Beyond the hype: Three cultural traps blocking AI adoption in Latin American newsrooms
Between 2023 and 2025, I worked on AI adoption with three Argentine newsrooms. The tools evolved. The need for cultural change remained constant. This is what I learned in the process.
Most articles about AI in newsrooms tend to come from the Global North. Case studies from major media outlets in the United States or Europe. They’re useful, it’s true, but they often miss something crucial. They describe what’s possible with abundant resources, large teams, and technical infrastructure that most newsrooms in the world will never have.
This is not that story.
Between 2023 and 2025, I worked with three Argentine newsrooms of different scales on AI adoption processes and delivered training to another twenty Latin American media outlets. All marked by tight budgets, diverse teams, and constant economic uncertainty. Budget constraints functioned as a filter that revealed what really matters.
While major newsrooms were still running cautious pilots, these outlets were already iterating, failing, adapting, and learning. They succeeded because they had to get the strategy right from the start. The margin for error was narrow.
In those three newsrooms, the same pattern emerged, regardless of scale or budget. Tools evolved rapidly, but organizational culture moved slowly. That mismatch marked the difference between real technology adoption and pilots that faded over time.
Three newsrooms, three scales, one common pattern
Todo Jujuy (2023): Automating to free up time and return to local coverage
Todo Jujuy, a regional outlet in Jujuy province, faced a typical dilemma for local media. Its small team spent hours producing service content (weather, traffic, sports results, and national news) while some of the local stories that truly mattered to their community went uncovered.
We defined a clear objective from the start. Automate the repetitive so journalists could focus on what other outlets weren’t covering. We moved forward gradually, with human oversight at each stage, and developed clear internal guidelines on when and how to use AI.
Results came quickly. Within a year, Todo Jujuy went from zero AI-assisted articles to more than 500 per month. Page views nearly doubled, from 1.3 million to 2.4 million, while the number of users tripled. But more important than the numbers was the editorial shift. The team reclaimed time to produce higher-value journalistic content. AI worked because it responded to a clear editorial need, not because it was adopted due to hype pressure.
Read the full process documentation in this IJNET article.
0221 (2024): Two-level work and tools designed by journalists
With 0221, a local digital outlet in La Plata, we approached integration on two parallel levels. At the strategic level, we worked with management and senior editors to define a shared vision, establish priorities, and develop responsible use policies. At the operational level, the focus was on journalists.
The difference lay in the methodology. Instead of delivering generic tools, we facilitated workshops where journalists themselves learned to design effective prompts and create custom models for their specific workflows. They developed tailored assistants for tasks like interpreting official bulletins, analyzing police reports, and generating supplementary content drafts.
An internal survey revealed that 90% of the team valued the experience positively, citing ease of use and reduction of repetitive tasks. 70% reported that AI improved their creativity or inspired new ideas. Even more revealing was the production data. During the first quarter of 2025, “production stories” (higher-value articles developed in a single day) increased 129% compared to the same period in 2024, while total content volume grew just 2.2%. The key was that journalists designed their own tools rather than being limited to using predefined solutions.
Read the full process documentation in this Audiencers article.
Clarín (2025): An opportunity to rethink formats and audiences
With Clarín, one of Argentina’s leading national dailies, the approach was different. We didn’t design a complete integration but rather an intensive exploration workshop with more than 50 journalists and editors from different sections.
We structured the workshop around three conceptual frameworks. First, Dmitry Shishkin’s user needs model, which identifies why people consume news beyond just staying informed. Second, an analysis of changes in consumption habits (fragmented access, limited attention, modular narratives). Third, AI as a cross-cutting technology that can assist multiple stages of the editorial process.
Teams worked on concrete prototypes. An automated meeting summarizer to improve daily editorial planning. A system for rewriting food articles for key dates. A connector for developing stories with background from the CMS. An automatic reader of technical reports for the Auto section. The value lay in using AI to rethink the journalistic product based on audience needs, not in automating for automation’s sake.
Read the full process documentation in this INMA article.
Three cultural traps blocking adoption
The three cases mentioned represented completely different contexts. A provincial regional outlet, a hyperlocal digital with agile structures, and a national daily with complex infrastructure. The differences in scale, budget, and audience were enormous, but something remained constant.
Three cultural barriers emerged repeatedly, regardless of scale or resources. They weren’t technical obstacles solved by buying better tools or hiring more staff. And paradoxically, they’re not usually at the center of conversations about AI in media, yet they proved more determinant in sustaining an implementation than any budgetary or technological limitation.
Overlooking them kills pilots and generates tension scenarios that, without intervention, can become structural and dangerous for the industry. Because ultimately, AI adoption isn’t about tools. It’s about people, teams, and organizational culture.
The generational paradox
In many teams, a persistent gap appears. Younger journalists feel comfortable with tools but are still building editorial judgment. More experienced journalists have that judgment but often lack adequate support and training to incorporate these technologies into their daily practice. Between both groups, exchange is scarce.
Automation affects tasks that historically functioned as learning spaces for the craft. Writing briefs, covering results, or transcribing conferences allowed journalists to acquire rhythm, prioritization, and editorial discipline. Today those tasks can be resolved differently, but new training paths aren’t always designed.
The processes that worked best were those that promoted work between diverse profiles, without assuming incompetence by age. It’s not exclusively about the generational aspect. When experience actively participates in tool design, adoption becomes more solid.
Cognitive sedentarism
Another frequent trap is adopting closed solutions without deep understanding of how they work. Initial comfort often leads to dependence and rigidity.
When teams don’t understand how tools operate, they lose adaptability. Faced with a context change or editorial needs shift, the response is often abandonment of the technology, accompanied by the idea that “it doesn’t work for us.”
The most positive experiences incorporated controlled experimentation routines. Simple tools, designed internally, generated greater ownership than sophisticated solutions imposed from outside. The value lay in keeping editorial thinking about technology active.
Survivorship bias
The media industry tends to observe only the cases that succeeded. Large newsrooms in the Global North function as permanent references, even when their starting conditions are difficult to replicate.
The analysis excludes processes that didn’t prosper, abandoned pilots, and integrations that never achieved sustained use. Yet those failures contain key learnings about limits, focus errors, and implementation problems. Survivorship bias.
In the cases worked in Argentina, resource scarcity forced different questions from the start. The focus was on solving concrete problems with available means. That logic allowed progress with greater clarity and less dependence on external models to avoid the survivorship bias trap.
What actually worked (and why): five transferable principles
Beyond cultural traps, some positive patterns repeated in cases where adoption worked. They’re not universal recipes, but principles that can adapt to different contexts and scales.
Training as infrastructure, not as an isolated event
Training can’t be a two-hour workshop. In Todo Jujuy, 0221, and Clarín, training was continuous and staged. Initial workshops to understand basic concepts, practical sessions to design tools, consultation spaces to resolve doubts in real time, and follow-up to adjust when problems appeared.
Media outlets that commit to sustained training programs aren’t simply adding technical skills. They’re generating conditions for teams to adapt and lead change. Training understood as infrastructure, not as a one-time expense, marks the difference between superficial adoption and real transformation.
Safe spaces to experiment
Experimentation requires permission to make mistakes. In 0221, we created a controlled environment where teams could test, fail, and learn without errors having editorial or employment consequences. That freedom was key for them to identify not only AI’s potential but also its limits.
The best way to understand what shouldn’t be done with this technology is to test it in real contexts but with safety nets. Safe spaces build trust faster than any best practices manual. Fear blocks innovation. Permission to fail enables it.
Participatory design by journalists
In my experience, the best and most sustained results emerge when journalists themselves conceive and adopt personalized tools according to real production and coverage needs. In 0221 and Clarín, those who designed their own solutions used them more consistently than those who received predefined tools.
Ownership is critical. When the team designs, it understands. When it understands, it adapts. When it adapts, it sustains. Even simple solutions created internally generate more ownership than sophisticated platforms imposed from outside. Transferring control from vendor to journalist is the step many implementations never take, which is why they fail.
Process before tool
Prioritizing “what for” before “what with” helps avoid technological fetishism. In all three cases, the starting point was never “what AI tool should we adopt?” but rather “what problem do we need to solve?”
AI understood as part of cultural change, not as a fad or gadget, has a better chance of genuinely integrating. The right questions at the process start make all the difference. What we do, who we do it for, and how we could do it more effectively without losing quality. Technology is the means, never the end.
Metrics and constant follow-up
None of the three cases limited itself to implementing and waiting. There was constant measurement. In Todo Jujuy, we tracked automated content volume, audience growth, and time freed for higher-value production. In 0221, we conducted internal surveys on usage perception, impact on creativity, and team satisfaction.
If you don’t measure impact and don’t return to adjust based on that data, the project loses momentum. Evaluation isn’t bureaucracy. It’s the only way to know if what you’re doing works or needs to change course. Implementations that don’t measure end up navigating blindly, not knowing if they’re advancing or stagnating.
Ethical and responsible use policies
Defining clear guidelines for transparency, human review, and equity from the start establishes a framework of trust and internal legitimacy for teams. In Todo Jujuy and 0221, guidelines weren’t a final document but a living process built parallel to adoption.
Policies served a dual function. Internally, they established clear responsibilities about who validates what, when AI intervenes and when it doesn’t. Externally, they made the process transparent to the audience. Todo Jujuy published the first AI guidelines from an Argentine outlet. 0221 adapted them to its specific context. Both documents evolved with practice.
Guidelines that work are those that are discussed, reviewed, and adjusted. A static PDF filed in a folder doesn’t build governance. Constant conversation about limits, uses, and responsibilities does.
Generating uncomfortable pauses
Pausing the logic of immediate content allows space to think and rethink. In contexts where urgency dictates rhythm, that pause generates discomfort. But it was the starting point for observing processes with distance and regaining control over tools.
All three cases required moments to stop. In-person workshops where the team left daily production to experiment. Strategic meetings where executives debated priorities without publish pressure. Evaluation sessions where what worked and didn’t was reviewed.
It’s a mistake to view those pauses as time wasted. They were investments in clarity. Newsrooms that don’t stop to think end up adopting technology reactively, without direction or criteria. The uncomfortable pause is the price of conscious implementation.
Human-centered AI
We spend a lot of time talking about machines, but we need to talk about the humans who will use those machines. Integrating artificial intelligence from a human perspective allows maintaining focus on journalistic purpose, prioritizing creativity, ethics, and editorial judgment over automation.
In all three cases, human oversight was non-negotiable. Nothing generated or assisted by AI was published without journalist validation. Workflows were designed with human checkpoints from the start, not as later additions. Technology assisted, but final decisions always remained in human hands.
The human approach also meant stopping anthropomorphizing AI. It doesn’t “lie” or “make mistakes” in the human sense. It predicts text based on statistical patterns. Understanding that changes how processes are designed. It’s not about trusting or distrusting AI, but about structuring flows that assume its technical characteristics and keep human responsibility at the center.
A new relationship with time: AI’s true potential
Artificial intelligence’s true potential in media doesn’t lie in automation per se. In contexts marked by constant acceleration and information saturation, this technology’s most valuable contribution isn’t its capacity to make everything faster, but its potential to enable a new relationship with time.
A relationship that allows newsrooms to recover spaces for analysis, look at processes with distance, and prioritize more reflective editorial decisions. In Todo Jujuy, automating the repetitive freed time to cover local stories that previously went unattended. In 0221, personalized tools allowed journalists to produce higher-value content without increasing total volume. In Clarín, rethinking formats with AI’s help opened strategic questions about what audiences really need.
Each newsroom has its own rhythm, culture, and urgencies. There’s no universal playbook. But there’s a cross-cutting question we should all ask ourselves. What tasks could we solve better if we had custom-designed editorial assistance. The starting point isn’t technology. It’s judgment. What we do, who we do it for, and how we could do it more effectively without losing quality.
The Global South’s budget constraints, far from being obstacles, can function as filters that force asking the right questions before better-resourced newsrooms do. That agility, that need to get strategy right from the start, can become a competitive advantage if sustained with training, experimentation, and open documentation.
Something deeper is changing. AI is no longer just a set of tools. It’s becoming a new interface for accessing knowledge. Conversational chatbots are reconfiguring how people get informed. Audiences no longer search, they converse. They no longer expect to find an article. They expect someone, or something, to answer them. Journalism finds itself in that transition today. And it’s not a technological transition. It’s structural.
Therefore, more than adopting technologies, what we need is to build criteria. And more than seeking definitive solutions, what a process like this leaves are open questions. What can we do better. How do we make it clearer, more useful, more human.
