Integrating AI into daily editorial workflows at Clarín

Over the course of four sessions, more than 50 journalists and editors from Clarín explored how to apply artificial intelligence in their workflows, experimenting with new formats and designing practical solutions for daily use, with sessions built around 3 frameworks: 
- A user-needs-driven approach
- An analysis of shifting consumption habits
- A strategic approach to AI as a cross-functional technology

Why Understanding the Architecture of New Formats Matters Before Using Them

The first section of the workshop focused on understanding the core principles behind today’s digital formats — before diving into concrete exercises. We analyzed the information architecture of articles, modular structures, and the layered logic of narrative design. The goal was to rethink how information can be organized to better guide readers through content: what they need to know first, what they might explore next, and what they might want to save for later, depending on their context.

Beyond the structural side, the emphasis was on the editorial reasoning behind every decision.

A key reference in this phase was the User Needs Model developed by Dmitry Shishkin during his time at BBC World Service. The model outlines eight core reasons people consume news: to stay informed, to learn something new, to solve a practical problem, to be inspired, to be entertained, to find community, to affirm their identity, or to follow a developing story. Understanding these motivations shifts the way content is created. It’s no longer just about producing a clear and timely article — it’s about addressing a specific user need, at the right time, on the right platform.

In this context, artificial intelligence was approached as a tool to improve the newsroom’s ability to meet audience needs. The goal was not to automate for the sake of automation, but to use AI to create content that is more useful, accessible, and personalized.

The guiding questions during this phase were deeply editorial in nature:

• Who is this content intended for?
• What specific need is it trying to address?
• How does that need change if the story appears on the homepage, on social media, or is shared via WhatsApp?

In a media ecosystem where access to information is no longer linear, and where large language models are beginning to mediate that access, thinking in terms of user needs is no longer a competitive advantage. It’s a baseline requirement for sustaining editorial relevance over time.

> Implement user needs in your newsroom: email course for publishers

How We Selected and Applied AI Tools with Editorial Criteria

The hands-on section of the workshop focused on demystifying how generative AI works and offering practical tools for its editorial use. A key part of the work centered on designing effective prompts — not as magical formulas, but as strategic instructions that help guide the model to produce outputs aligned with Clarín’s specific editorial standards.

Participants learned how to write prompts tailored to different content formats, such as explainers, timelines, and listicles, and how to adapt those prompts depending on the type of informational need they wanted to address. The sessions also covered the risks of AI “hallucinations” and how to reduce errors by crafting clearer instructions, selecting relevant source documents, referencing base texts, and specifying tone, style, or length.

A small but powerful set of accessible tools was also introduced, suited to different newsroom roles:

• ChatGPT: The focus was on the use of custom GPTs, trained with internal documents and specific instructions for editorial and operational tasks within the newsroom.
• NotebookLM: This tool was used to analyze large volumes of documents and turn them into interactive, searchable knowledge bases. Its application as a “smart reader” for long materials — such as government bulletins or public reports — was demonstrated, along with its ability to generate synthetic podcasts from that content.
• Inspiration Materials: A practical reference document was shared with examples of how these tools can be integrated into real workflows without disrupting editorial dynamics, helping to save time and free up capacity for more analytical or creative work.

This part of the workshop also addressed the ethical dimension of AI use. We discussed acceptable use cases, transparency standards, and the importance of maintaining human oversight at every stage of the process.

Designing Custom Assistants: Using AI to Solve Real Editorial Tasks

In the third stage of the workshop, each group was guided through the process of identifying a specific workflow from their daily routine where they could experiment with artificial intelligence tools. The starting point was simple: find a real, everyday problem — something time-consuming or poorly solved — and explore how a custom tool could help improve it.

Rather than working with generic platforms, the experimentation was driven by a set of editorial questions designed to guide the process:

• Where are we losing time that could be better spent elsewhere?
• What tasks are repetitive but necessary?
• Which steps could be automated without sacrificing control or quality?

With that mindset, the teams designed custom assistants tailored to their own workflows. These assistants were built from collaboratively developed instructions. The aim was to experiment, apply ideas in practice, and learn by doing. The approach remained iterative and open, encouraging each team to take on the role of creator — not just user.

Technology was framed as a partner in solving concrete editorial needs. Each prototype was developed with a clear intention, focused on improving some aspect of daily newsroom work. While the solutions varied in scope and style, they all shared a common goal: to enhance capabilities, reduce tedious tasks, and free up time for more meaningful, human-driven contributions to journalism.

Four AI Solutions Co-Developed with Journalists to Optimize Daily Workflows

One of the most impactful aspects of the workshop was giving journalists the opportunity to identify where AI could be meaningfully applied within their own workflows. Rather than starting from generic tools, each team chose a day-to-day process they wanted to optimize, automate, or enhance. From there, they designed tailored solutions — many of them using custom GPTs trained with specific instructions for each task.

Among all the prototypes developed, four projects stood out for their clarity, practical application, and editorial value:

Summary Meeting Assistant

Identified Problem: Daily editorial meetings bring together journalists from different sections to share what they’re working on. Without a shared structure or standardized documentation, these meetings often become overwhelming for editors, who have to reconstruct the day’s coverage priorities on the fly — identifying overlaps, gaps, and urgent stories without the support of a clear system.

Solution: Journalists created a custom assistant that allows each reporter to submit their planned stories in advance via a simple form. Based on that input, the model generates a structured report that organizes submissions by section, status, and urgency. This gives editors a clear, editable overview of the day’s agenda.

Benefit: It improves daily editorial planning, reduces the operational burden on editors, and creates a digital log of ongoing coverage that’s accessible even outside the meetings.

Repurposing Articles for Key Dates

Identified Problem: Clarín’s Food and Recipes team realized that many valuable articles were being underused. When holidays or seasonal events come up — such as national celebrations or traditional festivities — reframing existing content to match the occasion takes time and manual effort. As a result, useful material often goes unused instead of being repurposed.

Solution: The team developed a custom GPT model trained on previously published articles. This assistant helps rewrite existing food-related content by adapting it to specific holidays or events. It maintains the editorial style while adjusting the angle to fit the new context.

Benefit: It boosts productivity, allows for consistent and high-quality content reuse, and reduces production time without compromising editorial tone or quality.

Live News Context Connector Using CMS Archives

Identified Problem: In the Breaking News section, where publication is fast-paced and continuous, journalists often work on developing stories with little time to search for background information. In extended or long-term coverage, this can lead to new articles lacking context, omitting relevant references, or repeating previously published content — adding complexity for the journalist on deadline.

Solution: The team created a custom assistant that analyzes drafts of in-progress articles, identifies key elements such as names, dates, and entities, and then suggests previously published pieces related to the same topic. While the system is not yet fully integrated with the CMS, the model has been structured and documented in a way that makes future integration possible.

Benefit: It strengthens narrative continuity in fragmented coverage, gives reporters easier access to background materials during writing, and lays the foundation for a future integration that could streamline newsroom workflows.

Automated Report Reader for Specialized Coverage

Identified Problem: In the automotive section, journalists regularly analyze monthly reports from ACARA (the Association of Automotive Dealers of the Argentine Republic), which contain detailed data on vehicle registrations, brands, and models. These reports are published in PDF format and require careful reading and manual extraction of figures, delaying the publication process.

Solution: The team developed a custom assistant trained on previous reports and their related articles. The tool allows journalists to upload the latest monthly PDF, automatically extract key data, and generate a structured draft of a news article, ready for final editing.

Benefit: It speeds up the production of technical content, reduces the manual workload involved in reading and summarizing long documents, and enables more timely publication without sacrificing accuracy.

In addition to the prototypes developed during the in-person sessions, the custom AI models also served as a continuity tool — extending the impact of the workshop beyond its scheduled meetings. To achieve this, a personalized assistant was created and trained with all the content covered during the sessions, along with a full list of participants, including their names and the editorial sections in which they work.

This assistant allows journalists to ask questions, revisit key concepts, or explore use cases using natural language. By recognizing who is making the inquiry and what area they work in, the model can deliver contextualized responses tailored to each user’s specific needs and interests. In this way, the shared knowledge from the workshop remains available as an active resource in day-to-day newsroom work.

As a complement, a format translator was also provided to help facilitate the adoption of the narrative styles introduced during the sessions. This tool allows journalists to take a draft or already-published article — such as a standard news report — and convert it into a different narrative format, like a timeline, explainer, or key-point summary. The goal is to speed up experimentation with more engaging and functional structures, all aligned with the editorial style guide created during the workshop, without having to start from scratch.

It’s important to note that, in many cases, the prototypes weren’t designed to be fully finished products by the end of the session. Instead, they were conceived as starting points for future development and implementation. What mattered most was that each team identified a real need within their editorial workflows, envisioned a tangible solution, and took the first steps toward building it. The true value of the process lay not in delivering a polished product, but in helping participants learn how to think about artificial intelligence from a journalistic perspective — focused on real problems, guided by editorial judgment, and oriented toward practical, meaningful adoption.

Final Reflections: AI as an Editorial Ally, Not a Technological Destination

The experience of working with Clarín’s journalists reaffirmed something I had already observed in previous collaborations with newsrooms across the region.

The true potential of artificial intelligence in the media doesn’t lie in automation for its own sake. In a context shaped by constant acceleration and information overload, AI’s greatest value is not its ability to make everything faster, but its power to help reshape our relationship with time. A relationship that allows newsrooms to reclaim space for analysis, revisit their processes with perspective, and prioritize more deliberate editorial decisions.

Every newsroom has its own pace, culture, and pressures. But there’s one question that cuts across them all: what tasks could we handle better if we had an editorial assistant built specifically for our needs? The starting point isn’t technology — it’s judgment. What are we doing, who are we doing it for, and how could we do it more effectively without compromising quality?

Designing a workshop like this one meant first and foremost hitting pause on the logic of constant production and creating space to think. For many newsrooms, that pause can feel uncomfortable. But if there’s one thing this experience made clear, it’s that the goal isn’t to turn journalists into AI experts. It’s to help them regain control over the tools, adapt them to their own context, and make them part of their editorial language.

Something deeper is shifting. AI is no longer just a set of tools. It’s becoming a new interface for accessing knowledge. Conversational chatbots are changing the way people consume information. Audiences are no longer just searching — they’re engaging in dialogue. They’re not waiting to find an article. They expect someone — or something — to respond. That’s the crossroads journalism faces today. And it’s not a technological transition. It’s a structural one.

That’s why what we need isn’t just to adopt new technologies, but to develop strong editorial criteria. And instead of looking for definitive answers, what a process like this leaves behind are better questions. How can we do things better? How can we make them clearer, more useful, more human?

The answers aren’t in the models. They’re in the journalism.