Use Cases in Media:
decisions, processes and lessons from newsrooms that integrated AI

The Paraguayan outlet developed a generative AI chatbot that operates in the Guaraní language to promote access to knowledge and community participation in indigenous communities. Developing a chatbot in Guaraní faces considerable technical challenges, given that the language has limited representation in the training datasets of dominant language models. The project demonstrates that AI can be applied with linguistic and cultural equity criteria, and that media outlets can lead digital inclusion processes for communities historically marginalized from the information ecosystem.

Ojo Público developed a platform that uses AI to automate the production of verified content in Quechua, Aymara and Awajún. The tool translates and synthesizes journalistic content into indigenous languages and converts it into audio formats distributed through radio stations in nine Peruvian regions. The technical challenges are considerable, given that these languages have scarce representation in the training data of dominant models, but the social impact is direct. It points to a direction few are exploring: AI as a tool for linguistic inclusion in journalism, with concrete reach into communities that traditional media does not cover.

The International Centre for Investigative Reporting in Nigeria developed NativeAI, a platform that transcribes and translates journalistic content into local Nigerian languages with the goal of extending the reach of quality journalism beyond English. The system combines automatic transcription with translation models adapted to low-resource languages — a significant technical challenge given that many African local languages have minimal representation in dominant model training data. The case shares with El Surti and Quispe Chequea the same working direction: AI as a tool for delivering verified information to communities historically underrepresented in media coverage, where the linguistic gap is also a gap in access to public information.

Fogo Cruzado developed VANIA, an AI chatbot that allows citizens, journalists and researchers to query real-time data on violence incidents in Brazilian urban areas. The system translates natural language queries into searches on the incident database that Fogo Cruzado keeps updated, democratizing access to information that would otherwise require technical knowledge to analyze. The case combines two dimensions that rarely appear together: public utility in service of civil society and institutional accountability.

Factchequeado developed ChatMigrante, an AI chatbot integrated into WhatsApp to guide Latin American migrants with verified information about their rights, immigration processes and available services. The system combines generative AI with verified journalistic content to answer questions in everyday language about immigration topics such as work permits, asylum, deportation and access to public services. The choice of WhatsApp responds to its high penetration among Latin American migrant communities. The project was recognized at the 2025 Online Journalism Awards as a tool of social impact based on journalistic AI.

Maldita.es deployed three complementary tool layers. A WhatsApp bot allows citizens to send verification queries directly, democratizing access to fact-checking as a service. A narrative detector identifies recurring disinformation patterns on social networks and messaging apps before they escalate. An internal RAG-based assistant allows fact-checkers to query the archive of previous fact-checks in natural language, reducing background research time. The set of tools functions as a systemic integration in which each layer solves a different problem within the same verification workflow.

Newtral developed three proprietary tools. ClaimHunter identifies verifiable claims in speeches and publications. ClaimCheck cross-references those claims against databases of previous verifications. FactFlow AI detects disinformation narratives on Telegram and generates real-time alerts for the fact-checking team. The Telegram case is particularly relevant because false narratives spread faster than any team can monitor manually, and AI speed is structurally necessary in that ecosystem. Newtral represents the fact-checking organization model that combines human journalistic judgment with its own automated detection infrastructure.

Full Fact developed two complementary systems. Government Tracker systematically monitors government promises and generates alerts when it detects discrepancies between what was stated and official data, scaling verification capacity for government commitments without expanding the team. The second system, multimodal detection, identifies and prioritizes false or harmful health claims in audiovisual content, including social media videos, podcasts and television programs. The multimodal dimension is key because a significant portion of health disinformation circulates in audiovisual format, which means systems based solely on text processing have limited reach.

The Associated Press presented a platform that brings together in a single dashboard the verification tools its journalists were using in a scattered way. The platform includes reverse image search, frame-by-frame analysis, geolocation, object recognition, AI-generated text detection and social media monitoring. Among the cases documented during its testing: verification of Texas flood footage, tracing a viral meteor sighting to its original publication, and detecting a video of violence at a football match that actually corresponded to an event in Greece a month earlier. As Aimee Rinehart, AP's senior manager of AI strategy, noted, the tools only work alongside a human and serve as a guide, not a verdict.

Chequeado created an interdisciplinary generative AI lab aimed at experimenting with tools that accelerate and improve fact-checking work, with one distinctive feature that sets it apart from most similar initiatives: it documents and publishes each experiment so that other fact-checking teams and journalists can replicate or adapt the learnings. The lab has published experiments on simplifying complex concepts, writing social media threads, generating video scripts, updating articles with new information, analyzing databases and producing video clips. The approach is incremental and transparent. Each publication describes what was tested, what worked and what didn't, turning the experimentation process into an open resource for the journalistic community in the region. In an ecosystem where most outlets accumulate learnings privately, Chequeado bets on collective intelligence as its development model.

The Brazilian investigative journalism agency developed an internal workflow using ChatGPT to automate the monitoring and classification of mentions of its investigations in other outlets, with human oversight to verify accuracy and quality. The goal was to solve a concrete institutional governance problem: how to measure the real impact of investigative journalism beyond republication counts. The system allows Agência Pública to document more precisely how its investigations circulate, which types of outlets pick them up, and what effects they generate in the public agenda. It is an inward-facing use of AI, to strengthen a journalistic organization's capacity to account for its own work.

The Argentine outlet 0221 developed an AI integration process based on a combination of governance, cross-functional training and practical experimentation within the newsroom. The implementation included internal workshops, development of custom tools, responsible use guidelines and testing spaces where journalists and editors could explore concrete AI applications linked to their own work routines. One central aspect of the process was creating a controlled environment to experiment, make mistakes and document learnings before incorporating the tools into daily operations. The approach allowed adoption to not remain concentrated solely in technical profiles or innovation areas, but to expand across different sections of the outlet, strengthening a more organic integration aligned with the newsroom's real editorial needs.

Finland's largest daily implemented Juttuapuri, a draft-generation system using ChatGPT, Gemini and Anthropic models to produce lower editorial complexity news, with mandatory human review before publication. What distinguishes the case in governance terms is the process that accompanied it. The launch generated internal debate about the limits of AI use in journalistic production, and the outlet opted for a gradual approach with strict editorial controls and a reader transparency policy indicating when an article was AI-assisted. The combination of new technology with documented internal deliberation and explicit public communication is an adoption model that few outlets replicated with the same systematicity.

The Finnish publishing group implemented AI tools for three functions: drafting articles from press releases, analyzing its own podcasts to suggest coverage topics, and an ethical evaluation layer integrated into the workflow that alerts journalists when generated content might violate the outlet's standards. This third component is the case's most relevant element. Alma Media attempted to codify its editorial values within the technological infrastructure, turning journalistic ethics into a system function rather than a declarative policy. The case opens questions worth asking: who defines the ethical model's parameters, how are they updated when standards evolve, and whether a system can internalize the judgment that an experienced editor applies with nuances no algorithm captures.

Estonia's largest media group integrated AI as coordinated infrastructure across all its operations: live translation, transcription, automatic summaries, searchable archive, and editorial and advertising assistants within a single technological ecosystem. The group built a platform where different AI modules share data and coordinate with each other, rather than adopting isolated tools. The adoption process included intensive team training and the definition of clear internal policies on which functions are automatable and which require mandatory human oversight. It is one of the most advanced examples of AI integration as media infrastructure in the Baltic-Scandinavian space, and one of the few where the technology bet was accompanied by an explicit institutional architecture from day one.

Svenska Dagbladet, a Schibsted group daily, implemented a tool that evaluates the journalistic value of its articles and assigns automatic scores to support decisions about which content deserves greater investment or promotion. The system analyzes variables such as informational depth, exclusivity, relevance to the subscriber audience and engagement potential. The case belongs to governance rather than production because what is at stake is editorial. The underlying question is who defines the model's parameters, what criteria weigh each variable and what consequences it has for newsroom culture that an algorithm assigns value to journalistic decisions. Svenska Dagbladet reached 130,000 digital subscribers ahead of schedule with that system, making the case an empirical argument for those who consider that quantifying editorial value is incompatible with quality journalism.

Onet, one of Poland's leading news portals and part of the Ringier Axel Springer group, integrated generative AI tools directly into its consumption platform, including interactive news chat, article summaries and personalized digests, all built on proprietary editorial content and with explicit editorial criteria as a control layer. The case is relevant for multi-platform groups managing both journalistic production and digital distribution, because AI governance in that context involves simultaneous decisions about what AI can do with editorial content and how that transformed content is presented to a mass audience. Ringier Axel Springer chose to anchor both decisions in verified journalistic content as the exclusive source, which establishes a clear limit but also a scale restriction that the system must manage.

The National Library of Norway signed an agreement with Kopinor, the organization representing the country's newspapers, to use journalistic content for training language models in Norwegian and Sami languages. The Norwegian government pays 45 million kroner annually for that access, the resulting models are freely available for public and private use, and publishers can exclude content if they wish. It is the first country to establish this type of scheme and represents a thesis on digital sovereignty and cultural identity that no private agreement between media and technology companies can replicate. The state acts as guarantor that AI trained on national journalistic production is not captured by private interests. The case opens a question that Latin American media should ask themselves: what models of collective governance are possible in contexts where the state lacks the same investment capacity or the same traditions of collaboration with media.

ProPublica became one of the first American outlets where AI integration formally entered collective bargaining. During discussions over the newsroom's first union contract, the guild pushed for clauses linked to AI's impact on working conditions, including limits on possible staff reductions associated with its adoption. The case is relevant because it incorporates the labor dimension into AI governance in media, broadening the discussion beyond tools, editorial policies or technological infrastructure. The experience shows that the institutionalization of AI also implies defining rules about team participation, the scope of automation and the protection of roles within newsrooms.

The BBC developed a language model trained on thousands of its own articles to automatically adapt news texts to the outlet's editorial style guide. The system functions as an editing assistant that verifies that the language, tone and format of the text meet BBC editorial standards before publication. The most interesting governance element is the training decision: the BBC built a system that internalizes its specific linguistic and editorial conventions, rather than resorting to a generalist model. That process involves deciding which archive articles represent the desired standard, which style aspects are stable enough to be codified, and who has authority to update the model when criteria evolve. These are questions any outlet considering a similar path will have to answer before writing a line of code.

Business Insider implemented a generative AI system trained on its own archive to draft brief news articles, with mandatory human review and approval before each publication. What is relevant about the case from a governance perspective is the decision to accompany that implementation with an active transparency policy toward readers. Each AI-assisted article carries an explicit indication of that circumstance. In an ecosystem where most outlets used AI without disclosing it, Business Insider bet on making the process visible. That bet generated industry debate about applicable standards and contributed to normalizing the conversation about labeling AI-assisted content as part of the sector's editorial criteria.

In March 2025, News Corp Australia began deploying NewsGPT, a proprietary model combining ChatGPT, Google Gemini and Claude, with the stated goal of empowering staff work. Months later, journalists in the group raised concrete concerns about two issues: the possibility that the system would learn to imitate their individual writing styles, and the introduction of Story Cutter, an additional tool that threatened sub-editing functions. The case is valuable for what it documents: the labor tensions that emerge when large-scale AI integration is not accompanied by transparent negotiation with the teams. It is one of the few publicly documented cases where internal resistance took organized form — and for that reason, it is more instructive than many cases that only record positive results.