From urgency to protocol
The widespread adoption of generative AI tools in newsrooms from 2023 onward did not find the media industry unprepared, but it did find it without a shared framework. The speed with which models like GPT-4, Gemini or Claude became part of journalists' daily workflows generated an urgent need that the sector had not faced since the digital transition: the need to define, in writing, what can be done with these tools, under what conditions, and with what oversight.
When the first edition of this guide was published in 2023, the regulatory landscape was still emerging. The World Press Trends Outlook report by WAN-IFRA found that 46% of editorial roles were already experimenting with generative AI, while only 28% had a clear framework for its use. Much of the transformation was taking place informally, driven by individuals or small teams exploring available tools without an institutional framework to guide them.
This urgency produced something unprecedented in the recent history of journalism: a massive, simultaneous and global proliferation of editorial guidelines. From large international newsrooms to regional digital media, from public broadcasters to wire agencies, the industry began drafting documents that attempted to answer the same questions under very different contexts and with very different resources.
Transparency stopped being a mere ethical aspiration and became an editorial asset. Audiences began demanding to know when AI had been used in the production of a piece; regulators advanced in different jurisdictions; and media organizations realized that the absence of a clear policy was itself a policy decision, and generally a bad one.
But editorial guidelines are not neutral documents. They reflect power structures, access to technology, levels of institutional development and, above all, the vision each organization has of its own journalistic mission. A guideline from the BBC or the New York Times is not directly transplantable to a regional digital outlet in Latin America, even if both face similar technological challenges.
According to a 2023 WAN-IFRA report, 85% of media organizations said they had adopted or were developing AI policies that year. However, the same research revealed a structural gap: the media with the most resources were those best equipped to translate those policies into concrete operational processes, while smaller or less-resourced organizations tended to adopt generic principles without the means to implement them.
This chapter maps that landscape. Starting from an expanded corpus of more than 90 editorial guidelines from 25 countries, it identifies common patterns, blind spots, and new demands that the 2026 version of the guidelines framework must address. The goal is not to create a single universal guideline—which would be illusory and counterproductive—but to provide a comparative framework that allows each newsroom to build its own policy from real evidence and concrete context.
The evolution of guidelines: three generations of response
The comparative analysis of the corpus allows us to identify at least three generations of guidelines, each with distinct characteristics and priorities.
The first wave of guidelines was characterized by its exploratory and principlist nature. Organizations like Wired, Associated Press, DPA and the Financial Times published documents that established general principles—transparency, human supervision, prohibition of fabrications—but offered few operational indications. These guidelines were, above all, positioning statements: signals to audiences and the market about the organization's values in the face of a new and disruptive technology.
The second generation marked a turn toward operational specificity. The New York Times, The Guardian, BBC, Reuters and institutional frameworks like the Council of Europe's began incorporating differentiated uses (what is permitted, what is restricted, what is prohibited), data protection provisions, audience transparency protocols and, in some cases, clauses regarding intellectual property and use of journalistic sources to train models. The document ceased to be a declaration and began to function as an internal regulatory framework.
The third generation—which this edition maps for the first time—incorporates organizations from Latin America, Ibero-America and the Global South with a markedly more complex approach: governance structures, specific protocols for images and audio generated by AI, provisions on technological dependency and supplier management, and differentiated clauses by editorial role and type of content. Organizations like Chequeado, Adepa, El País, Newtral, Grupo Globo, Folha, Estadão, RTVE and La Nación, but also the World Economic Forum and more recent guidelines from major international media, are part of this generation.
The expanded corpus
For this update, guidelines from 90 media organizations, agencies, professional associations and international bodies from more than 25 countries were collected and analyzed. As in the previous edition, the work followed three sequential stages: identifying the distinctive features of each guideline, recognizing shared guiding principles, and developing a proposed usage policy for newsrooms.
The geographic distribution of the corpus reflects both the global advance of AI adoption in media and the growing participation of Latin America and Ibero-America. The most represented countries are Spain, Brazil, the United States, Argentina, the United Kingdom and Germany, followed by Belgium, France, Sweden, Switzerland, Finland, the Netherlands and Portugal. The corpus also includes Puerto Rico, Colombia, Mexico, the Czech Republic, Italy, Kenya, South Africa, Turkey, China and Norway, alongside international organizations such as the Council of Europe and the World Economic Forum.
Latin America’s emergence on this map is one of the most significant changes since the previous edition. In 2023, Latin American representation in the corpus was limited to Todo Jujuy, Núcleo and SIP. In this edition, Argentina contributes seven guidelines with the addition of Chequeado, Adepa, 0221, La Nación, ADN Sur and Cenital.
Brazil grows from two to twelve guidelines, including Folha de S.Paulo, Grupo Globo, Agência Pública, Aos Fatos, Agência Mural, Gênero e Número, O Tempo, Correio Sabiá, Revista AzMina and Vocativo. Also appearing for the first time are El Nuevo Día (Puerto Rico), El Espectador (Colombia) and Siete24 (Mexico), forming a Latin American map that was almost nonexistent three years ago.
This regional presence is joined by a dense network of guidelines from the Iberian Peninsula, including major outlets such as El País, RTVE, Agencia EFE, Atresmedia, Newtral, Jotdown and El Diario.es, and professional associations such as Asociación de la Prensa de Madrid, Colexio de Xornalistas de Galicia, Col·legi de Periodistes de Catalunya and Asociación Profesional de Periodistas Valencianos.
The full list of surveyed media and organizations, grouped by country of origin, includes: Bavarian Broadcasting, DJV, DPA and Springer Nature (Germany); Todo Jujuy, Chequeado, Adepa, 0221, La Nación, ADN Sur and Cenital (Argentina); Flemish Organization for Journalism / RVDJ and Mediahuis (Belgium); Núcleo, Estadão, Folha de S.Paulo, Grupo Globo, Agência Pública, Aos Fatos, Agência Mural, Gênero e Número, O Tempo, Correio Sabiá, Revista AzMina and Vocativo (Brazil); Hong Kong Free Press (China); El Espectador (Colombia); Seznam Zprávy (Czech Republic); Wired, Business Insider, News Media Alliance, Reuters, The Atlantic, USA Today, NPR, RTDNA, Associated Press, The New York Times, Afro LA, SIP, ICFJ and CBC (USA/Canada); Catalan Media Council, Agencia EFE, El País, RTVE, Newtral, Atresmedia, Jotdown, El Diario.es, Asociación de la Prensa de Madrid, Asociación Profesional de Periodistas Valencianos, Colegio Vasco de Periodistas, EITB, Colexio de Xornalistas, Col-legi de Periodistes de Catalunya and Berria (Spain); STT and Sanoma (Finland); Le Parisien, Le Figaro and RSF (France); Panhellenic Federation of Journalists’ Unions (Greece); Col-legi de Periodistes d'Italia (Italy); Media Council Kenya (Kenya); Siete24 (Mexico); Norwegian Tabloid (Norway); ANP and Volkskrant (Netherlands); Mensagem de Lisboa and Expresso (Portugal); El Nuevo Día (Puerto Rico); BBC, Financial Times, The Guardian, Telegraph, The Economist, The Globe and Mail and PINF (United Kingdom); Press Council of South Africa (South Africa); Aftonbladet/Schibsted, Svenska Dagbladet, TT Nyhetsbyran and Nordic AI Journalism (Sweden); Heidi.News, Ringier and Republik (Switzerland); Anadolu Agency (Turkey); Council of Europe / CDMSI (European Union); and World Economic Forum (International).
Featured guidelines
Among the 90+ documented guidelines, the following stand out for their originality, operational depth, or contribution to defining new standards. The selection is not exhaustive or hierarchical: it seeks to illustrate the diversity of approaches and the richness of the accumulated global production.
Quotations below are translated from the Spanish edition of this guide. Consult the linked policies for their original wording.
The fact-checking organization published one of the most complete and structured policies in the Spanish-speaking world. It distinguishes between generative AI tools developed in-house to automate specific tasks, such as El Explorador, Desgrabador and Chequeabot, and external models used experimentally or as a complement. Its policy organizes uses into explicit categories: transcription, translation, style corrections according to its own manual, database processing, support for checking misinformation and assisted drafting of institutional communications. It defines with equal clarity what AI is not used for and incorporates accountability, equity and non-discrimination principles that distinguish it from more generic guidelines.
“We began experimenting with AI in 2016 and, as part of our founding commitment to transparency, we explore and apply artificial intelligence tools responsibly, aware of their opportunities and risks.”
The recommendations of the Asociación de Entidades Periodísticas Argentinas emerged from a participatory process that brought together key figures in the national industry over several months. The document sets out eight concrete recommendations, from prioritizing AI as a tool for quality to audience trust and the promotion of appropriate legal frameworks. Its collaborative approach, adapted to media with different scales and resources, gives it particular value as a regional reference.
“Designate teams responsible for implementing AI projects in the outlet’s products and managing all the aspects detailed in editorial and organizational processes. Promote ethical leadership that encourages a culture of innovation and continuous training.”
One of the country’s largest-circulation newspapers published its newsroom AI guide, joining a growing Argentine policy ecosystem that includes Chequeado, Adepa, Todo Jujuy, 0221, ADN Sur and Cenital. Together they make Argentina one of the Latin American countries with the strongest concentrations of formalized guidelines.
“To counter public perceptions of AI as an autonomous force operating without editorial control, LA NACION is encouraged to be specific when telling its audience what type of AI we use, including mentioning it in the methodology of each AI-based production. This can serve an educational function, demystifying AI and communicating about technology as a tool.”
The La Plata digital outlet published guidelines distinguished by regular audits of implemented tools to detect bias and ensure alignment with editorial principles. Its policy includes commitments to continuous training, clear protocols for correcting AI-related errors and an explicit approach to continuous improvement as an iterative process. The specific tasks covered, from audio transcription to real-time data analysis for identifying trends, reflect practical and considered integration into daily routines.
“AI implementation will be a continuous process of learning and adaptation, focused on improving the information experience for our audience and optimizing internal workflows. We will strive to identify new opportunities for innovation that allow us to meet audiences’ changing needs without losing sight of our duty to inform with accuracy, relevance and humanity.”
It was one of the first Argentine outlets to implement a specific style manual for AI-generated content. Its approach prioritizes regular training for the journalism team and explicitly prohibits using AI to generate complete content or act as the final editor, maintaining human-led content production as its guiding principle.
“We will provide regular training for our professional journalism team on the potential that adopting this technology can bring to their work, as well as the crucial role their decision-making will play in selecting and reviewing the content they use, just as it currently does with any other source of information.”
It incorporated an AI protocol into its Style Book based on two organizing principles: every item generated with AI must be supervised by a journalist who takes responsibility for it, and the newspaper must communicate its use of the technology transparently. AI-assisted illustrations are allowed as a complement, provided that the generated elements are secondary and their use is disclosed. The policy also requires all staff to declare the AI tools they use in their work.
“Images, audio or any other synthetic content generated by artificial intelligence that could be confused with reality, mislead, spread disinformation or violate privacy will not be used.”
It integrated AI guidelines into its Libro del Estilo Urgente, giving them the same normative weight as the agency’s other editorial rules. The agency prohibits using AI to create images illustrating current news and actively monitors for AI-generated images entering its services through third-party contributions. Content generated automatically from identified sources, such as sports or stock market data, requires an explanatory note at the end of the text.
“Artificial intelligence systems may, in some cases, support the work of EFE journalists, but they can never do journalism; that work belongs to us.”
The fact-checking outlet’s ethical guide stands out for its emphasis on developing its own technologies rather than depending on external suppliers. The policy upholds human talent as its guiding principle and permits AI-generated images only as an occasional complement, always subject to copyright.
“At Newtral, we do not use opaque third-party algorithms: we focus on producing them ourselves and supervising their development to control their architecture and content.”
Spain’s public broadcaster places its AI guidelines within the international regulatory context, from the Bletchley Declaration to the European Union AI Act and Spanish National Artificial Intelligence Strategy. Its contribution is particularly relevant as a model for public media that must reconcile technological innovation with public service obligations and democratic accountability.
“Having a standard for AI use promotes transparency and internal and external understanding of what is done, with what tools and for what purpose.”
Latin America’s largest media conglomerate updated its editorial principles to incorporate specific guidance on artificial intelligence. Integrating AI guidelines into its general editorial principles, rather than a separate document, reflects a mature understanding of technology as a constituent part of journalism practice rather than an external addition.
“The use of artificial intelligence must always be supervised by a human being, and no content produced with this technology should be distributed without such oversight. This does not mean that every automatically generated item will be reviewed by a human before distribution. That requirement would negate much of the efficiency and reach the tool enables. The accuracy and quality of this content will be ensured through oversight of the process, sample readings and other forms of verification.”
It included AI guidelines in its Newsroom Manual, within the professional conduct section. As with Agencia EFE in Spain, integrating AI rules into the existing style manual gives them the same normative standing as the newspaper’s other editorial rules.
“Folha professionals may use artificial intelligence applications in their work. This tool does not replace human judgment or exempt journalists from responsibility for the final result, but, used well, it offers several opportunities to increase newsroom efficiency.”
Grupo Estado was one of the first Brazilian media organizations to formalize an AI policy. Its most significant regional contribution is an Artificial Intelligence Committee with representatives from Technology, Newsroom, Product, Legal and Audit. The committee analyzes AI-related editorial projects, establishing a model of shared governance that other newsrooms can adapt. The policy also prohibits entering business data, trade secrets, proprietary databases and journalistic scoops into external AI tools.
“The aim is to provide, in the right measure, full encouragement to experiment with these new technologies and the guarantee that final responsibility for journalistic output will always belong to a person, a professional of flesh and blood.”
The Brazilian fact-checking organization adds its AI policy to the country’s growing ecosystem of guidelines. With twelve formalized policies, Brazil has the densest such ecosystem in Latin America and one of the most developed in the Global South.
“Artificial intelligence will only be used to adapt or paraphrase content already written, edited and published by the Aos Fatos journalism team.”
It sets precise limits on AI use, allowing it as a consultation tool to improve specific parts of the editorial process and suggest alternative social media posts, while prohibiting its use to generate complete content or act as the final editor.
“Using AI to create illustrations is a highly urgent and exceptional case, and preference is always given to images created by in-house or independent artists, illustrators and designers.”
The appearance of guidelines in these three cases shows that formalizing AI policies is no longer limited to the largest media organizations. Each contributes to a Latin American policy ecosystem that was almost nonexistent three years ago and now spans six countries with more than twenty formalized guidelines.
It organized its principles around four areas: AI as a tool serving the journalistic mission, continuous human guidance and review, transparency and ethics, and protection of intellectual property. The Times introduces digitally voiced articles and automated translation as new forms of accessibility, pointing to applications beyond text production.
“We see the technology not as a magic solution but as a powerful tool that, like many technological advances before it, can be used in service of our mission. The expertise and judgment of our journalists are competitive advantages that machines simply cannot match.”
It establishes that any use of AI in creating, presenting or distributing content must align with existing editorial guidelines, including impartiality, accuracy, fairness and privacy.
“Where AI is used for editorial purposes, there must be a senior editorial figure who is responsible and accountable for overseeing its implementation and ongoing use.”
It requires human oversight and explicit permission from a senior editor to include significant AI-generated elements in any published work.
“We will use generative AI tools for editorial purposes only when they contribute to the creation and distribution of original journalism. We will guard against the risks of bias embedded in generative tools and their training datasets.”
It stands out for institutionalizing an internal experimentation register documenting all AI-related newsroom activity, including external suppliers. The outlet prohibits publishing photorealistic AI-generated images, while exploring enhanced images for infographics and diagrams with their origin always disclosed.
“Journalism at the Financial Times will continue to be produced by highly skilled human professionals who report on and analyze the world accurately and fairly.”
As a global news agency, its guidelines have a multiplying effect across thousands of outlets using its content. AP establishes human oversight standards that numerous organizations worldwide have adopted as a reference.
“Accuracy, fairness and speed are the values that guide AP’s news reporting, and we believe the thoughtful use of artificial intelligence can serve these values and, over time, improve how we work.”
The guidelines adopted by the Steering Committee on Media and Information Society in November 2023 constitute the most comprehensive institutional framework at a supranational level. Grounded in Article 10 of the European Convention on Human Rights, they connect AI use in journalism with freedom of expression, privacy, human dignity and non-discrimination. They also include a checklist for procuring AI systems, translating abstract principles into operational decision-making criteria.
Promoted by Reporters Without Borders and 16 partner organizations, this charter establishes principles recognizing AI’s transformative implications for the right to information. It advocates worldwide cooperation to ensure AI respects human rights, peace and democracy, and offers a reference for media seeking to place their practices within globally shared values.
Although included in the corpus since the guide’s first edition, the Inter American Press Association’s principles also deserve recognition as an institutional framework. They cover media across the Americas, and their endorsement of global AI principles makes them a reference for press associations throughout the region.
Common guiding principles: a new version
The comparative analysis of the 90+ guidelines makes it possible to identify a set of principles that appear, with variations in wording and emphasis, across a wide majority of documented frameworks. These are not universal agreements, but significant convergences that reflect an emerging industry consensus on the foundational values that should guide the use of AI in journalism.
All documented guidelines establish, explicitly or implicitly, that final editorial responsibility rests with human journalists and their organizations. AI is presented as a tool that assists, accelerates or augments human capacity, but never as an autonomous agent with editorial authority. This principle appears with particular force in guidelines for news, verification and sensitive coverage contexts.
The obligation to inform audiences when AI has been used in the production of content is one of the most consistent principles in the corpus, even when the specific implementation modalities vary significantly. Some organizations opt for systematic labeling; others establish selective disclosure thresholds; and others limit communication to cases of significant AI use. The "how" varies; the "that" is consensus.
Guidelines converge on the need to ensure that AI use does not compromise editorial integrity, journalistic ethics, or the rights of people involved in journalistic coverage. This includes specific provisions on the protection of sources, the privacy of journalistic subjects, the prohibition of AI use to generate disinformation or manipulate public opinion, and the protection of the rights of creators whose work may have been used to train AI models.
The guidelines don't reject technological innovation, but condition it: AI adoption must serve the journalistic mission, not replace it. This principle appears in a wide variety of formulations—from the restrictive ("AI will not be used for content that requires editorial judgment") to the open ("AI can be used when it improves our capacity to serve the public interest"). What they share is the primacy of mission over efficiency.
The most developed guidelines—especially in the third generation—include explicit provisions on training, professional development and capacity building for the use of AI tools. This principle reflects the recognition that the responsible adoption of AI requires not only rules, but skills: editorial teams must understand what the tools do, what their limits are, and how to evaluate their outputs.
An emerging principle—visible especially in institutional frameworks and sectoral associations—is the recognition that some challenges of AI in journalism exceed the capacity of individual organizations and require collaborative responses: common standards, shared research, industry-level regulatory frameworks and cooperation mechanisms that allow smaller media to access resources and knowledge they couldn't develop on their own.
From principles to decisions
| Declarative principle | Operational decisions in the newsroom | |
|---|---|---|
Human oversight |
Who approves AI-generated content? What is the minimum review required? Who is responsible if it fails? |
|
Transparency |
When do we disclose AI use? How? Where does the label go? Does it apply to all content or only some? |
|
Avoiding bias |
How do we test tools before adopting them? Who evaluates outputs for bias? What do we do when we detect problems? |
|
Editorial responsibility |
What uses of AI require prior authorization? Who grants that authorization? What is the review process? |
|
Responsible use |
What tools are approved? What criteria must a tool meet to be approved? Who maintains the approved list? |
|
Quality and accuracy |
What verification standards apply to AI-generated content? Are they different from those for human-produced content? |
|
Data protection |
What information can be entered into AI tools? What is prohibited? How do we manage confidential sources? |
|
Innovation |
How do we evaluate and test new tools? Who decides to incorporate them? What is the trial period? |
New axes: what 2026 added to the ethical map
Beyond the principles that already appeared in the first generation of guidelines, the comparative analysis of the 2026 corpus reveals at least four new axes that had not been addressed—or had been addressed only superficially—in previous frameworks. These axes reflect new dimensions of AI adoption in journalism that the industry's accumulated experience has made visible.
The most sophisticated third-generation guidelines no longer just declare principles: they describe governance structures. Who is responsible for AI policy? Who approves new tools? Who reviews compliance? Who manages incidents? The transition from a document of principles to a governance architecture represents the most significant maturation observed in the 2026 corpus. Organizations like Grupo Globo, Estadão, La Nación and the BBC have incorporated committees, review processes and defined responsibilities that make guidelines operational rather than merely declarative.
The proliferation of tools for generating images, videos and synthetic audio produced a gap in first and second generation guidelines that the 2026 corpus begins to fill. The generation of realistic images of people who don't exist or events that didn't happen; synthetic voice applied to real public figures; deepfakes and manipulated videos: these uses require specific protocols that general transparency guidelines don't cover adequately. The most complete 2026 guidelines include specific sections for these cases, with differentiated use restrictions by context (editorial news, illustration, entertainment, historical reconstruction) and specific audience communication requirements.
As AI regulation advances—especially the European Union's AI Act, but also regulations in Brazil, Canada and other jurisdictions—the guidelines begin to incorporate explicit regulatory compliance provisions. International references such as the UNESCO Recommendation on the Ethics of Artificial Intelligence and the OECD recommendations also appear as context. This is a significant shift: the first guidelines were purely self-regulatory; the most recent ones must operate within an emerging legal framework that imposes obligations and limitations that go beyond editorial discretion. The interaction between self-regulation and hard regulation is one of the most relevant dynamics in the evolution of the field.
The first guidelines tended to establish general principles applicable to all AI use. The most sophisticated 2026 guidelines distinguish between uses: what is permitted for summarizing press releases is not necessarily permitted for interviewing sources; what is acceptable for translating content may not be acceptable for generating images for coverage of armed conflicts; what applies to the sports section may be different from what applies to political information. This differentiation by task, context and risk level represents a significant increase in operational sophistication.
The blind spots of guidelines
Despite the maturation of the corpus, the comparative analysis also reveals significant blind spots: areas where even the most developed guidelines offer insufficient or absent responses. Identifying these gaps is as important as documenting what has been built, because it shows where the work of the industry—and of each organization individually—remains unfinished.
An almost universal blind spot in the documented corpus: no media policy adequately addresses what to do when an AI supplier changes its terms of service, raises its prices, restricts access to its models, or disappears from the market. The growing dependence of many editorial processes on a small number of technology suppliers—primarily OpenAI, Google, Anthropic and Microsoft—creates systemic risks that current guidelines don't address. What happens to a newsroom that has integrated AI into critical processes if a supplier decides to restrict access to its models? None of the 90+ documented guidelines offers a clear answer.
Most guidelines say what should be done; very few explain how to verify that it is actually being done. The absence of audit mechanisms, compliance indicators and review processes means that many guidelines function more as statements of intention than as operating systems. How does a newsroom know that its journalists are applying the guidelines? How does it detect deviations? How does it correct them? The gap between what is written in the document and what happens daily in the newsroom is one of the most significant challenges the industry faces, and it is largely a gap in compliance architecture rather than principles.
The reference models in the corpus come predominantly from large media with significant resources: the BBC, the New York Times, Estadão. These organizations can afford to develop detailed governance structures, invest in staff training, and allocate people specifically to AI oversight. Small regional media, freelance journalists and resource-limited digital newsrooms face very different challenges. The industry urgently needs models of minimal but effective guidelines adapted to organizations that cannot replicate large-media structures. The current corpus is, in this sense, systematically biased toward organizations with more resources.
These blind spots are not merely theoretical. They have practical consequences: a newsroom that doesn't plan for technological dependency may find itself paralyzed if a supplier changes its terms; one that doesn't verify compliance may be declaring principles it doesn't apply; and one that adopts a guideline designed for a large media organization may find it useless or counterproductive in its specific context. Identifying these gaps is the first step toward addressing them.
Proposed guidelines for newsrooms: 2026 version
Based on the comparative analysis of the 90+ documented guidelines and the identification of best practices, blind spots and new demands, we present below a proposed updated guidelines framework for 2026. This framework is not intended as a universal model to be adopted as-is, but as a starting point that each organization can adapt to its context, size, resources and journalistic mission.
Guidelines for the use of artificial intelligence in [outlet name]
At [outlet name], we understand artificial intelligence as a tool serving our journalistic mission. Its integration into our editorial processes is governed by human oversight, transparency, rigor, equity and responsibility. Every AI application will remain subject to our professionals’ editorial judgment and our commitment to truthfulness, accuracy and audience trust.
AI-assisted content production will be supervised by journalists and editors at every stage. Final responsibility for publication always rests with the editorial team. Every AI-assisted production will be reviewed and approved by a responsible person before publication. Any AI-related error will be addressed through a clear correction and communication protocol.
We will communicate clearly when and how AI is used in content production. Content generated automatically from identified sources will include an explanatory note. We will explain the tools used and the criteria guiding their application. We will remain open to our audience’s questions about the use of these technologies.
Permitted uses (with standard editorial oversight): transcription of audio and interviews, translation assistance, grammar and style correction, summaries of long documents, drafts for social media posts, database analysis and processing, and support for journalistic research.
Restricted uses (requiring authorization from an editor or designated person): headline and standfirst generation, assisted drafting of news texts, data visualizations, AI-enhanced images for infographics or visualizations, and experimentation with innovative narrative formats.
Prohibited uses: publishing entirely AI-generated texts without editorial oversight; synthetic images, audio or video that could be confused with reality; entering confidential data, anonymous sources, journalistic scoops or personal information into external AI platforms; and using AI as the final editor or a primary information source.
AI-generated images will be used cautiously and visibly labeled. Photorealistic AI-generated images are prohibited as illustrations for current news. AI-assisted illustrations are allowed when generated elements are secondary and their use is disclosed. Active monitoring will prevent synthetic material from entering our content through external sources.
Sensitive information will not be entered into external AI platforms. This includes personal data about sources or audiences, confidential documents, strategic business information and unpublished journalistic material. Applicable data protection rules will be followed, and security measures will be implemented for handling information.
[Outlet name] will designate an editorial AI lead to supervise implementation, document uses, coordinate training and serve as a point of contact for the team. Larger organizations are encouraged to create an AI committee with representatives from editorial, technology, legal and product teams. An internal register of AI tools, ongoing projects and evaluations will be maintained.
Continuous training programs on ethical and effective AI use will be implemented for the entire journalism team. Training will cover both the potential and limitations of available tools and will be updated periodically to incorporate new technological developments and lessons from practice.
Intellectual property rules will be respected in every use of AI. Priority will be given to tools that incorporate authorization and transparency criteria in model training. Using AI does not entail relinquishing any rights over the outlet’s original content.
Before adopting any AI tool, the outlet will assess the supplier’s contractual terms, including pricing policies, service continuity and data portability. Significant changes will trigger an assessment of whether to continue, replace or adapt its use. Outlets with limited resources may simplify this process but cannot omit it: scale shapes the means, not the principles.
These guidelines will be reviewed periodically to incorporate lessons from their application, technological advances and changes in the regulatory framework. We will maintain an ongoing dialogue with our community to ensure that our AI practices align with our audience’s expectations and needs.
Governance and AI Guidelines Assistant
Tool to design and analyze AI usage policies for news organizations. Based on a comparative survey of more than 90 international guidelines from 25 countries, it helps develop editorial criteria, governance frameworks and processes for responsible AI integration in newsrooms.
Open assistant in ChatGPTThe gap between document and practice
Writing a guideline is the beginning, not the end. The real challenge for media organizations is not producing a document—a task that, as this corpus demonstrates, thousands of organizations have already accomplished—but sustaining a practice. A guideline that exists only in a PDF or an internal page has limited value if it is not known, applied, and periodically reviewed by the people whose work it is supposed to guide.
The distance between what is declared in a guideline and what actually happens in a newsroom's daily life is one of the most significant challenges the industry faces. It is not necessarily a problem of bad faith: it can be a problem of resources (there is not enough time to train), of organizational culture (habits change slowly), of governance (it is not clear who is responsible for compliance), or simply of communication (journalists have not internalized what the guideline says).
The distance between what is declared and what is done is, perhaps, the pending ethical gap of this stage.
The guidelines that seem most promising in the 2026 corpus are not necessarily the most exhaustive, but the most operational: those that translate principles into specific processes, define clear responsibilities, include training mechanisms, and establish periodic review. A brief but concrete guideline that is actually applied is more valuable than a comprehensive one that serves only as a reputational signal.
The next step for the industry is to build the infrastructure that makes compliance possible and verifiable. This means developing specific training programs, creating peer oversight mechanisms, sharing learnings between organizations, and investing in research that helps understand what works and what doesn't. The question is no longer whether AI should be used in journalism—that debate was settled by practice. The question is how to ensure that its use serves the values that give journalism its social purpose: truth, independence, and public service.