Chapter 26: Misinformation, Disinformation, and Fact-Checking

Opening: The Tweet That Wasn't True — and Spread Anyway

On a Tuesday morning in October, five weeks before Election Day in the Garza-Whitfield congressional race, a screenshot circulated across local Facebook groups, neighborhood WhatsApp chains, and several regional news aggregators. The image purported to show a quote from Representative Elena Garza stating that she had voted to defund local police departments — a position she had never held and a vote that, by every verifiable record, had never occurred.

Within four hours, Adaeze Nwosu — executive director of OpenDemocracy Analytics — had received the screenshot via three separate tiplines. Her data journalist Sam Harding had already queued up a preliminary tracking analysis. By noon, the false claim had been shared more than 12,000 times across platforms. By the time ODA published a formal fact-check at 4 p.m., two local television affiliates had already aired segments mentioning the "controversy" — neither of which had confirmed the underlying claim.

"The correction," Sam told Adaeze that evening, "reached maybe a fifth of the people who saw the original. And some of them got more entrenched."

This chapter is about why that dynamic is so persistent, so politically consequential, and so difficult to disrupt — and what analysts, journalists, and civic technologists can do about it anyway.

26.1 Taxonomy of False and Misleading Information

Political discourse has always included falsehoods. What has changed in the contemporary information environment is the speed of dissemination, the fragmentation of shared reference points, and the industrialization of deliberate deception. Before we can analyze misinformation rigorously, we need a precise vocabulary.

26.1.1 The Core Typology

Scholars Claire Wardle and Hossein Derakhshan, in their foundational 2017 report for the Council of Europe, proposed a typology organized along two dimensions: the degree of falsity in the content and the intent to harm of the person sharing it. This gives us a more nuanced vocabulary than the colloquial "fake news."

Misinformation refers to false or inaccurate information that is shared without deliberate intent to deceive. The person spreading it may genuinely believe it to be true. The neighborhood grandmother forwarding a screenshot about a candidate's alleged vote falls into this category. She is not running a disinformation operation; she is a victim of one.

Disinformation is false information that is deliberately created and disseminated to deceive, manipulate, or harm. The person or organization behind a coordinated influence campaign spreading fabricated quotes knows the content is false and shares it anyway. Intent is the distinguishing criterion.

Malinformation is information that is technically true but is deployed in a context designed to cause harm — doxxing a private individual, releasing legitimately obtained private communications to damage a political opponent, or presenting accurate statistics stripped of the context that would change their meaning.

Beyond these three core categories, Wardle identifies seven content types arranged by increasing falsity and increasing fabrication:

1. Satire and parody — No intent to deceive, but potential to be mistaken for genuine news when stripped of satirical markers
2. Misleading content — Misleading framing of genuine information (a real video, a real quote, but presented in a context that distorts its meaning)
3. Imposter content — Genuine sources impersonated (a fabricated tweet designed to look like it came from a real journalist's account)
4. Fabricated content — New, entirely false content designed to deceive (the Garza quote screenshot)
5. False connection — Headlines, visuals, or captions that don't support the content they accompany
6. False context — Genuine content shared with false contextual information (a photograph from a 2012 protest relabeled as a current riot)
7. Manipulated content — Genuine content that has been altered (a real speech with audio editing, a real photograph with digital manipulation)

💡 Intuition: Why the Typology Matters for Analysis Different types of misinformation require different detection and correction strategies. Fabricated content requires source verification. False context requires reverse image search and timeline reconstruction. Misleading content requires access to the full original document. Conflating these types produces confused responses. ODA's fact-checking workflow assigns each incoming claim to one of Wardle's seven types before any further analysis begins, because the evidentiary standard and correction approach differs for each.

26.1.2 The False Binary of "True/False"

One of the most consequential methodological debates in fact-checking concerns the assumption that claims exist on a simple true/false binary. PolitiFact's "Truth-O-Meter" uses a six-point scale from True through Mostly True, Half True, Mostly False, False, and Pants on Fire. The Washington Post Fact Checker uses a one-to-four Pinocchio scale. These scales acknowledge that many political claims are partially accurate, selectively presented, or contextually dependent.

The category of "technically true but misleading" creates particular difficulty for automated detection systems. A campaign advertisement might accurately cite a congressman's vote on a procedural motion while framing it as a substantive policy vote. A headline might report a real unemployment statistic while omitting that it measures a different population than the one described. These claims pass simple factual verification while systematically distorting the picture they paint.

📊 Real-World Application: The Manipulation Index The Oxford Internet Institute's Computational Propaganda Project developed a "Junk News" index for cross-national comparison. Their criteria rated outlets on: (1) professional transparency, (2) source transparency, (3) funding and ownership disclosure, (4) sensationalism and credibility, (5) bias, (6) falsehood frequency, and (7) counterfeit content. Their findings, published across studies from 2017–2022, documented that during major election cycles in the US, UK, France, and Germany, junk news circulated as widely or more widely than legitimate news on social media — despite constituting a small fraction of total content producers.

26.2 Disinformation vs. Misinformation: The Role of Intent

The intent distinction matters for at least two reasons: moral responsibility and strategic response.

26.2.1 Moral and Legal Responsibility

A person who forwards a false claim without knowing it is false bears a different moral responsibility than the actor who fabricated and seeded the claim in the first place. Legal frameworks have generally been reluctant to impose liability on good-faith forwarders, while recognizing that deliberate fabricators may face defamation liability, election law violations, or — in some jurisdictions — criminal charges for deliberate interference with electoral processes.

The intent question is also analytically difficult to resolve empirically. We often cannot observe the mental states of actors. We can sometimes infer intent from behavioral patterns — coordinated inauthentic behavior, use of bot networks, rapid amplification from newly created accounts — but these are indicators, not proof. Platform researchers have developed probabilistic approaches: the likelihood that a pattern of behavior was uncoordinated versus orchestrated. Stanford Internet Observatory's research on coordinated inauthentic behavior campaigns established methods for distinguishing organic (if rapid) organic sharing from artificially amplified networks.

26.2.2 State-Sponsored vs. Domestic Disinformation

A significant portion of academic and journalistic attention has focused on foreign state-sponsored disinformation — Russia's Internet Research Agency during the 2016 US election, China's operations targeting Taiwan, Iran's operations against various opposition movements. These campaigns are real and consequential.

However, the research record increasingly demonstrates that domestic disinformation — produced by homegrown political operatives, partisan media, and hyperpartisan influencers — drives more of the false information that reaches ordinary voters. A 2019 study by researchers at MIT found that false news traveled faster, deeper, and more broadly on Twitter than true news, and that humans — not bots — were primarily responsible for the virality. The false news was more novel, more emotionally arousing, and thus more likely to be shared by real people than by automated accounts.

🔵 Debate: Does Distinguishing Intent Actually Matter for Policy? One school of argument holds that the distinction between misinformation and disinformation is analytically important but practically irrelevant for platform policy: false information causes harm whether the person sharing it intended harm or not. Platforms should focus on the content, not the intent. The opposing view argues that conflating them produces overbroad censorship regimes that cannot distinguish satire from fabrication, mistake from malice — with severe consequences for free expression. The policy debate over this distinction has intensified as platforms have implemented (and rolled back, and re-implemented) various content moderation approaches.

26.3 How Misinformation Spreads: Virality, Networks, and Motivated Reasoning

Understanding the spread of misinformation requires integrating insights from three distinct fields: network science, cognitive psychology, and communication studies.

26.3.1 Network Structure and Cascade Dynamics

Social media platforms create information cascades through resharing. When a piece of content is shared, it reaches a new audience, some fraction of whom share it again, producing exponential spread. The cascade structure — how many people share it, how quickly, through how many independent pathways — determines the ultimate reach of a piece of content.

Soroush Vosoughi, Deb Roy, and Sinan Aral's landmark 2018 Science paper on false news diffusion on Twitter analyzed approximately 126,000 rumor cascades from 2006 to 2017. Their findings were stark: false news spread faster, reached more people, and penetrated deeper into networks than true news. The median false news cascade reached 1,000 to 100,000 people; the top 1% of true news cascades rarely exceeded 1,000. False political news showed the strongest effect — it was 70% more likely to be retweeted than true political news.

Their analysis found that novelty was a key driver: false news was more novel (by measures of information distance from previous tweets) than true news, and novel information generated higher engagement. Emotional arousal — particularly negative emotions including fear and disgust, but also positive emotions like surprise — amplified sharing behavior.

26.3.2 Homophily, Echo Chambers, and Filter Bubbles

Social networks are not random. They are structured by homophily — the tendency to connect with people who share one's characteristics, beliefs, and information environment. In a politically homophilous network, false information that aligns with the dominant political priors of a cluster will circulate extensively within that cluster while receiving skeptical scrutiny (if any) from outside.

The concept of the "filter bubble" (associated with Eli Pariser) suggests that algorithmic personalization creates information environments tailored to individual preferences, reducing exposure to cross-cutting viewpoints. The empirical evidence on filter bubbles is more nuanced than popular accounts suggest: research by Axel Bruns and others has found that most users still encounter cross-cutting content, and that choice-based sorting (people selecting homophilous networks) often outweighs algorithmic effects. But even a partial filtering effect, when applied to billions of users, creates conditions for false information to persist and flourish within ideological communities.

26.3.3 Motivated Reasoning and Identity-Protective Cognition

The cognitive mechanisms underlying misinformation acceptance are well-documented in social and cognitive psychology. Motivated reasoning refers to the tendency to evaluate evidence in ways that protect or confirm prior beliefs and group identities, rather than in ways that track truth. Dan Kahan's work on "identity-protective cognition" shows that politically contested scientific information is assessed through a lens of tribal identification — people with higher scientific literacy actually showed greater polarization on contested empirical questions (like climate science) because they were better equipped to find flaws in unwelcome evidence.

This has direct implications for fact-checking: corrections that feel like attacks on identity may not only fail but may strengthen attachment to the original false belief — the "backfire effect" discussed in the next section.

Illusory truth effect: Repeated exposure to a claim increases its perceived truth, even when the claim has been previously flagged as false. Psychologist Lynn Hasher and colleagues documented this in the 1970s; subsequent research has found it operates in political contexts. The mere repetition of a false claim — even in a corrective context — increases its perceived truthfulness for a subset of hearers. This creates a profound challenge for fact-checking, which necessarily repeats the false claim while attempting to debunk it.

⚠️ Common Pitfall: Assuming High-Information Voters Are Immune One intuitive assumption is that voters who follow politics closely, read multiple news sources, and have higher formal education are resistant to misinformation. The research record does not support this. Motivated reasoning often operates more strongly in engaged partisans because they encounter more political information (and thus more opportunities to engage in partisan information processing) and because their political identity is more central to their self-concept. High-engagement voters can simultaneously be high-susceptibility-to-motivated-reasoning voters.

26.4 Why Corrections Often Fail — and When They Work

26.4.1 The Backfire Effect: A Revision

The "backfire effect" — the claim that correcting a false belief sometimes causes the believer to hold it more strongly — was popularized by Brendan Nyhan and Jason Reifler's 2010 study. It became one of the most cited findings in political communication research and contributed to widespread pessimism about the utility of fact-checking.

Subsequent replication attempts, including several by Nyhan himself, substantially revised the picture. Multiple high-powered studies failed to find backfire effects for corrections of political misinformation. A 2019 meta-analysis by Wood and Porter found no evidence of backfire effects across 52 experimental treatments covering a wide range of political topics. The current scholarly consensus is that backfire effects are rare, difficult to reliably produce in laboratory settings, and not the typical response to corrections.

This is actually good news for fact-checkers — but it comes with important qualifications. The absence of backfire does not mean corrections work in the strong sense of restoring accurate beliefs. The typical finding is that corrections produce partial, temporary belief updating that decays over time and may not translate into changed behavior.

26.4.2 Inoculation Theory and Prebunking

If debunking (correcting false information after exposure) is limited in effectiveness, prebunking — inoculating people against misinformation before they encounter it — has shown more consistent positive results. Inoculation theory, developed by William McGuire in the 1960s as a model of persuasion resistance, has been applied to misinformation by Sander van der Linden and colleagues.

The theory proposes that exposing people to a weakened form of a misleading argument, together with a refutation of that argument, confers resistance to the full-strength version — analogous to a vaccine. Applied research by van der Linden, John Cook, and Ullrich Ecker has found that "bad news prebunking" (explaining common manipulation techniques like fear-mongering, false experts, cherry-picking) reduced susceptibility to misinformation across political topics.

The Bad News game (getbadnews.com) and Harmony Square (designed for election misinformation specifically) use gamification to deliver inoculation at scale. Research on these interventions found significant reductions in sharing intention for misinformation.

✅ Best Practice: Fact-Check Message Design Corrections are more effective when they: (1) lead with the true information rather than repeating the false claim first, (2) provide a clear causal narrative that fills the explanatory gap left by removing the false belief, (3) come from sources the audience trusts, (4) are timely (the shorter the interval between false claim exposure and correction, the better), and (5) explicitly flag the correction as correcting a specific misleading claim (rather than just providing accurate information without context).

26.4.3 When Corrections Work

The circumstances under which corrections most reliably update beliefs include:

• Low partisan valence: Corrections to factual claims about topics not strongly linked to partisan identity produce more consistent updating
• Source credibility: Corrections from sources the recipient views as credible and non-partisan perform better
• Repetition of the correction: A single correction may not be sufficient; repeated corrections appear to consolidate updating
• Correction by a co-partisan source: For highly partisan topics, corrections from a source affiliated with the believer's own party are more effective than corrections from outgroup sources

🔗 Connection to Chapter 23 (Persuasion Research): The conditions for effective corrections closely parallel the conditions for effective political persuasion more generally — source credibility, narrative coherence, and relevance to existing beliefs and values. A fact-check that is technically accurate but communicatively ineffective achieves little.

26.5 The Fact-Checking Industry: Methods and Limitations

26.5.1 The Major Organizations

The fact-checking industry in the United States expanded dramatically after 2007, when PolitiFact (Tampa Bay Times) launched, followed by the Washington Post Fact Checker in 2007 and the expansion of FactCheck.org (operating since 2003 as part of the Annenberg Public Policy Center). By 2022, the Duke Reporters' Lab documented more than 350 active fact-checking organizations in 73 countries.

PolitiFact evaluates specific factual claims made by politicians and public figures. Trained journalists research the claim, consult primary sources and expert opinion, and rate the claim on the Truth-O-Meter scale. The organization operates state-specific bureaus (PolitiFact Georgia, PolitiFact Texas, etc.) that cover state-level races — a model directly relevant to the Garza-Whitfield race.

FactCheck.org takes a similar approach with a particular focus on federal candidates and national political advertising. Its Wire service fact-checks a wider range of claims, including viral social media posts. FactCheck maintains a separate "SciCheck" project focused specifically on claims about scientific topics.

Washington Post Fact Checker uses the Pinocchio scale (1–4 Pinocchios, plus a "Geppetto Checkmark" for claims that are unusually accurate and a "Bottomless Pinocchio" for claims repeated more than 20 times after being rated false). The Post has been particularly active in tracking cumulative false claims by political figures.

26.5.2 Methodological Challenges

Fact-checkers face several structural challenges that limit their reach and impact:

Claim selection bias: Fact-checkers select which claims to investigate, creating a non-random sample. Research by Brendan Nyhan and colleagues found that fact-checkers tend to focus on more easily verifiable, politically salient claims — which may systematically miss the more insidious forms of misleading communication that are harder to fact-check (innuendo, implication, framing).

Epistemological limits: Some claims are not falsifiable by available evidence, or involve genuine uncertainty. The distinction between "false" and "contested" is not always clear, and overly aggressive ratings risk undermining the credibility of the fact-checker itself.

Speed asymmetry: Misinformation can be produced and spread faster than it can be verified and corrected. The ODA example in this chapter's opening illustrates the asymmetry: four hours to produce a careful fact-check, versus four hours of unimpeded viral spread. A common finding in tracking studies is that corrections consistently lag false claims by hours to days.

Reach disparity: Fact-checks reach a relatively small, highly educated, already-politically-engaged audience — often the people least susceptible to the misinformation in the first place. Research on whether fact-checks reach the intended targets (people who have seen the false claim) consistently finds that they largely do not.

⚠️ Common Pitfall: Conflating Fact-Checking with Electoral Impact The research literature on fact-checking's effects on political beliefs, while showing modest positive effects in controlled experiments, provides much weaker evidence for effects on electoral outcomes. The fact that a claim was fact-checked and rated false does not mean the false claim ceased to affect the beliefs of the voters who saw it, nor that those corrected beliefs influenced their vote choice. The chain of causation is long, each link is probabilistic, and the evidence for downstream electoral effects is thin.

26.6 Platforms' Role: Amplification, Moderation, and Ad Policy

26.6.1 Algorithmic Amplification

Social media platforms optimize for engagement — clicks, shares, comments, reactions — because engagement is the mechanism through which advertising revenue is generated. False and emotionally arousing content tends to generate higher engagement than dry, accurate content. This creates a structural incentive for platforms to serve content that spreads misinformation, not because platforms want to spread misinformation, but because their incentive architecture does not penalize it.

The Facebook Files (documents published by the Wall Street Journal in 2021 based on leaked internal research) showed that Facebook's own researchers documented this dynamic internally — the platform's engagement-maximizing algorithm amplified angry, divisive content — but internal proposals to reform the algorithm were sidelined for business reasons.

Twitter/X's former head of trust and safety Yoel Roth documented in post-employment testimony and writings that the platform's engagement metrics systematically rewarded controversy, that "viral" false content consistently outperformed corrections, and that the business model created perverse incentives for the content moderation function.

26.6.2 Content Moderation Approaches

Platforms have deployed several approaches to misinformation, with varying effectiveness and substantial controversy:

Removal: Taking down false content entirely. Most platforms reserve this for the most egregious categories — content related to COVID-19 health misinformation, election fraud claims that directly call for illegal activity, deepfake manipulated media. Removal is the most aggressive intervention and generates the most "censorship" criticism.

Labeling: Attaching warning labels or contextual information to potentially false content without removing it. Twitter's application of warning labels to election-related misinformation in 2020 generated substantial attention. Research on labeling effectiveness found mixed results: labels did reduce sharing intention and belief for some audiences, but also produced "implied truth" effects — content without a label was perceived as more trustworthy, even when the platform simply had not gotten around to labeling it.

Friction: Slowing or discouraging sharing by inserting prompts ("Are you sure you want to share this?") before a piece of disputed content can be amplified. Experiments found that accuracy prompts — asking users to reflect on whether content is accurate before sharing — reduced sharing of misinformation by 5–15%.

Demotion: Reducing algorithmic distribution of flagged content without removing it. This has the advantage of limiting spread while avoiding the political controversy of removal, but is opaque and difficult to audit.

Deplatforming: Removing accounts that repeatedly violate policies, particularly superspreader accounts. Research following Donald Trump's Twitter deplatforming in January 2021 found a 73% reduction in online misinformation in the week following (as documented by the Election Integrity Partnership), though some portion of the traffic migrated to other platforms.

📊 Real-World Application: The 2020 Election Information Operations The Election Integrity Partnership (Stanford Internet Observatory, University of Washington, Graphika, and the Atlantic Council's Digital Forensic Research Lab) produced the most comprehensive real-time tracking of election misinformation in 2020. They documented 4,804 unique pieces of election misinformation meeting their threshold criteria, resulting in 639 platform actions. The post-election period saw the greatest concentration of false election fraud claims. Their research demonstrated that a small number of "superspreader" accounts accounted for a disproportionate share of misinformation reach — the top 21 accounts drove nearly a third of all engagement with false election claims.

26.6.3 Political Advertising and Platform Policy

Political advertising is subject to different (generally weaker) content standards than organic content on most platforms. Traditionally, political advertising has been exempt from the advertiser guidelines that apply to commercial advertising. In the run-up to 2020, Twitter banned political advertising entirely. Google restricted targeting capabilities for political ads. Facebook allowed political ads while applying limited fact-checking, a policy that generated significant criticism.

The political advertising question is thorny because of the First Amendment. In Buckley v. Valeo (1976) and subsequent cases, the Supreme Court established strong protections for political speech, including paid political advertising. Platforms applying fact-checking to political ads risk legal and regulatory challenges that do not apply to, say, misleading ads for consumer products.

26.7 The Misinformation Lifecycle in a Political Campaign

A campaign information environment involves multiple overlapping processes. Tracking misinformation requires understanding each phase.

26.7.1 Origin: Seeding and Laundering

False claims do not emerge randomly. They are typically produced by one of several actors: opposition campaigns (or allied super PACs) who produce and distribute negative information, some of which may be false or misleading; hyperpartisan media outlets who prioritize engagement over accuracy; domestic or foreign influence operations who deliberately fabricate content; or ordinary users who misremember, misread, or share without verification.

"Laundering" refers to the process by which a claim produced by a disreputable or anonymous source acquires the appearance of legitimacy by being repeated by progressively more credible outlets. A fabricated story on a fringe blog is picked up by a partisan aggregator, then mentioned (even dismissively) by a mainstream outlet in the context of coverage of "viral claims," and by the time the mainstream outlet mentions it, the claim has acquired a veneer of newsworthiness.

26.7.2 Amplification: Who Drives Spread?

Research consistently identifies two key amplification mechanisms: influential nodes in social networks (high-follower accounts, political elites, celebrities) and coordinated networks of accounts (bots, inauthentic accounts, troll farms). Elite cues are particularly powerful — when political leaders repeat or implicitly endorse false claims, their followers' acceptance of those claims increases substantially.

In the Garza-Whitfield race, ODA's tracking found that the false defund-the-police screenshot spread through three distinct networks: a local conservative Facebook group (organic sharing by community members who believed the claim), a regional radio host's Twitter feed (elite amplification), and a set of approximately 40 accounts created within the past 30 days that shared the same content in rapid sequence (coordinated inauthentic behavior).

26.7.3 Persistence: Why False Claims Outlive Corrections

Even after a claim is definitively fact-checked as false, it tends to persist in circulation. Several mechanisms explain this:

• Correction non-diffusion: Fact-checks reach a fraction of the audience that encountered the original false claim
• Memory updating failure: Even people who see a correction continue to be influenced by the original claim, particularly under time pressure or cognitive load
• Identity entrenchment: When a false claim aligns with existing beliefs, those beliefs provide ongoing psychological support for the claim even after explicit correction
• Reseeding: Coordinated actors can re-introduce already-debunked claims to new audiences continuously

ODA's post-race analysis of the Garza case found that the defund-the-police claim continued to be shared at significant rates for 18 days after their fact-check was published. Searches for Garza's name on social media continued to surface the original screenshot more prominently than the correction for weeks.

26.8 Case: False Claims in the Garza-Whitfield Race

26.8.1 ODA's Tracking Methodology

When the screenshot arrived via ODA's tiplines, Sam Harding began with what they called "provenance tracing." First, they verified the originating claim: had Representative Garza voted to defund police departments? They pulled voting records from GovTrack, the Congressional Record, and the representative's own legislative history. No such vote existed. They contacted the Garza campaign's press office, which confirmed the claim was fabricated and noted that the screenshot appeared to be an edited image — the font inconsistency was a giveaway.

Second, Sam traced the screenshot's spread using CrowdTangle (a Facebook-owned tool for tracking content across Facebook pages and groups) and social media tracking dashboards. They documented the original posting location (a now-deleted account), the first major amplification node (the local radio host), and the secondary spread patterns across Facebook groups.

Third, Adaeze made an editorial decision about how to publish the fact-check. Following the best-practices literature on correction design, ODA led with the accurate information: "Representative Garza has never voted to defund any police department. Records show she has voted to increase local law enforcement funding in two of the past three budget cycles." The false claim was quoted second, clearly flagged as fabricated, with the visual evidence of the image manipulation included.

26.8.2 Meridian's Experience: Polls Mischaracterized

At the same time, Meridian Research Group's Vivian Park was dealing with a different species of misinformation. Meridian had released a poll showing the Garza-Whitfield race as a toss-up — 47% Garza, 46% Whitfield, 7% undecided, within the margin of error. Within hours of publication, social media posts were characterizing the poll as showing Whitfield "surging" into the lead and Garza "in trouble."

The mischaracterization was technically understandable to a casual reader — within-margin-of-error results are counterintuitively difficult to communicate to non-specialist audiences — but was being spread by actors who clearly understood polling methodology and were deliberately stripping it of context.

"The claim that we showed Whitfield leading is false," Vivian told her team. "We showed a statistical tie. Those are genuinely different things, and the difference matters for how people interpret the race." Vivian published a thread on X explicitly correcting the mischaracterization, tagged the accounts that had spread it, and issued a press release. The correction reached some of the audience. The mischaracterization continued to circulate.

Carlos Mendez built a simple tracking dashboard that pulled together media mentions of Meridian's poll along with the characterization of the results — "leading," "tied," "lagging." The dashboard showed that 62% of mentions that week characterized the results as showing a Whitfield lead, despite this being flatly inconsistent with Meridian's published data.

26.8.3 Data Journalism Approaches to Tracking

ODA's approach to the Garza-Whitfield race combined several data journalism techniques:

Content monitoring: Using keyword alerts, CrowdTangle, and the Brandwatch social listening tool to track how specific claims and their corrections spread in real time.

Network analysis: Using network mapping (covered in depth in Chapter 27) to identify clusters of accounts spreading misinformation and distinguish organic from coordinated activity.

Longitudinal tracking: Documenting the half-life of misinformation — how quickly it decays — and comparing it to the diffusion curve of corrections.

Reach-vs-correction ratio: A simple but powerful metric: for every person who encountered the false claim, how many encountered the correction? In the Garza case, the ratio was approximately 5:1 — five people saw the false claim for every one who saw the fact-check.

🔴 Critical Thinking: Is the Problem Solvable? A structurally uncomfortable question underlies the entire fact-checking enterprise: if corrections consistently reach a fraction of the false claim's audience, if motivated reasoning limits belief updating, and if business model incentives favor engagement over accuracy, are current approaches to misinformation response fundamentally inadequate? Some scholars (like Zeynep Tufekci) argue that fact-checking treats a systemic problem as if it were a content problem — and that without structural changes to platform incentives, fact-checking at the margins is largely symbolic. Others argue that marginal improvements aggregate to consequential effects at scale, that inoculation offers a structural intervention, and that the alternative to imperfect correction is uncontested misinformation. This is not a resolved debate.

26.9 Structural Factors and the Political Economy of Misinformation

26.9.1 Why False Information Is Economically Rational

For certain producers, creating and spreading misinformation is financially and politically rational. Advertising-funded media ecosystems reward engagement over accuracy; a false, outrageous claim generates more engagement than a careful, nuanced correction. The Macedonian teen entrepreneurs who ran dozens of pro-Trump fake news sites during the 2016 election were motivated by AdSense revenue, not political ideology — they tried pro-Clinton sites first and found they generated lower engagement.

For political campaigns and operators, negative misinformation about opponents can suppress turnout among opponent-leaning voters, generate earned media coverage (even corrections attract attention to the original claim), and provide cheap opposition research at scale. The disincentive to produce misinformation — reputational cost, legal risk — is limited when fabricators can operate pseudonymously and when platform enforcement is inconsistent.

26.9.2 Epistemic Inequality and Differential Vulnerability

Not everyone is equally equipped to evaluate information quality. Research by Emily Thorson and others documents that false political information has larger and more persistent effects on people with lower prior knowledge of politics, less media literacy, and less access to diversified information sources. This creates a pattern of epistemic inequality: well-resourced, highly educated people with broad media access are more likely to have their false beliefs corrected; less-resourced people with narrower information diets may not.

The political implications are significant: if misinformation disproportionately affects less-educated voters, and if campaigns know this, there are incentives to target misinformation toward lower-engagement voters who are harder to reach with corrections.

⚖️ Ethical Analysis: ODA's Publication Choices When ODA fact-checks a false claim, they face a genuine ethical dilemma: publishing the fact-check increases the total number of people who learn about both the false claim and the correction. If the claim had only circulated in a small network, publishing a fact-check might bring it to the attention of a much larger audience. Adaeze has a policy of not fact-checking false claims that are circulating in fewer than 5,000 accounts — below that threshold, the amplification risk from the fact-check outweighs the correction benefit. This illustrates a genuine tension: the responsible journalism ideal of correcting all misinformation can, at the margin, cause more harm than silence.

26.10 Measurement and the Limits of What We Can Know

Studying misinformation empirically involves a recursive problem: we are trying to measure the spread of false information in a data environment that is itself the subject of information manipulation. Several challenges merit direct acknowledgment:

Platform data access: Most misinformation research has relied on data from platforms that researchers were allowed to access. Changes in Twitter's API policy in 2023 substantially limited academic access to social media data, reducing the ability to track misinformation in real time. CrowdTangle, Facebook's primary research tool, was discontinued in 2024. The research infrastructure that produced the 2020 election studies is substantially degraded.

Selection effects in detection: We study misinformation that was detected. The false claims that evaded detection are, by definition, outside our sample. Our estimates of the prevalence of misinformation are lower bounds.

Causation vs. correlation in effects research: Most research on the effects of misinformation on political behavior is observational. True experimental designs (randomly exposing some voters to misinformation and not others) face ethical and practical constraints. We can document that people who were exposed to more false content had different beliefs, but establishing that the exposure caused the belief difference is methodologically demanding.

📊 Real-World Application: Measuring Misinformation at Scale The Information Disorder Index developed by the Global Disinformation Lab combines several measurable indicators: velocity (how quickly claims spread), reach (how many accounts and users), persistence (how long they circulate), and correction penetration (what fraction of exposed users saw a fact-check). This composite measure allows comparison across claims, campaigns, and election cycles. ODA adapted this framework for the Garza-Whitfield race, producing one of the most detailed single-race tracking studies available at the time.

26.11 Toward a Data Literacy Response

The most robust long-term response to the misinformation problem is not debunking individual claims — a game that misinformation producers will always win on volume — but building the informational immune system of the electorate. Several evidence-based approaches are promising:

Media literacy education: Research on curriculum-based media literacy programs finds consistent positive effects on students' ability to identify misleading content, evaluate source credibility, and resist emotional manipulation. The effects are modest but durable.

Lateral reading: The practice adopted by professional fact-checkers of immediately opening multiple new browser tabs to check a source's reputation (rather than reading deeply within the source) has been shown, in training studies, to dramatically improve accuracy in evaluating web content — outperforming the approaches used by university professors and professional historians.

Platform design interventions: Frictions such as accuracy prompts, slowing down sharing, surfacing related context, and increasing the visibility of source information have shown positive effects in experiments. The challenge is scaling them without generating user resistance and avoidance.

Civic epistemics: The broader project — building shared norms for evaluating claims and standards for public discourse — goes beyond what any single intervention can accomplish. But the research suggests that social norms matter: people are less likely to share misinformation when they believe their social circle holds accurate-information-sharing as a value.

🌍 Global Perspective: Misinformation in Non-English Contexts Most of the academic research cited in this chapter was conducted in English-language contexts, primarily the United States. Misinformation dynamics differ substantially in other contexts. In Brazil, WhatsApp group sharing (rather than public social media) has been the primary vector for political misinformation, creating a private, encrypted environment that is nearly impossible to monitor or correct. In Myanmar, Facebook's algorithmic amplification of anti-Rohingya content contributed to incitement to ethnic violence — a case where misinformation had consequences incomparably more severe than electoral distortions. In Taiwan, a sophisticated government-civil society partnership (the counter-disinformation work of the mygov0.tw model) has developed rapid-response correction infrastructure that is frequently cited as the most effective national model.

26.12 Summary

Misinformation in political campaigns operates through a complex interplay of content properties (novelty, emotional valence), network structure (homophily, cascade dynamics), cognitive mechanisms (motivated reasoning, illusory truth), and institutional failures (platform incentives, reach asymmetries between false claims and corrections).

The fact-checking industry provides an important public service but faces structural constraints — claim selection bias, speed asymmetry, limited reach to target audiences — that reduce its aggregate impact. Platform interventions have produced measurable improvements in some settings but face business model and legal obstacles to comprehensive implementation. Inoculation approaches show promise as structural interventions that reduce susceptibility before exposure.

For political analysts, the key operational lessons are:

1. Track misinformation using systematic data collection methods, not ad hoc monitoring
2. Measure the reach-correction ratio, not just whether a fact-check exists
3. Design corrections using evidence-based message design principles
4. Recognize the limits of correction-based approaches and advocate for structural changes where appropriate
5. Account for epistemic inequality — different voter populations face different information environments and different vulnerabilities

The next chapter applies computational methods to analyze political text and media at scale, equipping analysts with the technical tools to operationalize the monitoring and analysis approaches described here.

Key Terms

Misinformation — False or inaccurate information shared without deliberate intent to deceive.

Disinformation — False information deliberately created and disseminated to deceive or manipulate.

Malinformation — True information deployed out of context to cause harm.

Backfire effect — The (now largely discredited) claim that corrections to false beliefs sometimes strengthen those beliefs.

Inoculation theory — The theory that pre-exposure to weakened forms of misinformation arguments confers resistance to full-strength versions.

Filter bubble — The information environment created by algorithmic personalization that reduces exposure to cross-cutting content.

Motivated reasoning — The tendency to evaluate evidence in ways that protect prior beliefs rather than in ways that track truth.

Illusory truth effect — The increase in perceived truthfulness of a claim through repeated exposure.

Laundering — The process by which misinformation produced by a disreputable source acquires the appearance of credibility through repetition in progressively more credible outlets.

Lateral reading — The fact-checking practice of opening external sources to evaluate a source's credibility rather than reading within the source itself.

Discussion Questions

1. The distinction between misinformation and disinformation rests on intent, which is often unobservable. How should analysts and fact-checkers handle this empirical limitation in practice? Does the distinction matter if the harms are equivalent?

2. Vivian Park's poll was technically accurately reported in its raw numbers but characterizing a within-margin-of-error result as a lead is a misrepresentation. How should platforms handle content that is factually correct but contextually misleading?

3. Given the evidence on correction effectiveness, what would you say to a funder who is considering investing in a fact-checking organization for an upcoming election? What realistic outcomes can they expect?

4. The "epistemic inequality" research suggests misinformation is not a uniform problem — it disproportionately affects less-educated, lower-information voters. What are the ethical implications for campaign strategy if this differential vulnerability can be exploited?

5. ODA's policy of not fact-checking claims circulating below 5,000 accounts reflects a judgment about amplification risk vs. correction benefit. How would you design a threshold policy? What factors would your threshold include?

26.13 The Empirical Research Agenda: What We Still Don't Know

The misinformation literature has grown explosively since 2016, but significant gaps in knowledge persist. Honest analysts acknowledge these gaps when framing research and policy recommendations.

26.13.1 Long-Run Effects of Exposure

Most experimental research on misinformation effects measures belief change in the short term — minutes to a few days after exposure. The question of whether false political beliefs persist over election-relevant time frames (weeks to months), and whether they survive subsequent corrective information encountered organically in the media environment, is largely unresolved. Longitudinal panel studies that track the same respondents through an election cycle and follow up afterward are methodologically demanding and expensive, which is why they are rare.

Matthew Levendusky and Dominik Stecula's 2021 book A Little Bit of Knowledge argues that media effects on political attitudes are generally weak and short-lived in the aggregate, while producing meaningful effects for specific subpopulations (low-information voters, people in information-poor environments). This perspective, if accurate, suggests that the electoral impact of individual misinformation campaigns is modest — though the aggregate effect of a sustained misinformation environment may still be consequential.

26.13.2 Cross-Platform Migration and Dark Social

A significant limitation of existing research is its near-exclusive focus on public social media platforms — Twitter/X and Facebook — for which researchers have had some data access. But a substantial fraction of political information sharing now occurs through "dark social" channels: private messaging applications (WhatsApp, Telegram, Signal), private Facebook groups, and closed Discord servers. These channels are practically invisible to researchers and essentially unmonitorable in real time.

The dark social problem is not merely a research inconvenience. Misinformation that circulates primarily through WhatsApp group chains — as is the case in Brazil, India, and other WhatsApp-dominant media environments — cannot be tracked using the methods developed for Twitter. It cannot be labeled by platforms without violating end-to-end encryption. It cannot be debunked through public fact-checks that recipients are unlikely to encounter. The entire architecture of misinformation response developed for public social media has limited application to encrypted private channels.

26.13.3 The Threshold Question

A question that the field has not adequately addressed is whether there is a threshold effect in misinformation exposure — a level below which exposure has no meaningful electoral effect and above which effects become consequential. If effects are linear (each additional exposure adds proportionally to belief change), the appropriate policy response is minimizing all exposure. If effects are threshold-based (exposure below some saturation level has negligible impact), then interventions can focus on preventing saturation rather than eliminating all exposure. The empirical evidence to distinguish these models is limited.

26.13.4 Heterogeneity in Response

The research literature appropriately emphasizes average effects across study populations, but political analysts are often more interested in heterogeneous effects — how different voter segments respond differently to the same misinformation. Research by Nyhan, Settle, Thorson, and collaborators has found that misinformation effects are substantially larger for low-knowledge voters and voters in information-poor environments, but the mechanisms driving this heterogeneity are not fully understood. Is it differential exposure? Differential ability to evaluate source credibility? Differential social network reinforcement? Understanding the mechanisms matters for intervention design.

📊 Real-World Application: The MSNBC Asymmetry One consistent and politically sensitive finding in the misinformation literature is that false political information circulating during the 2016 and 2020 election cycles was not evenly distributed across the ideological spectrum. Research by Guess, Nagler, and Tucker found that sharing of misinformation on Facebook was concentrated among older, conservative users. This is an empirical finding about a specific historical period and media environment; it does not establish a permanent asymmetry, and the mechanisms driving the difference are debated. Analysts who use this finding must be careful to distinguish the empirical finding from causal claims about why the asymmetry exists.

26.14 Practical Toolkit for Political Analysts

Political analysts, campaign data operatives, and civic media organizations need practical tools for engaging with misinformation. This section surveys the toolkit.

26.14.1 Monitoring Tools

CrowdTangle (Facebook, now Meta) tracked public content on Facebook pages, groups, and Instagram accounts. As of 2024, CrowdTangle was being replaced by the Content Library and Content Library API, which has more restricted access requirements.

Brandwatch, Sprinklr, Meltwater: Commercial social listening platforms that aggregate data across platforms and provide keyword tracking, sentiment analysis, and volume monitoring. These are widely used in campaign communications departments for real-time monitoring. They are expensive but provide more comprehensive coverage than academic tools.

Google Trends: Free and publicly available, Google Trends data can track the relative search volume for specific terms or claims over time. While it does not show the content of misinformation, spikes in search volume for false claim keywords can indicate when false claims are achieving broad public attention.

Botometer (OSoMe, Indiana University): A tool for estimating the probability that a Twitter account is a bot, based on account behavior, network structure, and content patterns. Useful for distinguishing coordinated inauthentic from organic spread, with appropriate uncertainty acknowledgment.

Snopes, PolitiFact, FactCheck.org APIs and feeds: These organizations provide structured data feeds of their fact-checks. Building a misinformation monitoring system that integrates existing fact-check databases allows analysts to check incoming claims against already-rated claims rather than starting from scratch.

26.14.2 Verification Techniques

Reverse image search: Google Images, TinEye, and Yandex Image Search allow analysts to find the earliest appearance of an image on the web, which is the first step in checking whether an image has been taken out of context or relabeled. TinEye is particularly useful for detecting doctored images, as it indexes images by visual similarity.

InVID/WeVerify: A browser plugin and web tool designed for video verification. It performs keyframe extraction for reverse video search, provides metadata about video files, and integrates fact-check databases. Widely used by fact-checkers for video verification.

EXIF data extraction: Digital photographs contain embedded metadata (Exchangeable Image File Format data) including camera make, timestamp, and often GPS coordinates. Verifying or challenging the claimed provenance of photographs often begins with EXIF data analysis. Tools like ExifTool and online EXIF viewers provide access to this metadata.

Archive.org Wayback Machine: For claims about what a website or social media post said in the past, Archive.org provides snapshots of web pages at various points in time. Useful for checking whether a website's "About" page matches its claimed identity, or whether a social media post was edited after the fact.

26.14.3 Network Analysis Tools

Gephi: Open-source network visualization software. When combined with social media data collected via API, Gephi can produce network maps showing how claims spread through social networks, identify key amplification nodes, and distinguish coordinated from organic sharing patterns. Chapter 33's dashboard project includes network visualization components.

Graphika: A professional network intelligence firm that provides social media network analysis as a service. Their public reports on influence operations (available at graphika.com) are models of rigorous, documented methodology.

NetworkX (Python): The Python library for network analysis, introduced conceptually in several earlier chapters. Sam's misinformation tracking pipeline integrates NetworkX for identifying clusters of accounts showing coordinated behavior patterns.

✅ Best Practice: Document Your Monitoring Protocol Any systematic misinformation monitoring effort should have a written protocol specifying: which platforms and channels are monitored, what keywords trigger review, how incoming claims are classified, what evidentiary standard is required before a claim is assigned a rating, how the correction message is designed, and how reach is tracked. Without a documented protocol, monitoring becomes ad hoc, inconsistent, and difficult to evaluate. ODA's monitoring protocol is updated before each election cycle based on lessons from the previous cycle and current threat landscape assessments.

26.14.4 Communicating Risk to Non-Technical Audiences

One underappreciated skill in political analytics is communicating misinformation risk to non-technical audiences — campaign managers, elected officials, party communications directors, and civic organization boards who need to understand the problem without being research consumers.

Effective communication of misinformation risk to these audiences requires:

Concreteness over abstraction: "This false claim has been shared 47,000 times in our district this week" is more actionable than "misinformation is a significant challenge in the current information environment."

Action framing: Briefings that explain the problem without suggesting actionable responses are frustrating for decision-makers. Every misinformation risk brief should include a "recommended response" section specifying what, if anything, the organization can and should do.

Honest uncertainty acknowledgment: Overstating what we know about the claim's reach or impact will damage credibility when estimates prove wrong. Including confidence intervals and explicitly flagging what is unknown builds trust with sophisticated audiences.

Avoiding the "sky is falling" trap: Research shows that framing misinformation as an overwhelming, unfixable problem can produce hopelessness and disengagement. Communicating the problem should be calibrated to promote appropriate response, not learned helplessness.

26.15 The Constitutional and Legal Landscape

Political misinformation in the United States operates within a constitutional framework that severely limits governmental responses, creating a stark contrast with some other democracies.

26.15.1 First Amendment Constraints

The First Amendment's protection of free speech has been interpreted by courts to cover political speech broadly, including false speech about political matters. In United States v. Alvarez (2012), the Supreme Court struck down the Stolen Valor Act, which criminalized false claims about receiving military honors. The plurality opinion held that false statements of fact do not categorically fall outside First Amendment protection. This case has been widely interpreted as limiting the government's authority to criminalize political misinformation.

Defamation law provides some remedy for false factual claims that damage reputations, but political claims about public figures are subject to the actual malice standard from New York Times v. Sullivan (1964): a public figure must demonstrate that the defendant knew the statement was false or acted with reckless disregard for its truth or falsity. This is a high standard that few political misinformation claims meet, particularly when the false claim is about a policy position or voting record that involves some interpretive judgment.

26.15.2 Election Law

Several states have enacted laws specifically targeting false statements in election campaigns, but these have faced First Amendment challenges. The Stolen Valor ruling and subsequent cases have put significant doubt on the constitutionality of broad false-campaign-speech statutes. Some narrower provisions — prohibiting false statements about election administration (polling place location, voting dates, registration requirements) — have survived legal challenge on the theory that they target a specific, narrow category of verifiable misinformation with direct electoral harm.

26.15.3 Platform Private Ordering

In the absence of government authority to regulate political misinformation, the primary regulatory actors are the platforms themselves — exercising editorial judgment as private companies. The Supreme Court's 2024 decisions in Moody v. NetChoice and NetChoice v. Paxton addressed whether states could compel platforms to carry speech (or prohibit them from moderating speech), affirming that platforms have significant First Amendment rights to make their own editorial judgments about content. This strengthens the constitutional basis for platform content moderation while limiting state authority to compel content carriage.

🔴 Critical Thinking: Who Governs the Information Space? The current governance structure for political misinformation is characterized by: government with limited constitutional authority to regulate content; platforms with maximum editorial discretion but business model incentives misaligned with information quality; civil society organizations (fact-checkers, media literacy educators) with limited resources and reach; and individual citizens with primary responsibility for their own information diet but limited tools for evaluating information quality. Critiques from both left and right find this structure inadequate — the left arguing platforms are insufficiently regulated and exercise their discretion to amplify harmful content, the right arguing platforms exercise their discretion to suppress conservative voices. The governance question remains among the most contested in contemporary democratic theory.

26.16 The Global Dimensions of Political Misinformation

The analysis in this chapter draws primarily on research conducted in the United States, with its distinctive First Amendment framework, its English-language social media ecosystem, and its particular pattern of partisan polarization. Misinformation operates differently — sometimes dramatically differently — in other national media environments, and political analysts working in or with international contexts must understand these differences rather than assuming American findings generalize.

26.16.1 Platform Infrastructure and Misinformation Vectors

The dominant misinformation vector varies significantly by country, and this variation shapes everything about effective response strategies.

In the United States and the United Kingdom, public social media platforms — Twitter/X, Facebook, YouTube — are the primary amplification environment for political misinformation. Content is public, searchable, and (in principle) monitorable by researchers and platforms. The toolkit of fact-checkers, content labeling, and network analysis was developed for this public social media environment.

In Brazil, India, Indonesia, and much of Latin America, the dominant misinformation vector is WhatsApp. Politically charged messages, images, and videos circulate primarily through private group chats rather than public timelines. This creates several profound differences from the American context:

Invisibility to research: WhatsApp messages in closed groups are end-to-end encrypted and not accessible to researchers, platforms, or fact-checkers in real time. The monitoring infrastructure developed for Twitter is simply inapplicable. Studies of WhatsApp misinformation must rely on tipped messages from researchers embedded in groups, post-hoc analysis of screenshots, or partnerships with platforms that provide aggregate metadata without content access.

Inability to label at point of exposure: Platform labeling — attaching a "Missing context" or "Disputed by fact-checkers" tag to a post — works when the platform controls the display interface. WhatsApp's encrypted message delivery means there is no interface point at which a platform can attach a label before the user sees the content. Correction must happen either before the message enters the group (through algorithmic detection before forwarding, which Meta has implemented in limited form) or externally, after the fact.

Social trust amplification: In the United States, political misinformation often reaches people through algorithmically curated feeds, which have lower social authority than direct personal connection. In WhatsApp group chains, misinformation arrives from a trusted contact — a family member, a friend, a community leader — which increases the perceived credibility of the content. Research on the Brazilian 2018 and 2022 elections found that WhatsApp-distributed misinformation was more durable than social media misinformation because its source was perceived as more trustworthy.

26.16.2 State-Sponsored Disinformation and the International Dimension

Political misinformation in many countries involves state actors — either the government itself producing and distributing disinformation about opponents, or foreign governments interfering in another country's electoral process. This dimension is relatively limited in the United States context (the First Amendment constrains domestic state speech interference) but is central to the misinformation landscape in many other countries.

Autocratic and hybrid-regime contexts: In countries with hybrid or autocratic governance — where formal democratic institutions exist alongside systematic government control of information — state-sponsored disinformation is not an exceptional event but a permanent feature of the political information environment. Research on Hungary, Turkey, India, the Philippines, and Brazil under Bolsonaro has documented government and government-aligned actors using social media (often WhatsApp) to systematically spread false information about opposition politicians, suppress accurate information about government failures, and create information environments that favor incumbents.

Cross-border influence operations: The Russian Internet Research Agency's interference in the 2016 U.S. election and subsequent influence operations in European elections established a template for state-sponsored cross-border disinformation: fake accounts, coordinated amplification of divisive content, and targeted messaging to specific vulnerable voter segments. EU countries have developed early-warning systems (EEAS East StratCom Task Force) specifically for tracking Russian-origin influence operations, reflecting the greater salience of cross-border disinformation in the European context.

26.16.3 Regulatory Divergence and the Global Governance Gap

While the United States relies primarily on platform voluntary action and faces First Amendment barriers to government regulation of political speech, other democracies have moved toward explicit legal frameworks for addressing political misinformation:

The European Union's Digital Services Act (DSA): Effective 2024, the DSA requires large platforms to assess and mitigate systemic risks — including disinformation risks — and subjects them to regulatory oversight. The DSA's transparency requirements include researcher data access provisions that partially address the research access problems created by Twitter/X's API restrictions and CrowdTangle's discontinuation. The DSA represents the most significant binding regulatory framework for platform misinformation management in any major democracy.

Taiwan's counter-disinformation model: Taiwan has developed what is frequently cited as the most effective national counter-disinformation system. The Presidential Hackathon, the civic technology organization g0v, and the government's "humor over rumor" rapid-response strategy — deploying fast, accessible corrections in meme format within hours of false claim detection — have demonstrated measurably better correction diffusion than traditional fact-checking. The Taiwan model works in part because of specific cultural and institutional factors that may not transfer directly to other contexts, but its core principles — speed, accessibility, non-partisan framing, government-civil society partnership — are broadly applicable.

🌍 Comparative Note for Analysts Analysts who apply American misinformation research to non-American contexts risk significant analytical errors. The assumption that public social media is the dominant vector, that platform labeling is the primary tool, that the legal framework limits government intervention, and that partisan polarization follows American patterns are all context-specific. Before applying findings from this chapter to international political analytics work, assess: What is the dominant distribution platform in this context? What is the regulatory framework? What is the relationship between state actors and misinformation production? Who are the trusted fact-checking organizations in this media environment, and what is their reach relative to misinformation producers? The answers to these questions determine which tools and strategies from this chapter are applicable and which require fundamental adaptation.