As I previewed in an earlier post — which referenced Meta’s AI system on WhatsApp failing to cite sources and acknowledging the harm caused to the publishing sector — I decided to dig deeper into how the Big Tech companies active in the world of publishing have shaped their policies in the context of digital media. Who has chosen to stand by publishers, and who has instead taken advantage of them?
In 2025, the relationship between technology platforms and publishers is no longer only about site traffic and links — it is about the use of generative AI, the right of access to content for model training, and the economic sustainability of media. This is an intense confrontation involving legal, economic, and technological dimensions, with the various players adopting profoundly different strategies.
For publishers, this is a complex transitional phase: it is essential to fully understand the dynamics at play, assess risks and opportunities, and envision sustainable business models in a landscape dominated by Big Tech and artificial intelligence.
Here is an updated map of the situation.
Google: between billion-dollar investments and AI that (partly) cuts traffic
Google is the player with the most structured — and most ambivalent — strategy. On one hand, it has invested for years in programs such as News Showcase ($1 billion globally) and Extended News Preview, formalizing licenses with hundreds of European publishers. On the other, its evolution toward generative AI — with projects such as Search Generative Experience (SGE) — is a source of concern for many in the industry.
SGE in particular alters the dynamics of organic traffic: by delivering immediate, synthesized answers directly in the SERP, it reduces the need to click through to original sources. Estimates suggest a potential traffic decline of between 20% and 40% from Google toward certain publications, especially on evergreen topics or navigational queries.
The issue of citations and copyright remains under discussion: Bard rarely cites sources, while SGE includes references but not always in a granularly traceable way. Google defends its conduct by invoking fair use, but accepts negotiation when the law requires it (e.g., Article 15 of the EU Copyright Directive). In Italy, thanks to scrutiny from the AGCM, significant agreements have been reached with many publishers.
- News Showcase: active in Italy since 2021, involving 76 Italian publications through individual agreements, as part of a global investment of more than $1 billion aimed at “supporting quality journalism.”
- ENP (Extended News Preview): more than 300 European publications licensed to display extended snippets on Google, in compliance with Article 15 of the EU Copyright Directive.
- SGE (Search Generative Experience): Google’s new AI frontier integrates generative snapshots into the SERP. Industry estimates suggest a potential reduction in organic traffic of between 20% and 40% for some publishers.
- Patent US11769017B1: highlights the use of AI to generate summaries from web content with selected sources, not always clearly cited — at least for now.
- Official position: Google emphasizes the traffic value it returns to sites, but accepts (where obligated) economic agreements. In Italy it avoided sanctions through “good faith” negotiations with AGCOM.
Meta: progressive disengagement and withdrawal from fact-checking
Meta has followed a trajectory opposite to Google’s. After investing more than $300 million in editorial projects through the Meta Journalism Project, the company now appears to be in retreat. The closure of Facebook News in Europe, the suspension of the third-party fact-checking program, and the total blocking of news in Canada in 2023 all signal a distancing from the world of information.
The message is unambiguous: according to Meta, the presence of news on its platforms is neither a priority nor profitable. Only 3% of content seen on Facebook involves journalistic articles and, according to the company, it is publishers who benefit from the visibility — not the other way around. When challenged by certain publications over compensation, Meta has retreated behind the wall of fair use.
In the AI domain, too, Meta has taken a more “conservative” stance: rather than building an autonomous answer engine in the style of Bard or Copilot, it integrates sources from Google and Bing into its chatbot. However, the training of the Llama 2 model has raised questions about the use of protected editorial content.
- Meta Journalism Project: invested $300 million (2017–2022) in journalism, including training, accelerators, fact-checking support, and emergency funds (e.g., during COVID-19).
- Facebook News: launched in the US, UK, and Germany — now closed in Europe.
- Since 2024, Meta has ended partnerships with fact-checkers on Facebook/Instagram in favor of automated tools.
- Generative AI: Meta AI integrates with Bing and Google Search, displaying sources. No proprietary aggregation platform.
OpenAI: the new actor, between training data and first agreements
OpenAI represents a break from traditional players. It does not distribute content — it uses it to train its language models (GPT-4). It is the only major actor to have signed formal agreements for AI training: the landmark case is the one with the Associated Press, which opened the door to future deals with other groups.
OpenAI has introduced tools to improve transparency, such as GPTBot, which allows publishers to opt out of training, and a Browsing mode that shows sources for online answers. However, the vast majority of ChatGPT responses still derive from the pre-trained model, without explicit source references. This remains a critical issue, particularly in light of incidents where AI has reproduced content protected by paywalls.
Regulatory pressure — the AI Act in Europe, potential class actions in the US — is pushing OpenAI to revise its approach. The agreement with AP and the $10 million fund with Microsoft for local newsrooms represent the first signals of a more responsible strategy.
- Agreement with the Associated Press (July 2023): archive license + access to AI solutions.
- Microsoft-OpenAI $10 million fund for US local newsrooms (2024).
- GPTBot: web crawler launched in August 2023 with opt-out via robots.txt.
- ChatGPT: now includes a browsing mode with cited sources only for Plus users, but most responses are generated from pre-trained models without direct attribution.
- Critical incidents: the Browsing feature was suspended in 2023 after reproducing full paywalled articles. OpenAI subsequently introduced anti-abuse filters.
Microsoft: the “responsible partner” of publishing?
Microsoft — perhaps because of its minority position in search (Bing holds roughly 3% market share) — has carved out a profile as a journalism ally. Through the Start portal and its apps, it has collaborated for years with more than 1,200 publishers worldwide. The AI pivot, with Bing Chat and the launch of Copilot Daily, did not undermine this approach. On the contrary: Microsoft updated its contracts to include the use of content in AI products, signing new agreements with major publishers including Reuters, Axel Springer, Hearst, and the Financial Times. Every Bing Chat response contains numbered source references, even where content is partially generated.
Furthermore, Microsoft was among the few Big Tech companies to openly support legislation in favor of publishers, such as the News Bargaining Code in Australia. This positioning has consolidated its reputation as a “responsible partner,” reducing its exposure to controversy and criticism.
The Italian case: editorial rights, AI, and regulatory prospects
Italy occupies an intermediate position, with certain strengths and certain gaps. Thanks to the EU Copyright Directive (transposed in 2021), many Italian publishers have active agreements with Google. However, Meta has never signed formal agreements with Italian media, and OpenAI has not yet established structured partnerships.
FIEG and other industry bodies are closely monitoring the question of content use in AI models. In the coming months, the approval of the European AI Act could open new scenarios, imposing licensing obligations or transparency requirements on the use of data in generative models.
For Italian publishing, this is a critical window in which to assert greater protection and compensation — drawing on experiences such as those of Australia or Canada while seeking, where possible, negotiated and sustainable solutions.
The relationship between digital platforms and media is no longer confined to the question of traffic or visibility: it now involves the very structure of informational content, its value, and the value of those who produce it, within the AI ecosystem.
On one hand, artificial intelligence promises efficiency, synthesis, and new opportunities; on the other, it risks disintermediating the role of newsrooms, reducing traffic and undermining business models that are already fragile and destabilized by the constant algorithmic shifts on which they have long depended.
Publishers need a clear strategy — but above all a new equilibrium: greater transparency around the use of content, clear and verifiable licensing agreements, and active involvement in shaping AI products. The future requires forums for dialogue in the interest of the entire sector. That future cannot be written solely by those who develop the technology — it must also be shaped by those who produce quality information every day. Or who strive to.
Frequently asked questions
