About the study
- Published
- December 2025
- Project
- Inteligência Artificial no Jornalismo em Portugal (IAJP)
- Coordination
- Paulo Nuno Vicente, José Sotero
- Funding & support
- European Media and Information Fund (EMIF) & Fundação Calouste Gulbenkian
Executive summary
Background
This report provides a multi-institutional, comprehensive assessment of how artificial intelligence (AI) is being adopted, perceived, and governed within Portugal's journalism ecosystem. Prepared by a consortium led by the Universidade NOVA de Lisboa alongside 7 other national and international universities, the study combines quantitative surveys, focus groups, and co-design workshops to outline the modern structural shift occurring across Portuguese newsrooms.
Chapter 1
State of the Art & The “Efficiency Paradox”
The empirical data shows that AI adoption in Portuguese newsrooms is mostly operational, fragmented, and predominantly confined to low-risk administrative workflows rather than active content generation.
Usage frequency statistics
- Only 11.2% of media professionals use AI tools on a daily basis.
- 15.6% report frequent usage.
- The rest of the ecosystem uses it occasionally (28.1%), rarely (25.4%), or never (19.7%).
Medium-specific variations
- Online / Digital-Native Media: High penetration, with 57.6% of digital journalists utilizing AI tools daily.
- Television Media: Significantly slower adoption curve; 18.0% have never used AI and 15.0% use it rarely.
Top technical workflows delegated to AI
- Information search and data gathering: 56.1%
- Machine translation: 41.0%
- Audio transcription of interviews/speeches: 40.0%
- Direct generation of articles or creative writing remains marginal (ranging between 2.7% and 24.4%).
The “Efficiency Paradox”: The report introduces this concept to describe how time saved on automation is frequently consumed by the rigorous human verification, algorithmic checking, and training updates required to maintain editorial truth. Real-world quality gains only surface if organizations intentionally reinvest the saved hours back into deeper human reporting.
Chapter 2
Disinformation and Synthetic Media
Portuguese media professionals approach generative AI with deep apprehension, viewing it simultaneously as an accelerator of fake content and an industrial asset for counter-fact-checking.
Risk metrics
- 64.4% of professional journalists believe generative AI will exert a negative or highly negative impact on the overall information landscape by super-charging disinformation.
Medium-specific vulnerabilities
- Written Press: The most pessimistic sector, with 50.5% expecting an entirely negative impact.
- Television: Deep concerns center around video deepfakes and facial/voice manipulation.
- Radio: Strong fears regard the public’s eventual inability to distinguish synthetic automated radio hosts from real human voices.
Real-world incident case studies (2024)
- The SAPO24 Incident: Highlights the risk of automated translation/summarization loops that published material without human verification, prompting official warnings from the Regulatory Authority for Media (ERC).
- The "Pulsómetro" / CNN Portugal Case: Cited as a key example proving why strict human control and transparency protocols are mandatory when using experimental computational tools during live television broadcasts.
Chapter 3
Ethics, Deontology, and Corporate Governance Deficits
A primary finding of the report is the severe structural lack of formal guidelines, training programs, and compliance architecture across Portuguese media houses.
| Corporate governance parameter tracked | Reporting absolute absence |
|---|
| Contractual Clauses (AI constraints in labor contracts) | 80.4% |
| Clear Internal Liability Rules for AI Errors | 85.0% |
| Compulsory Employee AI Training Programs | 67.5% |
| Dedicated Editorial AI Supervision Committees | 65.8% |
| Specific AI Codes of Ethics/Conduct | 64.0% |
| Pre-Publication Systematic AI Validation Checklists | 63.4% |
Regional and local asymmetry: The total lack of corporate guidelines worsens dramatically within regional and local media organizations (p=0.017). This structural void fosters “invisible practices,” where isolated journalists use free, unvetted commercial tools without managerial oversight or compliance frameworks.
Chapter 4
Copyright, Intellectual Property, and Platform Dependency
The text stresses a major structural imbalance between national media companies and international technology companies (Big Tech), exposing extreme algorithmic dependencies.
Content scraping policies
- 52.2% of professionals note that their publishers have never formally granted permission for content to be scraped to train Large Language Models (LLMs).
- 19.0% are completely unaware of whether their company's intellectual property archives are being scraped without compensation.
Algorithmic autonomy gap
- A marginal 6.8% of journalists declare that their news organization owns or maintains full sovereign control over the algorithms they use.
- Only 26.4% believe their technical departments possess the expertise to evaluate or audit third-party generative products (OpenAI, Google, Microsoft) before deployment.
Policy roadmap
10 strategic directives
- National AI Media Framework: Developed under the direction of the ERC to build elastic, principles-based regulation that preserves journalistic autonomy without stifling technical innovation.
- Corporate AI Handbooks: Mandating that every newsroom draft clear operational guidelines defining strictly permitted, recommended, and prohibited uses of generative tools.
- Mandatory Generative Watermarking: Requiring clear, visible signaling and public disclosures whenever consumer-facing content is synthetically generated or heavily assisted by automated systems.
- Public Service Upskilling Campaigns: Utilizing the Public Broadcaster (RTP) alongside the GILM group to design media literacy courses, drawing inspiration from interactive gamified models like the BBC IReporter Game.
- Copyright Code Overhaul: Amending the Portuguese Copyright Code (CDADC - Decreto-Lei n.º 63/85) to secure robust, mandatory financial compensation models for publishers whose data is ingested by LLM developers.
- Incentivizing Local Laboratories: Dedicating at least 50% of public media innovation funds (#Portugal Media Lab) exclusively to local and regional media consortia to bridge technological divides.
- Exploratory Science Grants: Urging the Fundação para a Ciência e a Tecnologia (FCT) to finance targeted, applied research loops that embed AI pilot software directly inside active newsrooms.
- National AI & Journalism Observatory: Establishing an independent, cross-university research center to publish biannual empirical reviews tracking automated journalistic workflows.
- CENJOR Specialized Reskilling: Deploying short-form professional certification modules focusing on forensic deepfake verification (audio/video cloning) and advanced semantic search tools.
- Curricular Renewal by 2030: Restructuring higher education Journalism and Communication degrees across Portugal to embed computational thinking, algorithmic criticism, and machine ethics directly into core undergraduate curricula.
Conclusion
The 2025 White Paper emphasizes that AI is neither a simple solution for the financial crises plaguing modern newsrooms nor a catastrophic threat destined to replace reporters. Instead, the ultimate challenge lies in organizational scale. Because independent newsrooms lack the clout to negotiate with massive global tech platforms on their own, the preservation of public trust and journalistic integrity in Portugal requires immediate, collaborative action across regulatory bodies, academic networks, and media executives.