, up sharply from 2020pewresearch.org
138. Americans tune out news as social media becomes main source of information, Pew findsjpost.com
139. Social Media & the News - Research and data from Pew Research Centerpewresearch.org
140. Reuters Institute digital news report 2026 - ORA - Oxford University Research Archiveora.ox.ac.uk
141. FALSE NEWS IS BIG NEWS. Barely a day goes by without a new development aboutide.mit.edu
142. [PDF] The spread of true and false news onlinesemanticscholar.org
143. On Twitter, false news travels faster than true stories, study findssciencedaily.com
144. DIGITAL NEWS PROJECT JANUARY 2026 Journalism and Technologyreutersinstitute.politics.ox.ac.uk
145. Reuters Institute releases 2026 Digital News Report - NCSnewscaststudio.com
146. Recommendation with Generative Modelsarxiv.org
147. Echo Chambers, Filter Bubbles, and Polarisation: a Literature Reviewroyalsociety.org
148. Echo chambers, filter bubbles, and polarisation: a literature review - ORA - Oxford University Research Archiveora.ox.ac.uk
149. Understanding Echo Chambers and Filter Bubbles: The Impact of Social Media on Diversification and Partisan Shifts in News Consumptionresearchgate.net
150. Social Media, Misinformation and Harmful Algorithmscommittees.parliament.uk
151. (PDF) How Many People Live in Political Bubbles on Social Media? Evidence From Linked Survey and Twitter Dataresearchgate.net
152. First Four Papers from US 2020 Facebook & Instagram Research Election Study Published in Science and Naturecsmapnyu.org
153. A study found Facebook’s algorithm didn’t promote political polarization. Critics have doubtsscience.org
154. How do social media feed algorithms affect attitudes and behavior in an election campaign?researchgate.net
155. social media and polarizationspia.princeton.edu
156. Frontiersfrontiersin.org
157. Echo Chambers and Filter Bubbles Don’t Reflect Our Media Environmentmediatechdemocracy.com
158. Opinion dynamics: Statistical physics and beyondarxiv.org
159. Like-minded sources on Facebook are prevalent but not polarizingnature.com
160. The Impact of Social Media on Adolescent Mental Health: A Meta-Analysisresearchgate.net
161. Social media use and its impact on adolescent mental health: An umbrella review of the evidence - ScienceDirectsciencedirect.com
162. A meta-analysis of the association between adolescent social media use and depressive symptoms - ScienceDirectsciencedirect.com
163. Social media has limited effects on teenage life satisfactionsciencedaily.com
164. (PDF) Social media’s enduring effect on adolescent life satisfactionresearchgate.net
165. Specification curve analysis shows that social media use is linked to poor mental health, especially among girls - ScienceDirectsciencedirect.com
166. Underestimating digital media harmnature.com
167. Digital Media Use and Adolescents' Mental Health During the Covid-19 Pandemic: A Systematic Review and Meta-Analysisncbi.nlm.nih.gov
168. Facebook Use Predicts Declines in Subjective Well-Being in Young Adultsjournals.plos.org
169. Passive Facebook Usage Undermines Affective Well-Being:sites.lsa.umich.edu
170. Social media use and adolescent health-risk behaviours: A systematic review and meta-analysismedrxiv.org
171. Memes in Digital Culture by Limor Shifmangoodreads.com
172. Memes in Digital Culture - Limor Shifman - Google Booksbooks.google.com
173. Book Review: Memes in Digital Culture by Limor Shifmanblogs.lse.ac.uk
174. www.cmu.educmu.edu
175. The Inventor Of The Twitter Hashtag Explains Why He Didn't Patent Itfinance.yahoo.com
176. Chris Messina, hashtag inventorcnbc.com
177. Hashtag History: From Chris Messina's Tweet to Word of Yearmeme.com
178. Chris Messina (inventor)en.wikipedia.org
179. Parasocial interaction - Xi'an Jiaotong-Liverpool Universityscholar.xjtlu.edu.cn
180. Parasocial Relationshipsencyclopedia.pub
181. Parasocial Interactionsebsco.com
182. The Surprising History of Twitter's Hashtag Originbuffer.com
183. Parasocial Interaction - AURA Lab - SIUEaura-lab.siue.edu
184. How to Properly Disclose FTC Endorsements: A Comprehensive Guide for E-commerce Sites and Influencers - PartnerCentricpartnercentric.com
185. FTC Endorsement Guides — Influencer Disclosure Requirements for Food and Supplement Brandstrytruli.com
186. FTC Influencer Guidelines: Brands and Creators Must Be Aware Before Postingtraverselegal.com
187. A Year Later—Revisiting FTC’s Updated Endorsement Guideshallrender.com
188. Creator Economy Payments: How Platforms Pay Creators and Why It's Brokenspark.money
189. Making a Living in the Creator Economy: A Large-Scale Study of Linking on YouTube - Bernhard Rieder, Erik Borra, Òscar Coromina, Ariadna Matamoros-Fernández, 2023journals.sagepub.com
190. Creator Economy Earnings by Platform 2026: What Creators Actually Get Paid - Axis Intelligenceaxis-intelligence.com
191. How the digital creator economy reshapes power, profit, and culture today — Networth Areahub.stellantis.com
192. Content Moderators Can Sue Meta In Kenya—Court Dismisses Tech Giant's Appealpeopleofcolorintech.com
193. AI’s Dark Secrets: Inside the Global Exploitation of Data Workersthecairoreview.com
194. Newsletter Newsletters Events Events Podcasts Videos Africanewseuronews.com
195. Dependence and Precarity in the Platform Economybc.edu
196. Dependence and precarity in the platform economy - PMCpmc.ncbi.nlm.nih.gov
197. Facebook Content Moderators Sue Meta in Kenyatime.com
198. Social Media Might Have To Rethink Platform Design And Features As Courts Reject Communications Decency Act, Section 230 Defensebeneschlaw.com
199. Massachusetts Supreme Judicial Court Recognizes Section 230 Is No Bar to Social Media Design Claimsepic.org
200. A US court just cleared thousands of social media harm lawsuits to proceedthenextweb.com
201. Section 230eff.org
202. As social media lawsuits move forward, is Section 230 facing an existential threat? - Tubefiltertubefilter.com
203. Section 230 and Platform Liability in 2026: What Operators Need to Knowtoslawyer.com
204. European Commission finds Meta in breach of the Digital Markets Act - Society for Computers & Lawscl.org
205. Commission finds Apple and Meta in breach of the Digital Markets Actdigital-strategy.ec.europa.eu
206. Guarding the Gates: The Digital Markets Act and Lessons in Ex Ante Regulationcsis.org
207. Digital Markets Act Impact 2026: Gatekeepers, Fines, and Future of EU Tech Regulation - News and Statistics - IndexBoxindexbox.io
208. EU Court Backs Meta Challenge Over Marketplace Gatekeeper Designation Under Digital Markets Act - News and Statistics - IndexBoxindexbox.io
209. Australia to strengthen enforcement of under-16 social media banfortune.com
210. Australia moves to strengthen under-16 social media ban amid evidence of enforcement concerns - JURIST - Newsjurist.org
211. Australia proposes increasing fines for social media companies in under-16 bancbc.ca
212. Australia is trying to enforce the first teen social media ban. Governments worldwide are watching.cnbc.com
213. Australia’s under-16 social media ban failing, study shows: What it meansaljazeera.com
214. Meta Announces Shutdown of Its CrowdTangle Monitoring Appsocialmediatoday.com
215. Meta shutters tool used to fight disinformation, despite outcry : NPRnpr.org
216. DSA Art 40 Delegated Acteosc.eu
217. Research Opportunities and Challenges of the EU’s Digital Services Actarxiv.org
218. Researchers Consider the Impact of Meta's CrowdTangle Shutdowntechpolicy.press
219. Meta shuts down data tool widely used by journalistsaxios.com
220. Letter: Twitter’s New API Plans Will Devastate Public Interest Research - Coalition for Independent Technology Researchindependenttechresearch.org
221. Non-Public Data Access for Researchers: Challenges in the Draft Delegated Act of the Digital Services Acttechpolicy.press
222. Who gets to see inside? The EU’s Operational Rules on Data Access under Article 40 of the DSAhlc.com
223. Mastodon (social network)en.wikipedia.org
224. Bluesky vs Mastodon: How Are They Different? Key Differencessocialbee.com
225. Platform Alternatives: Mastodon, Bluesky, and the Fediversestateofsurveillance.org
226. Decentralized Social Media Finds A Footholdgfmag.com
227. The Fediverse and Bluesky: A Comparison of Decentralized Social Media - LostInMatrixlostinmatrix.com
228. It’s time for the European Union to rethink personal social networkingbruegel.org
229. What the EU's First Digital Markets Act Review Actually Changestechpolicy.press
230. "Okay, I've Actually Softened My Take on This": How People in Decentralized Social Media Reason about the Appropriateness of Generative AIarxiv.org
231. Authenticity in the Age of AIcmr.berkeley.edu
232. Human or AI? Evaluating Labels on AI-Generated Social Media Content - The Daisdais.ca
233. Labeling Synthetic Content: User Perceptions of Label Designs for AI-Generated Content on Social Mediadl.acm.org
234. Governance of AI-Generated Content: A Case Study on Social Media Platformsarxiv.org
235. Bluesky and the AT Protocol: Usable Decentralized Social Mediaarxiv.org
236. The Open Social Web in 2025: Mastodon, Threads, Bluesky & the Rise of the Fediverse - Best Solutionbestsoln.com
237. Reclaiming the Algorithm: A call for social media interoperability - DSA Observatorydsa-observatory.eu
238. FR#162connectedplaces.online