Professor Stephan Lewandowsky is a cognitive scientist at the University of Bristol (UK). His research explores people's responses to mis- or disinformation and propaganda, why people reject well-established scientific facts, such as climate change or the effectiveness of vaccinations, and the potential conflict between the architecture of our online information ecosystem and democracy. He co-authored the report Fractured Reality – how democracy can win the global struggle over the information space, commissioned by the European Commission's Joint Research Center (JRC).

---

How would you describe the attention economy, and how is it shaping the way people perceive and interpret reality? Can you provide an example?

Stephan Lewandowsky: The attention economy is a system in which platforms, in collaboration with creators and advertisers, deliberately design content to capture, retain, and hence monetise human attention. The business model of most large platforms therefore rests on 'engagement': the longer you stay on the platform, the more personalised advertising can be shown to you and the greater the platform's revenue. Unfortunately, this business model is, at best, indifferent to truth.

Because human attention is drawn to information that is negative, emotional, and conflictual, the engagement-based business model will likely favour precisely the content that corrodes democratic discourse. Misinformation evokes more outrage than trustworthy content, and outrage is algorithmically successful.

The engagement model will also lead to fragmentation of perceived realities because pre-existing differences in attitudes will be exacerbated by the algorithms and will drive people further apart over time.

This was illustrated in a recent field experiment on X, in which researchers redesigned users' feeds to down-rank (or up-rank) content expressing antidemocratic attitudes and partisan animosity. When polarising content was down-ranked, participants' hostility towards the other side dropped by an amount comparable to reversing roughly three years of rising affective polarisation in the US. That tells you that platform algorithms are actively shaping how people perceive their political opponents.

---

To which extent are algorithms to blame for the spread of misinformation and polarisation and to what extent do our cognitive biases play a role?

Stephan Lewandowsky: We need to avoid creating a false dichotomy between humans and algorithms – the problem arises from their interaction. Confirmation bias, our tendency to seek and favour belief-consistent information, has always existed. What has changed is that algorithms now detect and cater to that tendency at scale, multiplying its effects.

The best evidence comes from a mega-study of news consumption by around 208 million Facebook users: as a rough guide, about a third of ideological segregation was attributable to algorithmic curation, and two thirds to people preferentially engaging with attitude-aligned content. So yes, human psychology does much of the work – but people make those choices inside an architecture designed by platforms, without their knowledge or input, and optimised for engagement rather than accuracy.

This is why our report argues that blaming individuals for being misinformed or polarised is a misdiagnosis. The problems are systemic, and they require systemic solutions rather than exhortations for citizens to simply "be more critical."

---

Today's social media platforms rely heavily on content personalisation and engagement-driven business models, which critics argue can encourage addictive behaviours, particularly among young people. What can the EU do to foster viable alternatives?

Stephan Lewandowsky: First, enforce what already exists – the DSA, DMA, and now also the AI act – and implement the European Democracy Shield. Beyond that, our report highlights several levers. One is restoring user agency through interoperability and "in situ" data rights: users should be able to choose their recommender algorithm and other attributes of the platform from a menu of providers, rather than being locked into a single proprietary feed. That would open competition on quality rather than addictiveness.

Second, the EU can shift incentives through a progressive digital advertising tax – an idea supported by Nobel laureates Acemoglu and Johnson – nudging platforms towards subscription or quality-based models. The tax could even be scaled to the concentration of harmful content in a platform's output, much as we tax industrial pollution.

Third, Europe should invest in decentralised, value-based alternatives – the Fediverse, Eurosky and similar infrastructure – including public early adoption by administrations.

The addiction concern is real: about 31% of social media use reflects habit formation with some attributes of addiction, and studies show many people would actually pay to have social media collectively switched off – a classic sign of lock-in rather than genuine value. That is a market failure, and market failures justify public intervention, particularly to protect young people.

---

Fact-checking and debunking have not always proven effective in changing someone's misinformed beliefs. What is pre-bunking, and what potential do you see in this approach?

Stephan Lewandowsky: Pre-bunking, or psychological inoculation, follows the vaccination logic: instead of correcting falsehoods after the fact, you forewarn people that they may be misled and expose them to a weakened dose of the manipulation – for example, the tactics used by disinformers, such as the tobacco industry, or fake experts, false dichotomies, or emotional exploitation. Having explained and refuted the technique in a safe setting (such as a brief educational video), people become more resistant when they encounter misleading argumentation in the wild.

The great advantage of pre-bunking over debunking is that it does not chase individual falsehoods one by one – a hopeless task given the sheer volume of misleading content. Because it targets rhetorical techniques rather than specific claims, it generalises to falsehoods that have not even been invented yet. The evidence base is now substantial: a recent meta-analysis of 33 inoculation interventions confirmed their effectiveness, and inoculation can be delivered at scale through short videos and games.

Related article