The debate over whether AI belongs in newsrooms is over. The question now is how to use it efficiently and safely to generate content — and to serve as a co-pilot for reporters and editors.

That was the focus of a recent INMA Webinar, Artificial Intelligence in Newsrooms: From Experiment to Real Impact. The discussion centered on integrating AI without losing editorial focus and on how far to go with a tool many view as a giant that soon will decide topics, angles, and even write stories without human oversight.

Álvaro Liuzzi, an Argentine journalist and digital media consultant, offered a behind-the-scenes look at AI adoption in Latin American outlets. His work with Clarín, TodoJujuy, and 0221 provides a road map for newsrooms exploring similar paths.

From “what” to “why”

AI has the potential to expand not only the productivity but also the creativity of media, Liuzzi said. To explain its evolution, he borrowed from journalism’s 5W rule: what, when, how, where, and why.

For him, 2022 through 2024 marked the “what” phase, when newsrooms were testing the potential — and the pitfalls — of adding AI to their workflow.

The next phase, starting in 2025, is about “why” and “how.” That’s when outlets will need a more strategic, reflective approach to AI’s role in journalism, Liuzzi said.

AI is hardly new. It’s been in development for more than 70 years. Milestones include Alan Turing’s computer in 1950, the first symbolic programming language in 1958, and popular culture’s vision of autonomous machines in Hollywood productions such as 2001: A Space Odyssey and Blade Runner.

There was IBM’s Deep Blue defeating chess champion Garry Kasparov in 1996, the rise of voice assistants, and the conversational AI tools dominating today — ChatGPT, DeepSeek, Claude, and others.

ChatGPT’s debut in 2021 was a turning point, giving users the sense of speaking with another human. Generative AI is the result of humanity’s unique ability to create increasingly complex tools, a process anthropologists call “cumulative cultural evolution,” Liuzzi said.

That ability, he said, has been supercharged by three decades of digitalization, from PDFs and Web sites to the audio and video libraries that power Spotify and Netflix. For the media, AI is now a creativity catalyst — and a production engine on a scale the industry has never seen.

The power of algorithms

Algorithms have become the gatekeepers of information, Liuzzi said: “Never in human history has a dictator, president, or king had the power to concentrate and distribute information the way algorithms do.”

Everything read, watched, or listened to online today is mediated by an algorithm, he adds. “What matters less is what algorithms are and more what humans represent to them: a chain of clicks.”

Platforms such as Spotify, Netflix, and YouTube optimise those clicks by recommending endless loops of content. However, while effective in entertainment, news consumption is another matter. Algorithms push people toward increasingly extreme content by amplifying what they’ve already clicked.

Combine that with confirmation bias, filter bubbles, and polarisation in some countries, Liuzzi said, and it becomes “a dangerous issue for democracies.”

Three paths for news outlets

AI is now embedded in nearly every newsroom task: headline writing, summarising, translation, social media copy, video subtitling, podcast scripts, image generation, grammar checks, and graphics.

However, media companies are responding differently.

Some, like the Associated Press, have struck deals with AI firms. In 2023, AP partnered with OpenAI, giving it access to a 40-year archive while AP gained the ability to test generative AI in its news services.

Others, including The New York Times, have chosen to block access, even filing lawsuits against OpenAI. Their concern: Audiences will stop visiting Web sites and rely solely on AI tools for information, slashing traffic to news portals.

A third group, such as The Washington Post, has begun institutionalising AI with dedicated editorial roles. In Latin America, outlets are creating similar positions focused exclusively on AI adoption.

Adoption in the global south

A Thomson Reuters Foundation study covering Africa, South Asia, and Latin America found 81% of newsrooms already use AI daily, although only 13% follow formal policies. Nearly 60% of journalists learned AI tools on their own.

“If we want to work with AI in newsrooms, it’s essential to train journalists and develop usage policies,” Liuzzi said.

Ethical dilemmas are the top concern, cited by 57% of journalists using AI. Many also fear that global AI models fail to reflect local realities.

The Tow Center’s AI in the News report, based on 134 interviews with outlets in the United States, Germany, and the United Kingdom, adds another warning: Growing reliance on AI could compromise editorial independence, align media more closely with tech firms, and stifle innovation.

The SEO model collapses

AI is also gutting a model that defined digital media for years: SEO.

Liuzzi noted Business Insider has lost half its organic traffic since 2022, coinciding with the rise of ChatGPT. The Washington Post is down 40%, The Wall Street Journal, 25%, and The Huffington Post has fallen off the charts altogether.

For years, sites wrote for algorithms rather than audiences, pumping out volume and relying on clickbait: “The model is collapsing.”

“Traffic isn’t coming back. Audiences no longer search — they converse,” Liuzzi said. “The challenge isn’t technological, it’s structural. We must stop propping up the old and start building the new.”

Four waves of AI

Liuzzi identified four stages of AI adoption:

1. Invisible AI (2010–2019): algorithms shaping distribution and monetisation through Facebook, YouTube, programmatic ads, and automated SEO.

2. AI as a tool (2020–2023): assisting with summaries, topic suggestions, simple reports, and graphics. Productivity rose, but structural integration was low.

3. AI as a producer (today): moving from accessory to content creator, raising concerns over ethics, copyright, trust, and the loss of jobs.

4. AI in full workflows (late 2025 onward): detecting relevant topics, making autonomous editorial decisions, distributing content, and redefining newsroom structures as well as the human-machine relationship.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .