The second day of Media Party Barcelona, held at the Auditorium of La Caixa Forum, revolved around a single question approached from different angles: how does journalism survive — and reinvent itself — when the mediation between newsrooms and their audiences no longer depends only on people, but on artificial intelligence systems. Here’s a recap of the core ideas from each of the day’s main speakers.
Nic Newman: the end of referral traffic and the rise of personalities
The Reuters Institute researcher opened the day with data from the Digital News Report: for the first time, social media and video networks have overtaken television and news websites as the leading source of news globally, and no age group now prefers news websites as their main source. Search traffic to publishers has dropped by a third in the past year due to AI-generated summaries in search engines, and three-quarters of publishers expect agentic AI — assistants like Pulse or Huxe that deliver personalized briefings — to deepen that loss of direct visits.
Faced with this landscape, Newman described a two-pronged editorial strategy: concentrate resources on what AI cannot easily replicate — original investigations, contextual analysis, human stories, community — while scaling back generic or service-oriented content, which is increasingly commoditized. At the same time, many outlets are integrating AI internally (backend automation, transcription, tagging), though editors themselves admit the impact so far has been limited rather than transformational. He closed on the rise of content creators: while the general public sees them as less trustworthy than traditional media, those who already follow them rate them higher on every attribute, including trust. The dominant publisher response is clear: turn journalists into figures closer to the creator format.
Clara Jiménez Cruz: who watches those who now watch over public conversation
The co-founder and CEO of Maldita.es argued that the adoption of generative AI — twice as fast as any previous technology — coincides with platforms rolling back their own protections against disinformation. When AI models act as the new gatekeepers, problems emerge on three fronts: in training (prevalence of a claim doesn’t mean it’s real), in presentation (generative AI sounds authoritative and rarely expresses doubt), and in personalization (sycophancy toward the user, and bias by language or education level, as documented even by MIT).
Jiménez Cruz walked through recent Maldita investigations: a network of accounts that monetized 89 million views on TikTok with AI-generated protest videos across 18 countries, violating both the platform’s own rules and the EU’s Digital Services Act; 67,000 fraudulent ads on Meta legitimized by the verification badge; YouTube channels with 21 million subscribers monetizing climate disinformation despite the platform’s explicit ban; and an investigation into TikTok that documented more than 1,300 predator accounts, with the platform leaving 96% of reported accounts active. Several of these cases are now in the hands of Spain’s National Police, the Public Prosecutor’s Office, and the European Commission. Her conclusion: if journalism’s mission is to watch over those who organize public life, it cannot leave out the companies that today organize public conversation.
Elena Sanz: when science journalism becomes public policy
The editorial director of The Conversation Spain presented the “Digital Wellbeing and Minors” project as proof that journalistic impact can be measured in social change, not just clicks. More than 70 articles written by nearly 100 scientists and experts, with over 315,000 reads — more than half from outside Spain — were paired with 14 themed newsletters and nine in-person events in cities such as Madrid, Zaragoza, Valencia, and Barcelona, where specialists in psychiatry, education, law, and technology engaged directly with young people and families.
That work didn’t stay confined to the newsroom: The Conversation joined the Control Z platform, which brings together medical and scientific societies to promote responsible technology use among minors, and contributed to the legislative push behind Spain’s Law for the Protection of Minors in Digital Environments, expected to pass in Congress this autumn. The outlet is also finalizing digital-empowerment materials for children as young as six. Sanz offered the model as one replicable by any newsroom looking to stop being a spectator of its environment and start being an agent of social change.
Dmitry Shishkin: do agents have needs?
The independent media advisor closed the day’s main talks with a warning: AI agents are already retrieving, summarizing, and delivering journalism before a person ever reaches the original article — and this is not a future scenario. Agents don’t have needs of their own, but they inherit human ones, and in that process of compression and transformation, the signals that make journalism trustworthy get stripped away: attribution, uncertainty, source hierarchy, editorial intent.
Shishkin introduced his “Agent Readiness Layer” framework, which sits beneath the User Needs Model without replacing it, and puts content through eight tests: whether it can be retrieved, verified, transformed, credited, acted on, compared, unbundled, and situated in time. His question to newsrooms was direct: how much of your journalism would pass all eight tests today? The session closed with a hands-on exercise in which attendees audited their own content and designed a workflow change to implement next quarter.
Tomorrow, the conversation moves from talks to code. Media Party Barcelona wraps up its European edition with the hackathon at BIT Habitat, where teams of journalists, developers, data scientists, and designers will build open-source prototypes around workflow automation, the post-search publisher economy, and information integrity. Full details and registration here.
You can already start pitching ideas and forming teams on HackDash.
