“Artificial intelligence is no longer knocking on the door of journalism—it’s already inside the newsroom”. That was the resounding message during the recent “AI Trends in Journalism 2025” webinar, part of WAN-IFRA’s ongoing “AI in Media” initiative. The event brought together media innovators and AI strategists to discuss the evolving relationship between artificial intelligence and journalism, and the consensus was clear: AI is no longer a luxury or a side project. It is now a strategic necessity.
Leading the discussion was Ezra Eeman, WAN-IFRA’s AI Expert and Director of Strategy & Innovation at Dutch public broadcaster NPO, who delivered a powerful reminder of the stakes. “AI is no longer optional for newsrooms—it’s existential,” he said. His statement framed the webinar’s urgency as speakers unpacked how AI is reshaping everything from newsroom workflows to public trust.
Drawing on insights from a recent study tour to San Francisco—where Eeman and a small group met with OpenAI, Perplexity, and top venture capital firms—he outlined a future where traditional search traffic and social media referrals are fading. In their place, AI interfaces and personalized agents are emerging as the new gatekeepers of information. Publishers, he said, must now prepare for a “post-link” reality where users no longer click through to homepages but instead receive summaries and snippets through AI agents. This shift, while convenient for users, poses a direct threat to journalism’s economic model by bypassing ads, subscriptions, and branded experiences.
Adding to the complexity is the fragility of the AI ecosystem itself. The market remains volatile, as seen in the DeepSeek-induced dip earlier this year, and the monetization of content through AI remains immature. Few publishers have forged meaningful partnerships with AI platforms, and experiments with “machine-to-machine” monetization—where bots pay to access content—are still largely theoretical. Despite heavy investment from tech giants, a sustainable economic model remains elusive, leaving publishers scrambling for clarity.
Yet amid the uncertainty, there are promising paths forward—especially in open-source innovation. Nikita Roy, founder of the Newsroom Robots Lab and host of the Newsroom Robots podcast, argued that journalism is lagging behind other sectors in AI adoption. Tools like Langchain are common in finance and insurance but virtually unknown in most newsrooms. “There’s a huge missed opportunity there,” Roy said. “Many journalists aren’t yet equipped with the skills to explore these tools. But every journalist needs to understand AI—it’s as transformative as when computers first arrived in newsrooms.”
Roy noted that many news organizations remain stuck in an experimental “playground phase,” using AI in isolated ways without incorporating it into broader editorial strategies. She emphasized the need for a clear return on investment to move experimentation into implementation. Florent Daudens, Press Lead at Hugging Face, echoed this, stressing that successful AI integration requires newsroom-wide buy-in, not just individual tinkering. “AI projects don’t fail because of tech. They fail because no one owns them,” he said.
Both Roy and Daudens emphasized the power of open-source models. Unlike proprietary systems from Big Tech, open-source AI allows newsrooms to customize tools to suit their editorial values and ethical standards. Roy highlighted the case of iTromsø, a small Norwegian newsroom that built its own AI assistant, DJINN, to scan municipal documents. The system, developed with editorial input, now flags important items for review and helped two interns break five front-page stories in its first week. This, she said, is the kind of innovation that happens when journalism leads, rather than follows, in the AI space.
Eeman, Roy, and Daudens all pointed to the need for deeper newsroom understanding of AI models—not just using tools, but knowing how they work, what data they rely on, and where their limitations lie. As AI agents become more capable of summarizing and delivering news independently, the nature of content itself is being redefined. Roy demonstrated how ChatGPT’s Research Mode can extract structured databases from public content, turning stories into machine-readable data. In this emerging environment, publishers who structure their content accordingly will thrive, while others risk becoming invisible.
This shift also raises new challenges around trust, verification, and legal ownership. A recent BBC study cited by Daudens found that more than half of AI-generated content contained factual or ethical errors. The solution, he argued, lies in building systems on high-quality, verified datasets—a call to action for newsrooms to maintain their editorial rigor, even in automation. Roy highlighted the Reuters v. Ross Intelligence lawsuit as a case that could reshape how content is used to train AI, but both she and Daudens agreed that legal victories won’t be enough. The real power lies in proactive engagement, collaborative partnerships, and shared infrastructure like the prompt libraries being built at the Financial Times.
In a striking demonstration of AI’s evolving potential, the panel showcased real-world applications of AI agents that can browse the web, retrieve stories, and generate summaries without human input. Roy even built an AI clone of herself to answer workshop questions, while Daudens created a custom news brief generator using a large language model and plain language prompts. These examples underscored just how accessible and transformative AI technology has become—even for non-developers.
“AI is a choice,” Roy concluded. “You either build with it, or someone else builds over you.” That sentiment captured the spirit of the discussion: Journalism is at a crossroads. The decisions newsrooms make today—about training, strategy, transparency, and innovation—will define their relevance in an increasingly AI-mediated information world.
For students and aspiring journalists, the message is clear: learning AI is not just a tech skill, it’s a journalistic responsibility.
