How information is produced, packaged, sold, and believed — advertising, journalism, and social media
Part A · what media is — the infrastructure of public knowledge
About this module
Media is the water we swim in. Every day you are the recipient of hundreds of media messages — news articles, advertisements, social media posts, podcasts, TV segments — and almost none of them arrived by accident. Someone chose to publish it, someone chose how to frame it, and someone is paying for it. This module gives you the analytical vocabulary to understand all three choices.
Media literacy is not about cynicism. It is about understanding the system well enough to navigate it on your own terms.
what media does
Agenda-setting
Media does not tell you what to think. It tells you what to think about. The agenda-setting theory, developed by McCombs and Shaw during the 1968 US election, showed that the issues most prominently covered in news become the issues voters consider most important. The mechanism is simple: coverage creates salience. If every outlet runs a story about immigration for two weeks, the public rates immigration as the top problem — regardless of any objective change in immigration numbers.
Framing
The same set of facts can be framed as a crime story, an economic story, or a mental health story. Framing theory (Entman, 1993) argues that how a story is packaged defines which mental "file" the audience reaches for. The 2005 Hurricane Katrina coverage showed this starkly: identical images of people carrying food were described as "looting" for Black subjects and "finding supplies" for white subjects in different wire service captions. The frame shapes the conclusion.
Gatekeeping
Before a story reaches you, it has passed through multiple selection filters: the reporter decides to investigate it, an editor decides to run it, a platform algorithm decides to distribute it. Each filter reflects institutional values, commercial pressures, and personal judgments. The gatekeeping model (White, 1950) showed that a single wire editor rejected 90% of stories not for factual problems but for relevance, taste, and space. Algorithmically curated feeds have not removed gatekeeping; they have automated and obscured it.
Cultivation theory
George Gerbner's long-running research from the 1970s found that heavy TV viewers develop distorted beliefs about reality: they consistently overestimate crime rates, social distrust, and the percentage of the population that are professionals, because that is what TV over-represents. Cultivation is slow and cumulative. The effect is not from any single program but from years of cumulative immersion in a mediated world that diverges systematically from statistical reality.
a brief history of mass media
Evolution of mass media — dominant eras by century
The printing press (1450s)
Gutenberg's movable-type press did not merely make books cheaper: it fundamentally broke the church's monopoly on literacy and authoritative knowledge. Within 50 years of 1450, more books had been produced than in all previous centuries combined. Martin Luther exploited it deliberately — his 95 Theses circulated as pamphlets across Germany within weeks of being nailed to the Wittenberg door in 1517. The Reformation was, in part, a media event.
The penny press (1830s)
Until 1833, American newspapers cost six cents and were read almost exclusively by merchants. The New York Sun changed this by dropping the price to one cent, funded by advertising rather than subscriptions. Mass circulation followed. By 1835, it claimed 15,000 daily readers. This business model — content subsidised by advertising, delivered free or cheaply to build audience — is the same model that Facebook and Google use today.
Broadcasting (1920s–1990s)
Radio and then television created a mass audience of unprecedented scale and simultaneity. On 30 October 1938, Orson Welles's War of the Worlds broadcast panicked thousands of Americans into believing a Martian invasion was real — demonstrating the raw persuasive power of the medium. By 1965, 93% of American homes had a TV set. The three US networks (ABC, CBS, NBC) each commanded audiences of 30–40 million nightly during the 1970s. This degree of shared public attention has never been replicated.
The internet and platform era (1990s–)
The internet initially promised radical decentralisation: anyone could publish. By 2010, it had reconcentrated into a handful of platforms more powerful than any previous media company. Facebook's algorithm reaches 3 billion monthly users. YouTube's recommendation engine accounts for 70% of viewing time on the platform. The printing press took 50 years to destabilise the medieval church; the internet took 20 years to destabilise the newspaper industry. The full consequences are still unfolding.
media ownership
Major media conglomerates — click to explore
Select a conglomerate above.
Concentration of media ownership matters because editorial decisions are shaped by owner interests. Rupert Murdoch's News Corp papers supported every political leader in Australia between 1972 and 2022 who won the election — except one. That is not a coincidence; it is the exercise of structural power.
State and public service media
Not all media is commercial. The BBC, funded by a compulsory licence fee (£169.50 per household in 2024), operates with formal independence from government and reaches 92% of UK adults each week. Al Jazeera, funded by the Qatari government since 1996, demonstrated that state-funded does not automatically mean propaganda — its coverage of the Arab Spring was often more critical of Arab governments than Western outlets. China's CCTV and Russia's RT, however, represent the other pole: state media used actively as tools of foreign influence.
Concentration trends
In 1983, 50 companies controlled the majority of US media. By 2012, that number was six. By 2024, a handful of conglomerates — Comcast, Disney, Warner Bros. Discovery, News Corp/Fox, Paramount/National Amusements — account for most film, TV, and newspaper output in the English-speaking world. This matters for diversity: a range of apparent choices can mask a narrower range of actual ownership perspectives. Local news has suffered most, with 2,000 US local newspapers closing between 2004 and 2020.
media business models
US digital advertising revenue by platform type, 2023
Source: IAB / eMarketer estimates. Meta = Facebook + Instagram. "Other" includes Amazon Ads, TikTok, Snap, Pinterest, and others.
Advertising model
The oldest mass media model: content is made attractive to audiences, and advertisers pay for access to that audience. The New York Times in its print heyday earned roughly 80% of revenue from advertising. The internet destroyed this by fragmenting audiences and redirecting ad dollars to Google and Facebook. By 2020, the Times earned 60% of revenue from subscriptions — a deliberate, difficult pivot that most local papers could not replicate. Advertising still dominates TV, radio, and free-to-access digital media.
Subscription and paywalls
The Financial Times was the first major newspaper to erect a digital paywall, in 2002, on the theory that its readers (financial professionals) had high enough willingness to pay. Others followed reluctantly. The New York Times crossed 10 million digital subscribers in 2023 — more than any newspaper in history — proving the model works for strong brands with unique content. The Washington Post, owned by Jeff Bezos since 2013, lost money in 2023 after a subscription growth stall, illustrating that the model is not automatic.
The platform model
Google and Meta do not produce content. They produce infrastructure for content distribution and sell targeted advertising against it. Google earned $237 billion in 2023, mostly from search advertising. Meta earned $117 billion. Neither employs journalists in the traditional sense, yet both earn multiples of what the entire newspaper industry earns globally. This asymmetry — content produced by journalists, revenue captured by platforms — is the central economic problem of the current media era.
The Substack / newsletter model
Since 2017, Substack has demonstrated that individual journalists can monetise directly without institutional backing. The top 10 Substack writers collectively earn over $25 million per year. This model works for journalists with strong existing audiences or a distinctive niche. It does not work for local accountability journalism, investigative projects requiring teams, or coverage requiring physical presence in conflict zones. The newsletter renaissance is real but does not replace what newsrooms do at scale.
Part B · journalism — the theory and practice of reporting the world
The fourth estate
The term "fourth estate" dates to 18th-century Britain, where Edmund Burke reportedly gestured at the press gallery in Parliament and declared it more important than the three estates (clergy, nobility, commons). The concept holds that journalism's democratic function is to scrutinise power on behalf of the public — not to be a neutral recorder of events, but an adversarial investigator. The Washington Post's investigation of Watergate (1972–74), which relied on a single anonymous source "Deep Throat" and two junior reporters, resulted in a sitting president's resignation. That is the fourth estate in action.
Journalism vs commentary
A common confusion: journalism reports what happened; commentary argues what it means. Both are legitimate and important, but conflating them is how misinformation spreads. A news article should be verifiable from its sources. A column should be clearly labelled as opinion. The problem is that cable news and social media have largely dissolved this distinction. Fox News's most-watched programmes are classified by the network as "opinion" rather than news — which means different legal standards apply — yet many viewers receive them as factual reporting.
news values
Why some things become news — Galtung and Ruge's news values (1965), updated
journalistic standards
Approaches to editorial philosophy — spectrum from objectivity to advocacy
Strict objectivity (view from nowhere)Advocacy journalism
The objectivity norm
The objectivity norm emerged in American journalism in the early 20th century as a commercial strategy as much as an ethical one: if a paper claimed no political affiliation, it could sell to readers across the political spectrum. By the mid-century it had hardened into professional dogma. The model requires reporters to present "both sides" and attribute claims rather than assess their truth. Its core problem: it treats balance as synonymous with accuracy. Climate science coverage, for decades, gave equal weight to 97% of climate scientists and 3% of contrarians — a structurally false balance.
Transparency as the better alternative
Media scholars like Jay Rosen argue that objectivity should be replaced by transparency: reporters should disclose their methods, sources, funding, and perspectives rather than pretending to have none. The Guardian and ProPublica both publish detailed funding and editorial independence statements. This does not solve bias, but it allows readers to assess it. The opposite — claiming objectivity while concealing perspective — is more dangerous because it is invisible. Readers who know a source has a viewpoint can calibrate; readers who believe a source is neutral cannot.
False balance problem
Giving "equal time" to unequal positions misrepresents the truth. On vaccination safety, scientific consensus and anti-vaccine advocates are not equivalent sources of evidence. This matters because journalism's default reflex — find the opposing quote — was designed for political disputes where reasonable people disagree. Applied to empirical questions, it manufactures artificial controversy. Nature's 2023 analysis of climate coverage found that between 2007 and 2020, US newspapers reduced false-balance coverage by 75% after sustained criticism from scientists and media critics.
types of journalism
press freedom
RSF Press Freedom Index 2024 — selected countries (rank out of 180)
Lower score = higher press freedom
Source: Reporters Without Borders (RSF) 2024 World Press Freedom Index. Norway has ranked first for 8 consecutive years.
How journalism is suppressed
State suppression is the most visible form: North Korea, Eritrea, and Iran regularly imprison journalists, and Turkey jailed more journalists than any other country between 2016 and 2022. But the more common mechanism in democracies is legal. SLAPP suits (Strategic Lawsuits Against Public Participation) are defamation claims filed not to win but to drain a journalist's resources through prolonged litigation. The UK's libel laws were so plaintiff-friendly that "libel tourism" — foreign oligarchs suing British journalists in English courts — was a recognised problem until 2013 reforms. Since then, a new wave has emerged through data privacy law.
Self-censorship
The most pervasive form of press suppression leaves no fingerprints. Reporters self-censor stories they know their editor will not run, that will attract advertiser pressure, or that will result in their access to powerful sources being revoked. The access problem is structural: a White House correspondent who writes a story embarrassing to the administration risks losing their press credentials. This is not censorship in the legal sense; it is an incentive structure that produces comparable results. Investigative units with funding independence — ProPublica, ICIJ — exist precisely to insulate reporters from these pressures.
the journalism crisis and misinformation
US newspaper newsroom employment, 2006–2023
71,000
20062023
The revenue collapse
The proximate cause of local journalism's collapse was classified advertising. In 2000, classified ads — job listings, real estate, cars — generated $19.6 billion for US newspapers annually. Craigslist, launched in 1995 and expanded nationally in 2000, gave away the same service for free. Within a decade, classified revenue had fallen by over 80%. Google and Facebook then captured the remaining display advertising. The resulting revenue collapse was not gradual. Between 2004 and 2023, the US lost over 2,500 local papers — one in four. These closures left 70 million Americans in "news deserts" with no local coverage.
Misinformation taxonomy
Claire Wardle's widely-used taxonomy distinguishes: misinformation (false content shared without intent to harm), disinformation (false content deliberately created to deceive), and malinformation (true content shared to cause harm, like leaked private photos). These distinctions matter for responses. Misinformation calls for correction and education. Disinformation requires understanding adversarial actors and their goals. Malinformation calls for legal remedies. Treating all three identically — as "fake news" — obscures why each spreads and what can reduce it.
A 2018 MIT study (Vosoughi, Roy, and Aral, published in Science) found that false news spreads six times faster than true news on Twitter, reaches 10 times more people at its peak, and is 70% more likely to be retweeted. The mechanism: novelty and emotional arousal. False stories are more surprising and more emotion-inducing. This is a structural property of the platform, not a fixable behaviour problem.
how to evaluate a news source
The SIFT method — lateral reading in practice
SIFT was developed by Mike Caulfield at the Digital Polarization Initiative. The key insight of "lateral reading" is that you should leave the page you are evaluating and search for what others say about it — rather than reading it more carefully, which confirms whatever impression you already have.
Part C · advertising — the art and science of persuasion at scale
The fundamental transaction
Advertising is not a tax on business. It is the economic substrate that funds the majority of free content, news, and entertainment in the world. The transaction: audiences give attention; media companies sell that attention to advertisers; advertisers try to convert it into purchases. The phrase "if you're not paying for the product, you are the product" is memorable but imprecise. You are not the product — you are the audience being sold to advertisers. Your data is the targeting infrastructure. The product, in the sense of what changes hands for money, is the ad placement. Understanding this clarifies why platforms have incentives to maximise time-on-platform regardless of user wellbeing.
The attention economy
Herbert Simon noted in 1971 that "a wealth of information creates a poverty of attention." Every piece of media content competes for a fixed resource: human hours of attention. The economics of the attention economy produce predictable distortions: content that generates outrage, anxiety, or arousal attracts more attention than content that generates satisfaction. This is why both news and social media tilt negative — not because the world is predominantly negative, but because negativity is more attention-compelling. Tim Wu's 2016 book The Attention Merchants is the definitive history of this dynamic from 19th-century newspapers to Facebook.
how advertising actually works
Byron Sharp's counterintuitive findings
Marketing professor Byron Sharp's 2010 book How Brands Grow challenged 50 years of advertising conventional wisdom. His key findings: brand loyalty is largely a myth, as most brands grow through occasional light buyers not loyal ones; mental availability matters more than differentiation — brands grow by being easy to think of in purchase situations; and emotional, distinctive advertising that builds memory structures outperforms rational product-benefit advertising. Coca-Cola running ads for people who already drink Coca-Cola is not reinforcement — it is keeping the brand salient for the 80% of category buyers who rarely think about it.
The mere exposure effect
Robert Zajonc demonstrated in 1968 that people develop preferences for stimuli they have been repeatedly exposed to, even without conscious awareness of the exposure. This is the scientific basis for much of brand advertising: a consumer who has seen 10,000 Volkswagen ads over 30 years will, all else equal, reach for the VW over an equally good alternative at the point of purchase — not because they rationally weighed the ads, but because familiarity creates comfort. This is why large brands advertise even when they have nothing new to say. Absence from market creates unfamiliarity, which corrodes preference over time.
targeting and digital advertising
How programmatic advertising works — real-time bidding in under 200ms
The average web page loads ads from 20–50 different advertising technology companies. Ad fraud — bots simulating clicks and page views — accounts for an estimated 22% of digital advertising spend, roughly $84 billion globally in 2023 (CHEQ).
The Google/Meta duopoly
In 2023, Google and Meta together captured approximately 48% of all global digital advertising revenue — about $350 billion between them. Google dominates search advertising; Meta dominates social display. Their moats are their data scale: Google processes 8.5 billion searches daily; Facebook has 29 billion posts per month to train its targeting models against. No other advertising platform can match the granularity of their targeting. The result is that the CPM (cost per thousand impressions) that publishers earn from Google's ad network is typically 60–70% lower than what they could earn by selling directly — illustrating who captures value in the chain.
Cambridge Analytica and micro-targeting
In 2016–17, Cambridge Analytica harvested the Facebook data of 87 million users without consent through a personality quiz app. They used this to build psychographic profiles based on the OCEAN model (Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism) and targeted political ads to voters identified as high-neuroticism persuadables in battleground states. Whether this determined Trump's 2016 victory is disputed — the causal effect of micro-targeted political ads is smaller than Cambridge Analytica claimed in its marketing to clients. But the episode demonstrated that the behavioural profiling infrastructure built for commercial advertising could be trivially repurposed for political manipulation.
Digital advertising calculator — CPM, reach, and budget
Results appear here.
the creative side of advertising
Ad strategy spectrum — from pure rational to pure emotional
50
Pure rational (features, price, specs)Pure emotional (feeling, identity, belonging)
David Ogilvy's principles
David Ogilvy, founder of Ogilvy and Mather and the closest thing advertising has to a canonical theorist, believed that advertising must sell. Not entertain, not win awards: sell. His 1963 book Confessions of an Advertising Man remains a foundational text. Key principles: the headline is the most important element (80% of readers never pass the headline); long copy works for interested buyers; never write an ad you wouldn't want your family to see; and research before you create. He tested everything. His most famous campaign — "At 60 miles an hour the loudest noise in this new Rolls-Royce comes from the electric clock" — worked because the specific detail was credible and the implied claim (extraordinary quietness) was remarkable.
Iconic campaigns and what they did
Apple's "1984" Super Bowl ad (directed by Ridley Scott) is studied not because it ran once but because it defined the brand's identity as a counter-cultural force for the next 30 years. Nike's "Just Do It" (1988) was created in response to Reebok's aerobics dominance; by 1998, Nike's market share had risen from 18% to 43%. Dove's "Real Beauty" campaign (2004) grew sales from $2.5bn to $4bn in three years by deliberately rejecting conventional beauty advertising's models. Each campaign worked by owning a distinctive emotional territory rather than competing on product features.
advertising ethics and PR
Regulation landscape
UK advertising is regulated by the Advertising Standards Authority (ASA), which operates a self-regulatory system with real enforcement power — it can remove ads and ban advertisers. In the US, the Federal Trade Commission (FTC) regulates deceptive advertising at the federal level. Key prohibitions that vary by country: tobacco advertising is banned from broadcast in the UK since 1965 and in the US since 1971; alcohol advertising is restricted but not banned; political advertising in the UK is banned on broadcast but allowed in print and online; in the US, political advertising has near-absolute protection under the First Amendment, creating a distinctive environment where lies in political ads face almost no legal consequence.
Native advertising and sponsored content
Native advertising — paid content designed to look like editorial — has existed since advertorials in 1960s magazines. The internet supercharged it. BuzzFeed built an $800m business primarily on native content. The FTC requires "sponsored" or "advertisement" labels, but research consistently shows that 60–70% of consumers do not notice these labels or do not understand what they mean. The more invisible the disclosure, the more persuasive the content — which creates an obvious incentive to make disclosures invisible. Instagram influencer posts remain systematically under-labelled: a 2023 UK ASA investigation found 50% of sampled influencer posts had inadequate disclosure of paid partnerships.
PR vs advertising
Advertising is paid media: you control the message and pay for placement. PR is earned media: you pitch journalists and hope they run your story, with no control over what they write. The distinction matters because earned media is more credible — readers discount explicit advertising but trust editorial coverage. A single positive feature in The Economist is worth more brand trust than 100 equivalent-reach display ads. PR also includes crisis communications: the Tylenol poisoning crisis (1982) is the textbook case. Johnson and Johnson's immediate recall of 31 million bottles at $100m cost, and their transparent communication, is credited with saving the brand. A cover-up would have ended it.
Part D · social media — the attention machine
How social media differs from all previous media
Every previous mass medium was asymmetric: one broadcaster, millions of passive receivers. Social media is the first genuinely networked medium, where every user is simultaneously a receiver and a potential broadcaster. This is not just a technical change; it transforms the logic of information spread. Virality — the mechanism by which content propagates through networks — has no analogue in broadcast. A tweet by a private individual can reach more people than a CNN broadcast if it is retweeted by the right nodes. The second change is algorithmic curation: unlike broadcast, where the same content is sent to all viewers, every social media feed is individually customised by an engagement-optimising algorithm. You and your neighbour watching the same TV channel see the same programme; you and your neighbour on the same platform see different feeds.
the major platforms
Platform model spectrum — from open public square to closed algorithmic feed
Open / chronological / user-controlledClosed / algorithmic / platform-controlled
the algorithm
Engagement optimisation
Every major social media algorithm is trained to maximise one metric: engagement — defined as likes, comments, shares, and time spent. This sounds benign until you recognise that outrage, anxiety, and moral indignation are among the most engagement-generating emotions. A 2021 internal Facebook study, revealed by whistleblower Frances Haugen, found that the platform's own research showed content that provoked "angry" reactions was distributed 5x as widely as content that provoked "like" reactions. The company had a proposed fix and did not implement it, because it would have reduced overall engagement metrics. This is not a bug; it is a rational response to the optimisation objective.
Filter bubbles — real but overstated
Eli Pariser's 2011 book The Filter Bubble popularised the idea that algorithmic curation creates ideological echo chambers that prevent exposure to challenging viewpoints. The hypothesis is intuitive and alarming. The evidence, however, is more mixed. A 2023 study by Meta's researchers published in Science, examining the actual content seen by 37 million Facebook users, found that while feed ranking did reduce cross-partisan exposure, the effect was smaller than feared, and that individual choice to avoid cross-cutting content mattered more than algorithmic filtering. The problem is real but its magnitude — and therefore its weight as a cause of polarisation — is contested.
YouTube's rabbit hole
YouTube's recommendation algorithm, which determines 70% of viewing time, was found by researchers Guillaume Chaslot (a former YouTube engineer) and others to systematically route viewers toward more extreme content. A viewer watching mainstream political content would, over multiple recommendations, be routed toward increasingly radical alternatives — not because the platform intended radicalisation, but because extreme content generates more watch time and subscriptions, which are the signals the algorithm rewards. YouTube adjusted its algorithm in 2019 to reduce "borderline content" recommendations, claiming a 70% reduction in such recommendations in the US. Independent research suggested the actual reduction was real but smaller.
the creator economy
Influencer tier explorer — adjust follower count to see typical metrics
100K
Nano (1K–10K)Mega (>1M)
Platform dependency
The creator economy's central vulnerability is that creators build audiences on infrastructure they do not own. A platform algorithm change can cut a creator's reach by 80% overnight. Vine shut down in 2016 with 10 days' notice, destroying the careers of hundreds of professional creators. TikTok's potential US ban, debated from 2020 and passed into US law in 2024, showed how geopolitical risk compounds platform dependency. The creators who have built most successfully tend to own their audience through email lists and subscriptions that are platform-independent. YouTube's Casey Neistat, Substack's Heather Cox Richardson, and podcast networks like Crooked Media all built direct audience relationships as insurance against platform shifts.
Creator burnout
The algorithmic imperative to produce at high frequency — YouTube's internal research suggests channels that post less than once a week see significant subscriber decline — creates intense production pressure. A 2022 survey by the Creator Economy Institute found 77% of professional creators reported feeling burnt out. The dynamic is structurally similar to piecework manufacturing: the platform captures most of the value while creators bear all the production cost and risk. YouTube takes 45% of ad revenue. Spotify pays artists an average of $0.003–0.005 per stream. The platforms that host the work capture more than the people who create it.
social media and mental health
What the data shows
The adolescent mental health deterioration documented in Western countries from roughly 2012 coincides with smartphone and social media penetration. In the US, emergency room admissions for self-harm among girls aged 10–14 increased 189% between 2009 and 2021. UK data from NHS Digital shows a similar trajectory. Jonathan Haidt's 2024 book The Anxious Generation attributes much of this to social media and argues for a phone-free childhood. His argument rests on the coincidence of timing, the severity of the effect in girls (who use visual platforms more than boys), and international correlation with smartphone penetration.
The counter-evidence
Several researchers contest the causal claim. Amy Orben and Andrew Przybylski's large-scale 2019 study, using UK data from 120,000 adolescents, found that social media use explained only 0.4% of the variance in wellbeing — comparable to the impact of wearing glasses or eating potatoes. Candice Odgers at UC Irvine has reviewed the literature and argues that correlational data consistently shows that adolescents who are already struggling use social media more, not that social media causes the struggle. The honest summary: there is a real association, the causal direction is disputed, the effect appears stronger for passive consumption and social comparison than for active communication, and girls appear more vulnerable than boys.
social media, democracy, and regulation
Democratic possibilities and risks
Social media enabled the Arab Spring uprisings of 2010–11: Wael Ghonim's Facebook page coordinating Tahrir Square protests became a template for networked activism. The Belarusian Sviatlana Tsikhanouskaya used Telegram to organise mass protests in 2020. Twitter's real-time information during crises — natural disasters, coup attempts, public emergencies — genuinely serves a democratic function that broadcast media cannot replicate. The risks are equally concrete: 2016 Russian Internet Research Agency operations placed 3,500 Facebook ads reaching 11.4 million Americans in the election period; state-sponsored troll farms in Nigeria, Brazil, the Philippines, and India have been documented by Stanford's Internet Observatory since 2017 in at least 30 countries.
Section 230 and platform liability
Section 230 of the US Communications Decency Act (1996) provides that platforms are not legally liable for content posted by users. This legal immunity is what made the internet as we know it possible: without it, Facebook could be sued for every defamatory post, and would either not exist or would moderate so aggressively that user expression would be impossible. Critics argue it has allowed platforms to disclaim responsibility while actively amplifying harmful content through recommendation. The EU's Digital Services Act (2022) takes a different approach: platforms over 45 million EU users face due diligence obligations, transparency requirements, and algorithmic auditing — without eliminating the Section 230-equivalent immunity for hosted content.
Decentralised alternatives
ActivityPub is an open protocol, adopted by Mastodon and now Meta's Threads, that allows social media servers to interoperate without requiring a centralised platform. Think of it as email for social media: you can have a Mastodon account on any server and still interact with users on other servers. The theoretical advantage is that no single company controls the protocol or can be acquired, shut down, or weaponised. The practical limitation: Mastodon had 8 million users in 2024 compared to Facebook's 3 billion. Network effects are structural; platforms that already have everyone are extremely hard to dislodge regardless of their technical or ethical inferiority.
Part E · media literacy — reading the media environment critically
What media literacy actually means
Media literacy is not about distrust. Distrusting everything is as epistemically disastrous as trusting everything — it leaves you unable to distinguish reliable from unreliable, and equally susceptible to bad information dressed in sceptical clothing. Media literacy means understanding the system: who produces content, with what incentives, using what methods, aimed at what audience. It means being able to apply appropriate scepticism calibrated to source reliability, not uniform cynicism. The inoculation theory of media literacy — teaching people the rhetorical techniques of manipulation before they encounter them in the wild — has shown significant effects in randomised trials. Recognising a manipulation technique before it is deployed reduces its persuasive effectiveness by 20–30%.
Confirmation bias and how to resist it
Confirmation bias — the tendency to seek and preferentially believe information that confirms existing beliefs — is the single most important cognitive distortion in media consumption. Critically, it operates on the evaluation of evidence, not just its seeking: people apply stricter scrutiny to evidence that contradicts their beliefs than to evidence that supports them. This is why fact-checking and corrections are less effective than they should be. The most evidence-backed intervention is not "seek different sources" (most people do not) but perspective-taking: asking "what evidence would I need to change my mind?" before evaluating a claim. Predetermining what would count as falsification before reading makes motivated reasoning harder.
understanding bias
Types of media bias — distinct mechanisms, not one thing
Select a bias type above.
media literacy in practice — a checklist
Who published this? Check the publication's Wikipedia page — lateral reading rather than reading the article itself more carefully. If a site was created in the last three years, has no physical address, and no named editorial staff, treat it as low-reliability regardless of how professional it looks.
Who funded it? Independent journalism funded by subscriptions has different incentives than journalism funded by a state, a billionaire, or an advocacy organisation. None of these is automatically unreliable, but the funding source tells you what pressures exist.
Is there a primary source? Most journalism reports secondary or tertiary accounts of events. If a story says "a study found...", find the actual study. Scientific claims filtered through journalism lose important caveats almost universally.
Are multiple reliable sources reporting it? Major true stories are covered by multiple independent outlets. If only one outlet carries a dramatic claim, that is not confirmation that it is false, but it is a reason to wait for corroboration before treating it as established.
What is the emotional register? Content designed to make you feel outrage, fear, or disgust before you think should be read more slowly, not shared. Emotional intensity is a signal to engage critical faculties, not to amplify.
What's not in the story? Framing works by exclusion as much as inclusion. A story about crime that never mentions socioeconomic context, a story about immigration that uses only economic framing, a story about protest that focuses on violence and not grievance: what is left out defines what conclusion is reachable.
the future of media — AI and synthetic content
AI-generated news
By 2024, several major publishers including Associated Press (since 2014) and Axel Springer were using AI systems to generate routine news content: financial earnings summaries, sports score reports, weather updates. The AP publishes approximately 4,000 AI-generated earnings stories per quarter. For structured, low-stakes content, this is arguably an efficiency gain. The risk is the extension of automated content generation to complex, contextual, interpretive journalism where errors are consequential. CNET's 2023 experiment with AI-generated financial articles produced multiple factual errors that went uncorrected until independent review. The problem is not just accuracy but the economics: if AI can generate plausible-seeming journalism at near-zero cost, the incentive for expensive human journalism is further eroded.
Deepfakes and synthetic media
In 2023, AI-generated images of the Pope in a white puffer jacket circulated widely as real. A deepfake audio of London Mayor Sadiq Khan making inflammatory comments circulated ahead of the Armistice Day protests. A fake image of an explosion near the Pentagon briefly moved financial markets. The technology to create photorealistic fake video now runs on consumer laptops. Detection tools exist but are in an adversarial race with generation tools. The strategic effect of deepfakes may be less about specific falsely attributed content and more about what researchers call the "liar's dividend" — the ability of anyone caught in genuine wrongdoing to claim the evidence is a deepfake. Synthetic media raises the cost of truth, not just the supply of falsehood.
What trust requires
In a world of abundant synthetic content, trust will increasingly attach to verified human identity, institutional reputation, and cryptographic provenance. The Coalition for Content Provenance and Authenticity (C2PA) standard, developed by Adobe, Microsoft, and camera manufacturers, allows digital content to carry a cryptographic "nutrition label" verifying where it was created, by whom, and whether it has been modified. The BBC, Reuters, and the New York Times have all joined. This does not solve the problem for the billions of items shared outside C2PA-compliant systems, but it creates a parallel track of verified content that may, over time, become the gold standard for high-stakes journalism and evidence.
The most dangerous near-term effect of AI in media is not a single convincing deepfake but the normalisation of a landscape in which nothing can be assumed to be real. When synthetic fabrication becomes cheap and ubiquitous, the default public response may become blanket scepticism — which benefits those who want to dismiss genuine documented wrongdoing as AI-generated.
Part F · Q&A
If media shapes what we think about, how can anyone form genuinely independent views?
The honest answer is that fully independent views are an ideal rather than a practical possibility. All cognition begins with received information; you cannot have opinions about events you have never heard of. The goal is not independence from media but critical engagement with it: knowing who chose what to cover and why, triangulating across multiple sources with different incentive structures, and maintaining epistemic humility about what you don't know. The researcher Eli Pariser showed that even the same search engine query returns different results to different users, which means that "objective" information access is not a given. This is an argument for deliberate media diversification, not for despair.
Is investigative journalism actually effective — does exposing wrongdoing produce change?
The record is mixed but real. The Washington Post's Watergate reporting brought down a president. The Boston Globe's Spotlight investigation into Catholic Church sexual abuse (2002) led to the resignation of Cardinal Bernard Law and legal reforms in multiple countries. ICIJ's Panama Papers and Pandora Papers investigations produced tax policy changes and criminal prosecutions in over a dozen countries. The pattern that works best: sustained investigation, not single stories; publication in multiple outlets simultaneously to prevent legal suppression; and stories that connect to a specific regulatory failure that officials can be pressured to address. Stories that simply name wrongdoing without actionable policy consequences tend to produce outrage but not change.
Why hasn't the internet made journalism better by allowing anyone to report anything?
It has made some things better — citizen journalism documented the Arab Spring and police brutality in ways traditional media could not have — but it has also made several things worse. First, professional journalism requires physical presence, legal defence, source cultivation, and sustained attention: you cannot investigate a city council's building contracts from your bedroom. Second, the attention economy rewards emotional engagement over accuracy, meaning accurate but less viscerally compelling stories are outcompeted. Third, the collapse in advertising revenue removed the economic engine that paid for news bureaux, foreign correspondents, and investigative teams. The internet democratised publication but did not fund the infrastructure that makes publication worthwhile.
Does advertising actually work — can you measure the return on investment?
Measurability is advertising's most disputed topic. Digital advertising is measurable at the click level: you can track who saw an ad and whether they purchased. But attribution is contested — the customer who clicked an ad after seeing it 14 times on different platforms was already going to buy. A landmark 2013 study by Randall Lewis and David Reiley at Google found that the average display ad campaign on Yahoo had a positive ROI of 34%, but only when tracked through actual sales data, not click-throughs. The problem with clicks as a metric is that they are gameable: ad fraud bots generate billions of fake clicks annually. Field experiments using randomly assigned control groups — the gold standard used by researchers like Brad Shapiro — consistently find advertising effects are real but substantially smaller than the industry claims.
Is advertising uniquely manipulative compared to other forms of communication?
Advertising is unusual in that it is explicitly persuasive, implicitly identified (you usually know it is an ad), and uses both rational and emotional techniques. By these criteria, it is more manipulative than straightforward information provision but no more manipulative than political rhetoric, religious persuasion, or social pressure — all of which are also designed to influence beliefs and behaviour. The distinct ethical problems with advertising are: (1) it is paid, creating an incentive to deceive that disinterested communication lacks; (2) children are especially vulnerable and less able to recognise persuasive intent; and (3) its scale means that systematic biases — toward materialism, body image distortion, stereotyping — have population-level effects. These are genuine harms, but they do not uniquely distinguish advertising from other persuasion.
If we know social media amplifies outrage and extremism, why haven't platforms fixed it?
Because fixing it would reduce engagement, and engagement is the revenue engine. The alignment between user wellbeing and platform revenue is partial, not complete. Content that makes users feel bad but keeps them scrolling is profitable; content that makes users feel satisfied and puts down the phone is not. The Frances Haugen documents showed Facebook had identified specific algorithmic changes that would reduce misinformation exposure but would also reduce overall engagement, and declined to implement them. This is not unique villainy — any shareholder-owned company with the same incentive structure and measurement infrastructure would face the same tension. The structural fix requires either changing the incentive (through regulation, competition, or taxation of attention externalities) or changing the metric. Several researchers have proposed that platforms be required to publish wellbeing metrics alongside engagement metrics, making the tradeoff visible.
Did social media cause the political polarisation of the 2010s?
Probably not primarily. The US has been politically polarising since the 1970s, well before the internet, driven by realignment of the parties along educational and cultural lines. The period of most dramatic polarisation in Congress actually precedes social media's mass adoption. Nolan McCarty, Keith Poole, and Howard Rosenthal's long-run data on congressional voting shows consistent polarisation trend lines from 1980. What social media appears to have done is accelerate emotional polarisation (the belief that the other side is evil rather than merely wrong) and given fringe voices access to audiences they couldn't previously reach. These effects are real and serious but different from causing polarisation from scratch. They amplify an existing trend rather than originating it.
Is media literacy enough — can education solve the misinformation problem?
Education is necessary but not sufficient. The evidence from inoculation studies, prebunking research, and Finland's media literacy curriculum (the most developed in the world, introduced in 2016 as a compulsory subject) shows real effects on individual critical evaluation. But several structural problems remain education-resistant. First, the volume of content is simply too large for critical evaluation of each item. Second, the people most confident in their ability to spot misinformation are systematically worse at it than those with appropriate epistemic humility — the Dunning-Kruger dynamic applies to media literacy. Third, the platforms profiting from misinformation spread are not education's ally. The research consensus is that platform design changes and regulatory requirements are necessary complements to education, not substitutes. The best analogy is food safety: consumer education helps, but you still need hygiene inspections.
Has any society successfully regulated social media — are there working models?
The EU's Digital Services Act (2022) is the most comprehensive attempt to date. It requires Very Large Online Platforms (VLOPs, defined as over 45 million EU users) to assess and mitigate systemic risks, submit to annual independent audits, provide researchers with access to their data, and offer users a "recommender-free" option. The Act's implementation is ongoing; early enforcement actions include investigations into Meta's handling of electoral content, Twitter/X's content moderation failures, and TikTok's protection of minors. Singapore's Online Safety Act (2022) and Germany's NetzDG (2017) represent different approaches. The honest assessment is that no comprehensive regulatory framework has yet been fully tested through the courts and enforcement cycle; we are in the early stages of democratic societies working out how to govern privately-owned global information infrastructure.
Given all of this, should you trust the media at all?
Yes, calibrated trust is both possible and necessary. The alternative — treating all media as equally unreliable — is not epistemically neutral; it is a gift to bad actors who can claim equivalence with credible sources. The empirical record shows that some media institutions are systematically more accurate than others. Repeated verification against reality over time is how you earn calibrated trust. The Reuters Institute at Oxford University tracks media trust across 46 countries annually: Finland, Denmark, and the Netherlands consistently show trust levels above 55%, achieved through press councils, strong professional standards, and public broadcaster models. Trust in media is not fixed; it is a function of media behaviour, institutional design, and the media literacy of the audience. All three are improvable.