Executive summary
A controlled online experiment led by researchers at the University of Copenhagen — with partners at TU Dresden and the Max Planck Institute — provides fresh, actionable evidence that seemingly minor changes in social‑feed algorithms can meaningfully alter political polarisation and the factual accuracy of users’ beliefs. For business leaders, investors and regulators, the study crystallises a recurring dilemma: algorithms that maximize short‑term engagement also amplify division and misinformation; alternatives exist that improve civic outcomes, but they pose trade‑offs for platform business models and market dynamics. The finding matters now because regulators in Europe and elsewhere are moving from moral debate to legal obligations for algorithmic risk mitigation, and because public trust, advertiser preferences and geopolitical information risks are increasingly material to corporate value.
The experiment in brief
Design: Two sequential online experiments with U.S. participants (first phase ~500 people, second phase ~1,000), recruited to represent equal numbers of self‑identified liberals and conservatives.
Treatment: Participants were exposed to feeds generated by different ranking strategies: an engagement‑based algorithm typical of many commercial platforms; a “bridging” algorithm that prioritizes posts that are popular across ideological lines; an “intelligence” algorithm designed to reduce factually misleading beliefs; and random ranking as a baseline.
Findings: The engagement‑based feed produced greater polarisation and lower factual accuracy among participants than some alternative rankings. The bridging algorithm improved cross‑ideological agreement in some settings; the intelligence algorithm increased accuracy of factual beliefs compared with both engagement and random rankings.
Interpretation: Small design choices in content ranking materially affect group belief formation and social cohesion — outcomes that extend beyond user experience to public goods such as information quality and democratic resilience.
Why this matters to business and policy
1. Engagement is a business lever — and a social risk
Most large social platforms have optimized for engagement metrics (likes, shares, time on site) because they map directly to advertising revenue. But the study reinforces an empirical link between engagement optimization and social harms: content that triggers strong emotional reaction tends to polarise and propagate inaccuracies. For boards and investors, this transforms a product design question into a balance of revenue, reputational and regulatory risk.
2. Regulation is catching up
In Europe, frameworks such as the Digital Services Act (DSA) now require very large online platforms to assess and mitigate systemic risks, including societal risks arising from their recommender systems. Broader AI governance — for example the EU’s AI Act — is also converging on obligations for transparency, auditing and risk reduction. The Copenhagen findings bolster the case for stricter regulatory scrutiny and provide concrete algorithmic interventions that regulators could encourage or mandate.
3. Advertisers and brands are sensitive to downstream effects
Brands increasingly care about the context in which their messages appear. Polarizing or deceptive environments can deter advertisers, accelerate shifts toward brand‑safe inventory, and create opportunities for platforms that can credibly demonstrate safer curation. That pressure alters the economics of feed design over time.
4. Geopolitics and system gaming
Algorithms that reduce polarization in one setting may be gamed or repurposed by bad actors; moreover, information operations exploit engagement-maximizing systems. National security agencies and foreign policy actors are watching how platform design choices affect public opinion and social stability.
Nordic and international context
Nordic societies benefit from high levels of digital literacy, institutional trust and strong public media — factors that can blunt but not eliminate the effects identified in the study. Nordic governments and regulators have been active in shaping European rules; Nordic tech firms and public institutions are well placed to pilot alternative feed models that prioritize civic outcomes. Internationally, markets differ: U.S. firms face federal political resistance to sweeping content rules, while the EU is moving toward formal obligations — creating a split regulatory landscape platforms must navigate.
Business implications: risks and opportunities
Risks
– Revenue dilution: Algorithms that deprioritize high‑engagement but polarizing content may reduce advertising yield in the short term.
– Competitive asymmetry: Platforms that refuse to change may retain engagement and ad revenue, creating a first‑mover disadvantage for firms that act on civic concerns unless regulation levels the playing field.
– Legal and reputational fallout: Failure to address systemic algorithmic harms can trigger fines, litigation and brand damage.
Opportunities
– Differentiated value proposition: Platforms can offer curated, quality‑focused feeds to premium users and advertisers, or provide algorithmic choice (engagement vs. civically oriented feeds).
– New compliance and audit markets: Independent algorithmic auditing, certificate services, and feed‑design consultancy can become growth sectors.
– Public‑private partnership: Governments and platforms can co‑fund and pilot “public interest” feeds or transparent algorithmic experiments, building trust and regulatory goodwill.
Practical steps for executives and policymakers
For platform leaders and product executives
– Institute portfolio thinking for recommender systems: test multiple ranking logics in parallel (engagement, bridge, accuracy‑focused) and measure societal as well as commercial metrics.
– Offer algorithmic choice to users and enterprise clients: transparency coupled with opt‑in alternatives may preserve user autonomy and reduce backlash.
– Invest in independent audits and publish risk assessments: proactive transparency reduces regulatory uncertainty and reputational risk.
For investors and boards
– Factor algorithmic externalities into risk assessments: quantify potential revenue impacts from regulatory compliance and advertiser flight.
– Demand scenario analysis: require management to present business plans for possible regulatory outcomes and alternative monetisation models (subscriptions, premium content, contextual ads).
For policymakers and regulators
– Encourage sandboxing and third‑party experimentation: regulatory sandboxes can accelerate evidence‑based design without imposing one‑size‑fits‑all mandates.
– Harmonise disclosure and audit standards: comparable audits across jurisdictions reduce regulatory arbitrage.
– Support public‑interest infrastructure: fund experiments with civic feeds, media literacy, and cross‑national monitoring of information operations.
Caveats, limitations and the path forward
No algorithm is perfectly neutral. Training data and human values inevitably bias machine learning systems; optimizing fairness for one group can reduce accuracy for another. The Copenhagen study is controlled and instructive, but real platforms operate at scales and under adversarial conditions that complicate straightforward translation. Alternative feed designs can also be manipulated — actors who benefit from chaos will adapt. The solution will be iterative: evidence gathering, policy calibration and continuous audit.
Conclusion — a strategic pivot point
The Copenhagen study adds empirical weight to a strategic imperative that has long been debated: algorithm design is not a purely technical or product choice — it is a governance decision with material consequences for markets, politics and society. For executives and investors, the question is not whether algorithmic curation matters, but how to manage the trade‑offs between engagement‑driven growth and the long‑term costs of polarisation and misinformation. For policymakers, the study signals that targeted regulation and incentives can feasibly nudge platforms toward healthier information ecologies without eliminating innovation. The near term will reward organisations that treat algorithmic governance as a core strategic capability — balancing user choice, transparency, compliance and new monetization models — because the alternatives are asymmetric: regulatory mandates, market repricing, and loss of public trust.