# Social Media

> Source: https://aiwiki.ai/wiki/social_media
> Updated: 2026-07-30
> Fact-checked: 2026-07-30
> Categories: AI Tools & Products, Machine Learning
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Social Media." aiwiki.ai, 30 Jul 2026. https://aiwiki.ai/wiki/social_media
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Social media** comprises internet-based services through which people and organizations create, distribute, discover, and respond to content in networked publics. The category includes social-network services, media-sharing platforms, discussion forums, live-streaming services, and parts of messaging applications. Its boundaries are disputed: a service may combine public posting, private communication, commerce, entertainment, and search, while users and researchers may apply the term to different subsets of those functions.[1][2]

Modern social media is a socio-technical system rather than a single kind of website. Its visible interfaces sit above identity and relationship graphs, storage and delivery infrastructure, recommendation and advertising systems, moderation processes, and governance rules. [Artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) is used in several of those layers, but it does not define social media and is not the sole cause of what users see. People choose whom to follow and what to publish; platforms set product rules and ranking objectives; advertisers and creators respond to incentives; and automated models estimate, classify, generate, or order content.

## Definition and scope

One influential definition of a social-network site emphasizes three capabilities: users construct profiles, articulate connections, and view or traverse those connections. Under that definition, SixDegrees.com, launched in 1997, was the first recognizable social-network site because it combined profiles, lists of friends, and traversal of those lists.[1] Social media is broader. Obar and Wildman describe common features rather than a rigid boundary: Web 2.0 applications, user-generated content, user or group profiles, and connections among those profiles.[2]

The distinction matters because not all social media is organized around reciprocal friendship. Video feeds may be centered on creators and inferred interests; forums may be organized by topic; messaging products may contain public channels; and federated networks may distribute posts among independently operated servers. A platform can also change category as it adds feeds, profiles, groups, shopping, or private messaging.

Common forms include:

| Form | Organizing unit | Typical interaction |
|---|---|---|
| Social-network service | Profile and connection graph | Following, friending, posting, commenting |
| Media-sharing service | Video, image, audio, or live stream | Viewing, reacting, remixing, subscribing |
| Forum or community | Topic, group, or thread | Posting, replying, voting, moderating |
| Creator or publishing network | Account and audience | Publishing, subscribing, sharing |
| Social messaging | Conversation, channel, or group | Direct messages, group chat, public broadcasts |
| Federated social network | Account on an independently operated server | Following and exchanging activities across servers |

These categories overlap. They describe product organization, not mutually exclusive legal or technical classes.

## Historical development

Online social interaction predates the term *social media*. Bulletin-board systems, Usenet, chat rooms, online forums, blogs, and personal home pages supplied many of the practices and technical features that later platforms combined. Social-network sites added a particularly visible representation of identity and interpersonal ties. In the late 1990s and early 2000s, services including SixDegrees, LiveJournal, Friendster, MySpace, and later Facebook made profiles and connection graphs central to the user experience.[1]

During the 2000s, broadband access, camera phones, and Web 2.0 publishing tools lowered the cost of posting and sharing media. Feeds gradually displaced visits to individual profile pages as a primary way to encounter content. Mobile applications, push notifications, live video, and short-form video then made social media a persistent rather than occasional channel. This history was not a single progression: forums, blogs, messaging, and federated communities continued alongside large centralized platforms.

Most commercial platforms centralize account management, storage, ranking, and policy enforcement. Federation offers a different architecture. The World Wide Web Consortium's ActivityPub Recommendation, published in 2018, defines both client-to-server and server-to-server protocols for creating, updating, delivering, and following activities across compatible services.[3] Federation can distribute operational control, but it does not eliminate moderation, privacy, security, or interoperability problems; it reallocates some of those decisions among server operators and users.

## Feeds and recommender systems

A chronological feed orders eligible items mainly by time. A ranked feed uses a [recommender system](https://aiwiki.ai/wiki/recommender_system) to select and order a subset. Eligibility can depend on explicit connections, subscriptions, geography, language, age settings, or policy filters. Ranking then estimates outcomes such as the probability of a view, watch duration, reaction, comment, share, follow, hide, or report. Platforms may combine multiple predictions and apply diversity, freshness, integrity, and business constraints.

Large recommenders often separate retrieval from ranking. A 2016 YouTube paper described one neural network that selected hundreds of candidates from a much larger corpus and another that ranked those candidates by expected watch time.[4] The architecture is historically important, but it is a published snapshot of one system, not a complete description of YouTube's current production stack or a universal design.

Platform disclosures illustrate how signals can differ. TikTok has said that its For You system considers user interactions, video information, and device or account settings; stronger signals, such as completing a longer video, can receive more weight than weaker contextual signals.[5] Meta has described multistage ranking systems that gather candidates, predict several forms of user response, combine those predictions, and apply additional integrity and diversity rules.[6] These are platform-authored overviews and do not expose all training data, model weights, experiments, or policy interventions.

Academic and industry papers add engineering detail without providing a permanent map of any service. Meta's Deep Learning Recommendation Model combines sparse categorical embeddings with dense features and an interaction layer.[7] ByteDance's Monolith paper describes collisionless embedding tables and an online-training architecture intended to respond to changing behavior.[8] Such publications show methods that can be used at scale; they should not be read as proof that a named model remains deployed unchanged.

Ranking is neither a neutral mirror of user preference nor an autonomous actor. The candidate pool and training labels are products of previous user behavior and platform rules. An objective such as watch time privileges measurable responses; a penalty for reports or repeated content changes the ordering; and interface design affects the behavior later used as training data. Feedback loops can therefore emerge among exposure, user action, creator strategy, and subsequent model updates.

### A simplified ranking pipeline

| Stage | Function | Examples of inputs | Important limitation |
|---|---|---|---|
| Eligibility | Exclude unavailable or ineligible items | Connection rules, age or region settings, blocks, policy status | Rules vary by service and account |
| Candidate retrieval | Find a manageable set from a large corpus | Follow graph, item embeddings, search or topic signals | Relevant items can be missed before ranking |
| Prediction | Estimate possible responses | Viewing history, item features, context | A prediction is not a direct measure of welfare or truth |
| Ranking and re-ranking | Combine estimates and constraints | Expected watch time, reactions, freshness, diversity | The objective and weights are platform choices |
| Delivery and feedback | Display items and record responses | Impression, dwell time, share, hide, report | Interface and prior exposure shape the feedback |

## Data, advertising, and measurement

Social-media services can collect information supplied by users, records of interactions, device and network data, inferred interests, and information from advertisers or other partners. Data practices differ by service, jurisdiction, account state, and product setting. Advertising is a major revenue source for many large platforms, but subscriptions, commerce fees, virtual goods, licensing, and enterprise services also exist; no single business model covers the entire category.

In 2024, the United States Federal Trade Commission published a staff report based on compulsory orders issued in 2020 to nine social-media and video-streaming companies. The report described extensive collection and retention of user and non-user data, limited data-minimization practices, and incentives linking surveillance to targeted advertising.[9] The findings concern the companies and response period examined. They are not an audit of every social-media service, and some commissioners disputed portions of the staff's policy analysis.

Ad targeting and ad delivery are separate stages. An advertiser may define an audience, while a platform's delivery system decides which eligible people receive impressions. A 2019 field study of Facebook advertising found skew by gender and race proxies in the delivery of employment and housing ads even when advertisers used neutral targeting parameters.[10] That result demonstrates a mechanism by which optimization can alter an audience; it does not establish that every campaign, platform, or later system behaves identically.

Aggregate metrics also need context. Monthly active users, views, impressions, and watch time are platform-defined measures. A change in counting rules, bot removal, product bundling, or geographic coverage can break comparisons. Surveys measure reported behavior in a sampled population rather than server logs. For example, Pew Research Center's United States fact sheet describes survey estimates for specified field periods and populations, not global platform totals.[11]

## Content moderation and integrity

Content moderation is the application of service rules to accounts, posts, messages, advertising, and other activity. It can include pre-publication restrictions, post-publication removal, reduced distribution, warning labels, age gates, account penalties, appeals, and referrals required by law. Decisions are made through combinations of user reports, automated systems, trusted flaggers, human reviewers, and rules for known material.

Automation is useful where the volume is high or matching is well defined. Hash matching can identify copies of known files; classifiers can prioritize likely spam, harassment, nudity, or violence; and language or vision models can provide scores for review queues. Meta, for example, has described a process in which matching systems and [machine-learning](https://aiwiki.ai/wiki/machine_learning) classifiers can detect content before or after reports, with human review used for some decisions and appeals.[12] That description is a company account of its own process, not independent evidence that all errors are found or that enforcement is consistent.

Moderation models face ambiguous language, cultural variation, adversarial adaptation, and unequal error costs. A false positive can suppress lawful expression; a false negative can leave harmful material available. Labels used for training may reflect policy judgments that change over time. Accuracy measured on a benchmark can also conceal different performance across languages, dialects, image contexts, or newly emerging behavior.

Automated-account detection has similar limits. Botometer combines multiple account and network features to produce scores that help researchers study coordinated or automated behavior. Its maintainers caution that a score is not necessarily a calibrated probability that an account is a bot, and that model output should be interpreted with domain, sampling, and platform-access limitations in mind.[13] Automation itself is not always abusive: scheduling tools, news alerts, and service accounts can be legitimate, while coordinated manipulation can involve people as well as software.

## Generative AI and synthetic media

From 2023, several platforms integrated chat assistants and content-generation tools. Snap introduced the experimental My AI chatbot in February 2023 and explicitly warned that it could give biased, incorrect, harmful, or misleading answers.[14] Meta announced conversational assistants and image tools across its applications in September 2023.[15] These launches illustrate a shift from using AI mainly behind the interface to offering [generative AI](https://aiwiki.ai/wiki/generative_ai) directly to users. They do not imply that every platform adopted the same features or safeguards.

Generated text, images, audio, and video can lower production costs for education, accessibility, translation, creative work, and advertising. The same tools can produce impersonation, fraud, spam, or non-consensual intimate imagery. A [deepfake](https://aiwiki.ai/wiki/deepfake) is one subset of synthetic media, usually involving a convincing representation of a real person's appearance, voice, or actions. Not every edited or AI-assisted work is a deepfake, and harm depends on content, context, consent, and distribution.

Platforms have adopted different disclosure policies. YouTube introduced creator disclosure for realistic altered or synthetic content in March 2024, with more prominent labels for certain sensitive topics.[16] Meta's 2024 approach used detected industry metadata and user disclosure, later changing the visible wording from “Made with AI” to “AI info” after the broad label captured some lightly edited photographs.[17] TikTok announced automatic labeling of some externally generated content carrying Content Credentials in May 2024.[18] These policies are neither identical nor comprehensive; metadata may be absent, removed, unsupported, or incorrectly applied.

[C2PA](https://aiwiki.ai/wiki/c2pa) Content Credentials provide a technical framework for signed assertions about an asset's origin and editing history. A valid credential can support attribution of a claim to its signer and reveal whether bound content changed after signing.[19] It does not by itself prove that a depicted event occurred, that every edit was disclosed, or that an unsigned file is synthetic. Provenance, forensic detection, contextual reporting, and platform policy address different questions.

## Social and political effects

Social media can connect geographically dispersed communities, support interpersonal relationships, distribute emergency information, and enable participation by people who face barriers in offline institutions. It can also enable harassment, surveillance, deceptive persuasion, and rapid distribution of false or low-quality material. Effects depend on the service, population, content, task, measurement window, and comparison condition.

“Echo chamber” is used for environments where selective exposure and social reinforcement concentrate similar views. A 2021 comparative study found homophilic clustering and selective exposure on several platforms, with important differences among Facebook, Reddit, Twitter, and Gab.[20] It was an observational analysis of specified datasets. It did not establish that all users occupy sealed information environments or that recommendation alone caused the measured clustering.

Randomized interventions complicate simple claims that ranked feeds uniformly polarize users. In a study conducted around the 2020 United States election, reducing exposure to Facebook posts from like-minded sources changed what participants saw but produced no detectable change in several measured political attitudes.[21] A companion experiment that removed reshared posts substantially changed exposure and on-platform engagement, with no detectable effect on the preregistered political attitudes.[22] Null estimates do not prove that feeds never affect attitudes; they bound effects under those interventions, users, measures, and time period.

A seven-week randomized experiment on X conducted in 2023 and published in 2026 found that enabling its algorithmic feed shifted some issue attitudes in a more conservative direction, while the study reported no significant effects on partisanship or affective polarization.[23] Taken together, these experiments argue against a universal platform effect. A system may alter exposure and behavior without changing a survey outcome, or may change particular attitudes in a particular political and product context.

Accuracy prompts are another narrow intervention. Experiments by Pennycook and colleagues found that asking people to consider accuracy improved the quality of news they said they would share.[24] The intervention concerns attention at the point of judgment. It is not a general solution for misinformation, which also involves source incentives, coordinated distribution, identity, trust, and platform enforcement.

## Adolescents and health

Social media can provide adolescents with friendship, creative expression, identity exploration, peer support, and access to information. It can also expose them to bullying, unwanted contact, sleep disruption, social comparison, or harmful content. Age, developmental stage, existing health, family environment, the nature of use, and the design of the service can change the balance. The American Psychological Association therefore advises against treating social-media use as inherently beneficial or harmful.[25]

The 2023 United States Surgeon General's advisory concluded that available evidence was insufficient to determine that social media was adequately safe for children and adolescents. It cited an association between more than three hours of daily use and roughly twice the risk of poor mental-health outcomes in one longitudinal cohort.[26] That figure is an observed association, not a universal causal threshold or a diagnosis for an individual.

The National Academies' 2024 consensus report reached a deliberately limited population-level conclusion: its review did not support saying that social media causes changes in adolescent health at the population level, and it called for better access to platform data and stronger longitudinal and experimental research.[27] A 2024 systematic review and meta-analysis covering 143 studies found small positive associations between social-media use and internalizing symptoms in community samples, alongside substantial heterogeneity and limited evidence from clinical populations.[28] These findings can coexist with serious harm to some individuals and benefit to others; averages do not identify who will be affected or which product features matter.

Useful research separates total time from specific experiences. Active communication with supportive peers is different from repeated exposure to self-harm content; nighttime notifications are different from daytime creative work; a measured association between heavy use and depression can reflect effects in either direction or shared causes. Product-specific, age-sensitive, and clinically meaningful outcomes are more informative than claims that social media as a whole either causes or cures a condition.

## Governance and regulation

Governance operates at several levels: service terms, app-store policies, technical standards, civil and criminal law, sector regulation, and independent oversight. Obligations vary by jurisdiction and by the size or function of a service. The following examples describe the position at the research cutoff of 28 July 2026; they are not a complete statement of any person's legal rights.

In the European Union, the Digital Services Act requires online platforms to explain the main parameters of recommender systems. Very large online platforms must offer at least one recommender option not based on profiling, assess systemic risks, and provide qualifying researchers with data under specified conditions.[29] The EU Artificial Intelligence Act was enacted in 2024. Its Article 50 transparency duties for certain synthetic content and deepfakes were scheduled to become applicable on 2 August 2026, after this article's cutoff; they should not be described as already applicable on 28 July.[30]

The United Kingdom's Online Safety Act duties entered into force in stages. Ofcom stated that illegal-content duties began applying on 17 March 2025 and that child-safety duties began applying on 25 July 2025 for services likely to be accessed by children.[31] In Australia, age-restricted social-media platforms have been required since 10 December 2025 to take reasonable steps to prevent people under 16 from holding accounts; the legal duty falls on platforms rather than children or parents.[32]

The United States has a mixture of federal and state rules rather than a general social-media statute. The federal TAKE IT DOWN Act, enacted on 19 May 2025, criminalizes specified publication of non-consensual intimate depictions and requires covered platforms to operate a notice-and-removal process, including removal within 48 hours after a valid request.[33] The Federal Trade Commission's 2025 amendments to the Children's Online Privacy Protection Rule added separate parental consent for disclosures connected to targeted advertising and strengthened retention limits; most amendments had a one-year compliance period after publication.[34] These laws address defined conduct and data practices, not every disputed claim or harmful recommendation.

Regulatory transparency is not the same as public reproducibility. A platform can describe ranking parameters without releasing personal data or proprietary models. Researchers may need secure access, privacy protections, documentation of product changes, and the ability to publish adverse findings. Conversely, unrestricted data release can expose users or enable abuse. Effective oversight has to manage both accountability and data protection.

## Studying social media

Research on social media is unusually sensitive to time. Interfaces, user populations, policies, APIs, and ranking models can change during a study or between data collection and publication. A result from one service in one country and election cannot automatically be transferred to another. Platform-authored engineering reports provide valuable detail but may omit negative results; leaked datasets may be incomplete; surveys are affected by recall; and observational traces rarely reveal why an item was shown.

Strong claims identify the unit of analysis, period, population, and comparison. They distinguish exposure from engagement, association from causation, and a company's stated policy from observed enforcement. They also preserve negative and null findings. Because social media joins technical systems to social institutions, no single metric—accuracy, watch time, removals, prevalence, or self-reported well-being—captures its overall effects.

## See also

- [Artificial Intelligence](https://aiwiki.ai/wiki/artificial_intelligence)
- [Recommender System](https://aiwiki.ai/wiki/recommender_system)
- [Machine Learning](https://aiwiki.ai/wiki/machine_learning)
- [Deep Learning](https://aiwiki.ai/wiki/deep_learning)
- [Generative AI](https://aiwiki.ai/wiki/generative_ai)
- [AI-Generated Content](https://aiwiki.ai/wiki/ai_generated_content)
- [Deepfake](https://aiwiki.ai/wiki/deepfake)
- [C2PA](https://aiwiki.ai/wiki/c2pa)
- [Meta AI](https://aiwiki.ai/wiki/meta_ai)
- [Social Media ChatGPT Plugins](https://aiwiki.ai/wiki/social_media_chatgpt_plugins)
- [TikTok](https://aiwiki.ai/wiki/tiktok)
- [YouTube](https://aiwiki.ai/wiki/youtube)
- [ByteDance](https://aiwiki.ai/wiki/bytedance)
- [Large Language Model](https://aiwiki.ai/wiki/large_language_model)

## References

1. boyd, danah m., and Nicole B. Ellison. “Social Network Sites: Definition, History, and Scholarship.” Journal of Computer-Mediated Communication 13, no. 1 (2007): 210–230. https://doi.org/10.1111/j.1083-6101.2007.00393.x
2. Obar, Jonathan A., and Steven S. Wildman. “Social Media Definition and the Governance Challenge: An Introduction to the Special Issue.” Telecommunications Policy 39, no. 9 (2015): 745–750. https://doi.org/10.1016/j.telpol.2015.07.014
3. World Wide Web Consortium. “ActivityPub.” W3C Recommendation, 23 January 2018. https://www.w3.org/TR/activitypub/
4. Covington, Paul, Jay Adams, and Emre Sargin. “Deep Neural Networks for YouTube Recommendations.” Proceedings of the 10th ACM Conference on Recommender Systems (2016): 191–198. https://research.google/pubs/deep-neural-networks-for-youtube-recommendations/
5. TikTok. “How TikTok Recommends Videos #ForYou.” 18 June 2020. https://newsroom.tiktok.com/how-tiktok-recommends-videos-for-you
6. Meta. “How AI Influences What You See on Facebook and Instagram.” 29 June 2023. https://about.fb.com/news/2023/06/how-ai-ranks-content-on-facebook-and-instagram/
7. Naumov, Maxim, et al. “Deep Learning Recommendation Model for Personalization and Recommendation Systems.” arXiv:1906.00091 (2019). https://doi.org/10.48550/arXiv.1906.00091
8. Liu, Zhiqiang, et al. “Monolith: Real Time Recommendation System With Collisionless Embedding Table.” arXiv:2209.07663 (2022). https://doi.org/10.48550/arXiv.2209.07663
9. Federal Trade Commission. A Look Behind the Screens: Examining the Data Practices of Social Media and Video Streaming Services. September 2024. https://www.ftc.gov/reports/look-behind-screens-examining-data-practices-social-media-video-streaming-services
10. Ali, Muhammad, et al. “Discrimination through Optimization: How Facebook's Ad Delivery Can Lead to Biased Outcomes.” Proceedings of the ACM on Human-Computer Interaction 3, CSCW (2019), article 199. https://doi.org/10.1145/3359301
11. Pew Research Center. “Social Media Fact Sheet.” 20 November 2025. https://www.pewresearch.org/internet/fact-sheet/social-media/
12. Meta. “How We Review Content.” 13 November 2019. https://about.fb.com/news/2019/11/community-standards-enforcement-report-nov-2019/
13. Yang, Kai-Cheng, et al. “Botometer 101: Social Bot Practicum for Computational Social Scientists.” Journal of Computational Social Science 5 (2022): 1511–1528. https://doi.org/10.1007/s42001-022-00177-5
14. Snap. “Say Hi to My AI.” 27 February 2023. https://newsroom.snap.com/say-hi-to-my-ai
15. Meta. “Introducing New AI Experiences Across Our Family of Apps and Devices.” 27 September 2023. https://about.fb.com/news/2023/09/introducing-ai-powered-assistants-characters-and-creative-tools/
16. YouTube. “How We're Helping Creators Disclose Altered or Synthetic Content.” 18 March 2024. https://blog.youtube/news-and-events/disclosing-ai-generated-content/
17. Meta. “Our Approach to Labeling AI-Generated Content and Manipulated Media.” 5 April 2024, updated 1 July and 12 September 2024. https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/
18. TikTok. “Partnering with Our Industry to Advance AI Transparency and Literacy.” 9 May 2024. https://newsroom.tiktok.com/partnering-with-industry-to-advance-ai-transparency-and-literacy
19. Coalition for Content Provenance and Authenticity. C2PA Technical Specification, version 2.2. May 2025. https://spec.c2pa.org/specifications/specifications/2.2/index.html
20. Cinelli, Matteo, et al. “The Echo Chamber Effect on Social Media.” Proceedings of the National Academy of Sciences 118, no. 9 (2021): e2023301118. https://doi.org/10.1073/pnas.2023301118
21. Nyhan, Brendan, et al. “Like-Minded Sources on Facebook Are Prevalent but Not Polarizing.” Nature 620 (2023): 137–144. https://doi.org/10.1038/s41586-023-06297-w
22. Guess, Andrew M., et al. “Reshares on Social Media Amplify Political News but Do Not Detectably Affect Beliefs or Opinions.” Science 381, no. 6656 (2023): 404–408. https://doi.org/10.1126/science.add8424
23. Gauthier, Germain, Roland Hodler, Philine Widmer, and Ekaterina Zhuravskaya. “The Political Effects of X's Feed Algorithm.” Nature 652, no. 8109 (2026): 416–423. https://doi.org/10.1038/s41586-026-10098-2
24. Pennycook, Gordon, et al. “Shifting Attention to Accuracy Can Reduce Misinformation Online.” Nature 592 (2021): 590–595. https://doi.org/10.1038/s41586-021-03344-2
25. American Psychological Association. “Health Advisory on Social Media Use in Adolescence.” May 2023. https://www.apa.org/topics/social-media-internet/health-advisory-adolescent-social-media-use
26. Office of the Surgeon General. Social Media and Youth Mental Health: The U.S. Surgeon General's Advisory. 2023. https://www.hhs.gov/surgeongeneral/reports-and-publications/youth-mental-health/social-media/index.html
27. National Academies of Sciences, Engineering, and Medicine. Social Media and Adolescent Health. Washington, DC: National Academies Press, 2024. https://doi.org/10.17226/27396
28. Fassi, Luisa, et al. “Social Media Use and Internalizing Symptoms in Clinical and Community Adolescent Samples: A Systematic Review and Meta-Analysis.” JAMA Pediatrics 178, no. 8 (2024): 814–822. https://doi.org/10.1001/jamapediatrics.2024.2078
29. European Parliament and Council. Regulation (EU) 2022/2065, Digital Services Act, Articles 27, 34, 35, 38, and 40. 19 October 2022. https://eur-lex.europa.eu/eli/reg/2022/2065/oj
30. European Parliament and Council. Regulation (EU) 2024/1689, Artificial Intelligence Act, Articles 50 and 113. 13 June 2024. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
31. Ofcom. “Platforms Must Start Tackling Illegal Material from Today.” 17 March 2025. https://www.ofcom.org.uk/online-safety/illegal-and-harmful-content/enforcing-the-online-safety-act-platforms-must-start-tackling-illegal-material-from-today ; “Age Assurance.” Updated 24 July 2025. https://www.ofcom.org.uk/online-safety/illegal-and-harmful-content/age-assurance
32. Australian eSafety Commissioner. “Social Media Age Restrictions.” Updated 23 July 2026. https://www.esafety.gov.au/about-us/industry-regulation/social-media-age-restrictions
33. United States. TAKE IT DOWN Act, Pub. L. No. 119-12, 139 Stat. 55. 19 May 2025. https://www.govinfo.gov/app/details/PLAW-119publ12
34. Federal Trade Commission. “FTC Finalizes Changes to Children's Privacy Rule Limiting Companies' Ability to Monetize Kids' Data.” 16 January 2025. https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-finalizes-changes-childrens-privacy-rule-limiting-companies-ability-monetize-kids-data