• False and misleading information spreads easily because current information systems reward speed and engagement, concentrate distribution power, and sit on top of eroding trust and media infrastructure.
• Treat information integrity as infrastructure. Build shared diagnostics, robust public-interest media, and standing fact-checking and coordination capacity so systems detect and correct distortions early.
• Realign incentives and feedback loops. Use transparency, diversity constraints, and circuit breakers for virality so recommendation systems are less likely to reward harmful cascades.
• Grow epistemic resilience from sturdy roots. Invest in community-based knowledge institutions and long-term media, digital, and civic literacy so communities can hold their own in turbulent information environments.

When a major outbreak, election, or geopolitical crisis hits, a familiar pattern plays out. Health agencies, election bodies, and platforms roll out fact-checks, advisories, and moderation efforts, yet rumors still surge through group chats and feeds, pulling attention away from official guidance and corroding trust.¹ During major health emergencies, the World Health Organization describes this as an “infodemic,” where misleading material and half-true claims stack up faster than people can sort out what is reliable.²

The problem is that the environment rewards speed and emotional punch. Economic models of online sharing show that when attention is scarce, and engagement is the main currency, people have strong incentives to circulate content that signals identity and emotion, even when it increases the risk of spreading misinformation.³ Empirical work on health communication confirms that vivid, emotionally charged posts often travel further than careful explanations and can gradually erode confidence in experts.⁴

Taken together, these findings point toward a systems problem rather than a series of isolated mistakes. Early work on cross-epistemic resilience described resilient knowledge systems as those that can absorb shocks, keep core functions like verification and correction running, and adapt without collapsing into confusion.⁵ Recent syntheses extend that logic to whole information environments. Epistemic resilience becomes a property of institutions, media structures, and social norms working together, rather than a matter of individual media savvy alone.⁶ Comparative research on epistemic vulnerability shows that countries with concentrated media ownership, high polarization, and weak safeguards are more exposed to sustained misinformation pressure than societies with plural, well-regulated media ecosystems.⁷ Under those conditions, even well-intentioned interventions land in a brittle environment.

Policy work at the OECD level echoes this shift in perspective. Analyses argue that scattered, reactive responses cannot replace stronger governance of information integrity that treats it as a long-term public good.⁸ Core guidance now frames disinformation and misinformation as systemic challenges that interact with media markets, regulatory choices, and civic culture.⁹ Reviews of policy responses to false and misleading digital content find that many measures remain piecemeal and crisis-driven, with limited baselines for tracking progress.¹⁰

Newer recommendations call for a more infrastructure-oriented approach that strengthens journalism, independent oversight, and shared diagnostics for information health.¹¹ Guidance on combating misinformation online emphasizes that tactical fixes like fact-checks, labels, and nudges need to operate inside ecosystems that already support trust, verification, and timely correction, rather than act as a thin patch over structural weaknesses.¹²

In many places, the basic infrastructure that supports shared knowledge has thinned. Local newspapers close, public-service broadcasters face political pressure, and independent fact-checkers struggle to secure stable funding. OECD work on governance responses to disinformation warns that these trends erode the “information integrity” that democratic systems rely on, especially when combined with weak regulatory safeguards.⁸

At the same time, people face growing volumes of digital content and a mix of legacy media, partisan outlets, influencers, and group chats. Official guidance from international bodies stresses that media, information, and digital literacy should sit inside broader strategies for dealing with misinformation, not on the margins, yet implementation often trails behind.⁹

Fact-checking organizations and verification networks now play a central role in many countries. They monitor claims, publish analyses, and provide structured evidence that journalists, platforms, and civic groups can reuse. OECD reviews describe these actors as key parts of the response to false and misleading digital content.¹⁰ Still, they are frequently treated as add-ons or short-term projects rather than as core utilities. When funding cycles end or partnerships lapse, capacity drops and must be rebuilt in the next crisis.

Newer policy frameworks argue for a more systematic approach. In 2024, the OECD adopted a Recommendation on Information Integrity that treats high-quality journalism, independent regulators, and resilient civic information as public goods that deserve long-term investment and policy coordination.¹¹ Yet even where such frameworks exist, many countries lack concrete diagnostics that show where infrastructure is strongest and where it is fraying. Without stable institutions and shared metrics, responses remain episodic. Each pandemic, election, or geopolitical shock triggers ad hoc coalitions between platforms, governments, and NGOs, followed by long periods of drift. The ecosystem never fully upgrades its baseline defenses.

Epistemic resilience improves when information integrity is governed with constant monitoring, investment, and maintenance rather than reactive, crisis-only action.⁸ One design move is to develop recurring diagnostics that track the health of the information ecosystem. These could include indices of media pluralism, measures of epistemic vulnerability, and indicators for fact-checking capacity and regulatory safeguards.⁷ Regular public reporting would make slow erosion visible and give policymakers a basis for targeted support.

A second move is to put core institutions on firmer footing. That means sustainable funding and independence protections for public-interest media, local newsrooms, and national fact-checking hubs, combined with transparency requirements that keep them accountable and credible.¹⁰

Third, cross-platform coordination protocols can be made permanent rather than improvised. During high-risk periods such as pandemics or national elections, existing guidance already calls for structured collaboration between health agencies, platforms, and civil society on harmful content.¹¹ Turning those arrangements into standing features, with clear triggers and evaluation plans, would help systems respond faster and more coherently when the next wave arrives.

Finally, community-level knowledge infrastructure deserves direct attention. Libraries, community radio, local NGOs, and grassroots networks often act as trusted intermediaries. When these actors have access to timely, reliable information and tools for monitoring rumors, they can localize responses in ways that feel legitimate to their audiences.8 These steps create durable scaffolding that makes accurate information easier to access, scrutinize, and reuse in moments of stress.

Scroll through the feed of any large platform, and you’ll see the design logic at work. The interface invites you to react, share, and keep going. Underneath, ranking systems learn from those signals. Economic models of online content sharing describe this as a strategic complementarity: when people gain status and connection from likes and shares, and when platforms earn revenue from attention, everyone is nudged toward amplifying whatever already looks popular.³ Engagement starts to act like gravity.

In that environment, content that is polarizing, sensational, or misleading enjoys a built-in advantage whenever it triggers strong emotion or group identity. Outrage, moral disgust, and in-group pride are fast, intuitive responses. From a behavioral perspective, they tap into availability and affect heuristics, which make vivid items feel more important and more true. Once these posts start to pick up momentum, algorithms tuned to maximize interaction interpret the spike as a success signal and show them to more people.³

Health communication during the COVID era is a concrete example. Analyses of platform data found that emotionally charged, low-quality health content could outperform measured guidance from official sources, especially when framed as insider knowledge or resistance to government control.⁴ People who engaged with a few such posts were more likely to see similar material later, because the system had learned that this cluster of themes and tones held their attention. Over time, this feedback loop fostered anti-science sentiment and eroded trust in experts and institutions, even among users who did not think of themselves as “anti-science.”⁴

Governance analyses show how hard this is to correct from the outside. Platforms seldom provide detailed, consistent reporting on how their recommendation systems affect diversity of exposure, source quality, or the prevalence of low-integrity content.⁸ When transparency appears, it often arrives as broad principle statements rather than stable metrics that researchers can track. External teams see the outcomes of algorithm changes in aggregate behavior or in public controversies, not in clear, shared dashboards. That opacity makes it difficult to study how design choices alter epistemic outcomes, and it leaves little room for structured feedback.⁸

OECD reviews of policy responses highlight the same structural tension. Transparency and accountability around recommendation systems are underdeveloped, while incentives to maximize engagement continue to shape product decisions.¹⁰ Teams are rewarded for growth in attention metrics, even when those gains come from content that undermines social cohesion or informed participation. From an epistemic resilience perspective, this is more than a bug in one component. It is a system-level design flaw: a learning architecture that can optimize for clicks while remaining largely blind to how its own feedback loops shift what people believe and whom they trust.¹⁰

Engineering epistemic resilience means changing what the system optimizes for, not only which posts it labels or removes. Policy work on disinformation points toward a shift from ad hoc fixes to measurable rules for how recommendation engines behave.⁸ One pillar is structured transparency. Platforms can publish regular, independently auditable reports that show how ranking algorithms treat different types of content and how changes alter exposure to trustworthy versus low-integrity sources.¹⁰ Those reports would include simple metrics for diversity of viewpoints, concentration of attention, and the share of traffic going to outlets that meet basic integrity standards. The goal is not to expose proprietary code, but to create a shared view of outcomes that regulators, researchers, and civil society can interrogate.

A second pillar is to bake quality and diversity constraints directly into ranking objectives. International guidance on information integrity stresses that credible journalism and plural media ecosystems are stabilizing forces in democracies.¹¹ Platforms can reflect this by ensuring that, especially on civic topics, people see a mix of reliable outlets and perspectives rather than a narrow stream of whatever happens to perform best on engagement alone. In behavioral terms, this broadens the “choice set” the system presents, so users have more than one salient narrative to anchor on. It also reduces the risk that feedback loops narrow information diets over time.

A third element is the use of circuit breakers for virality. Borrowed from financial markets, these are temporary brakes that slow the spread of content that starts to spike in sensitive domains while extra checks run in the background.8 Practically, that might mean slowing reshares after a certain threshold, prompting users to read an article before forwarding, or inserting a short cue that links to context from trusted sources when a claim begins to move rapidly. The behavioral aim is to interrupt the fastest, most reflexive sharing at the moment when social proof and urgency would otherwise dominate. Corrections then have a chance to reach the same audiences before the narrative hardens.

Over the long term, realignment will also touch business models. As long as revenue depends heavily on maximizing attention, there will be persistent pressure to favor content that drives intense engagement, even when it erodes trust or polarizes audiences. The OECD and civil society reports argue for more experimentation with subscription models, public-interest funding, and mixed revenue streams that reduce the structural pull toward sensationalism.¹¹ That might involve premium spaces for verified information, shared funding mechanisms for public service content, or incentives that reward stable, high-quality contributions rather than constant novelty.

None of these moves “fixes” misinformation in a clean, one-shot way. They change the background conditions under which content spreads and under which product teams make trade-offs. When transparency, diversity constraints, circuit breakers, and healthier business incentives sit together, the system becomes more capable of noticing when engagement gains come at the cost of epistemic damage and adjusting before the next crisis forces a blunt response.

Misinformation harms do not fall evenly. Health communication studies show that communities facing systemic obstacles, including lower access to care, language barriers, or histories of discrimination, are often more exposed to low-quality information and less likely to encounter timely, trusted corrections.¹²

Documents from the OECD and WHO on media and digital literacy make a similar point at the national scale. Countries with weaker public-interest media, low investment in literacy, and high polarization show greater vulnerability to sustained misinformation pressure.¹³ These societies lack both top-down and bottom-up mechanisms to anchor public discourse in shared baselines of fact.

The result is uneven epistemic resilience. Some groups can draw on dense networks of credible media, expert institutions, and community intermediaries. Others rely on narrow mixes of partisan talk shows, influencers, and encrypted messaging. When shocks arrive, like new vaccines or contentious elections, the first group can update beliefs with relative confidence. The second group faces conflicting messages and sees little reason to trust official narratives.

Earlier work by Acemoglu and colleagues suggests that in environments with strong social rewards for sharing and weak accountability, misinformation can become a stable feature of equilibria.¹⁴ That logic interacts with inequality: where people feel marginalized, incentives to align with alternative information networks grow stronger.

System-level resilience improves when communities are equipped and empowered to shape their own information environments, instead of being passive recipients of centrally designed interventions.

Long-term investment in media, digital, and civic literacy remains one of the most promising levers. OECD analyses describe literacy programs as essential enablers of citizen participation and informed engagement, especially when they address how recommendation systems and generative AI work.¹³ When people understand the basic mechanics behind their feeds, they are better able to contextualize what they see and to seek out diverse sources.

Equally important are community-based knowledge institutions. Libraries, local media, youth organizations, faith groups, and grassroots networks already act as validators and translators of complex information. The WHO and national guides on misinformation highlight the value of partnering with such intermediaries during crises, rather than relying solely on national campaigns.¹²

Participatory approaches extend this logic. When researchers, journalists, and community members co-produce data and narratives about local information challenges, they build social networks that can respond quickly when new rumors appear.¹¹ Co-designed monitoring tools, such as community rumor logs or local information observatories, help ensure that early signals travel both upward and sideways.

Finally, governance processes that include communities most affected by misinformation harms can reduce the risk that resilience measures unintentionally widen inequalities. Advisory councils, citizen panels, and structured consultations give marginalized groups a voice in how platforms and regulators define problems and evaluate interventions.¹¹ In combination, these strategies make it more likely that misleading narratives will meet skepticism and context at multiple points in the network.

System-level interventions are slow, political, and often contested. Building or reforming public-interest media, setting transparency requirements, or revising platform incentives requires legislation, regulation, and negotiation among actors with conflicting interests. International reviews stress that evidence on the long-run impact of these measures is still emerging and that evaluations can be difficult because changes roll out gradually and in bundles.¹⁰

There is also a risk that efforts to strengthen information integrity slide into over-centralization. If new rules concentrate power in a small set of state bodies or firms, or if transparency requirements are not matched by robust pluralism, epistemic resilience can suffer even as certain indicators improve.⁸ Societies need multiple, partially independent channels for verification and critique. 

Adversaries will keep adapting. Coordinated disinformation campaigns can migrate to under-regulated platforms, exploit new formats such as synthetic media, or learn to game transparency metrics.¹¹ A resilient system therefore needs continuous monitoring, learning, and revision.

Misinformation is often framed as a problem of bad actors or naive users. A systems view tells a more structural story. Many information ecosystems combine fragile media infrastructures, highly optimized engagement engines, and rising distrust, especially in communities that have good reasons to question authority.

Engineering epistemic resilience means changing how the system behaves when those tensions are high. It means treating information integrity as infrastructure worth measuring and funding, realigning incentives so that the most engaging content is not always the most visible, and building community capacity so that people can maintain shared baselines of reality even when narratives collide.

Content-level tools like fact-checks, labels, and informational nudges remain important. They work best when they sit inside resilient ecosystems that can detect patterns, coordinate responses, and support people who want to stay oriented toward truth. The Decision Lab works with organizations that see information integrity as both a behavioral design challenge and a systems problem. If your mandate is to move from patching feeds to strengthening the epistemic foundations beneath them, let us help you turn this lens on epistemic resilience into a plan for where to invest next.

Building a culture of innovation around Generative AI & LLMs

Get a practical look at standing up safe, fast LLM experimentation inside organizations. This guide covers lab structures, governance touchpoints, and how to turn small pilots into durable capability without losing speed.

How to Preserve Agency in an AI-Driven Future

An exploration of how AI systems can support human judgment instead of quietly steering it. Discover how institutional guardrails, interface transparency, and decision rights allow people to question, contest, and update information even as AI mediates more of what they see.

1. World Health Organization. (2024, February 6). Disinformation and public health: Questions and answers. World Health Organization.
   https://www.who.int/news-room/questions-and-answers/item/disinformation-and-public-health
2. World Health Organization. (n.d.). Infodemic. World Health Organization.
   https://www.who.int/health-topics/infodemic
3. Acemoglu, D., Ozdaglar, A., & Siderius, J. (2024). A model of online misinformation. The Review of Economic Studies, 91(6), 3117–3150.  https://doi.org/10.1093/restud/rdad111
4. Rodrigues, F., Fonseca, J. M., Figueiras, A., & Ladeira, I. (2024). The social media infodemic of health-related misinformation and technical solutions. Health Policy and Technology, 13(3), 100846.
   https://doi.org/10.1016/j.hlpt.2024.100846
5. Hamborg, S., Meya, J. N., Eisenack, K., & Raabe, T. (2020). A cross-epistemic resilience framework for interdisciplinary research on energy transitions. Environmental Innovation and Societal Transitions, 37, 331–345. 
6. Sustainability Directory. (2025). Epistemic resilience. Sustainability Directory.
7. Labarre, J. (2024). Epistemic vulnerability: Theory and measurement at the system level. Political Communication, 42(1), 6–26. https://doi.org/10.1080/10584609.2024.2363545
8. Matasick, C., Alfonsi, C., & Bellantoni, A. (2020). Governance responses to disinformation: How open government principles can inform policy options. OECD Working Papers on Public Governance, 39. OECD Publishing.  
9. Organisation for Economic Co-operation and Development. (n.d.). Disinformation and misinformation. OECD.
   https://www.oecd.org/en/topics/disinformation-and-misinformation.html
10. Hill, J. (2022). Policy responses to false and misleading digital content: A snapshot of children’s media literacy. OECD Education Working Papers, 275. OECD Publishing.
11. Organisation for Economic Co-operation and Development. (2024). Facts not fakes: Tackling disinformation, strengthening information integrity. OECD Publishing.
    https://doi.org/10.1787/d909ff7a-en
12. World Health Organization. (2023). Combatting misinformation online. World Health Organization. https://www.who.int/teams/digital-health-and-innovation/digital-channels/combatting-misinformation-online
13. Organisation for Economic Co-operation and Development. (2023). Media literacy education system. OECD.
14. Acemoglu, D., Ozdaglar, A., & Siderius, J. (2021). A model of online misinformation (NBER Working Paper No. 28884). National Bureau of Economic Research. https://doi.org/10.3386/w28884