The headline
OpenAI just announced its first media partnership in Brazil, bringing Grupo Folha and Grupo UOL—two of the country's biggest news organizations—into ChatGPT. Starting today, all 900 million weekly active ChatGPT users globally can see summaries and citations from Folha de S.Paulo and UOL's reporting directly in chat responses.
This is OpenAI's standard playbook now: license quality journalism, surface it with attribution in ChatGPT, and position the partnership as a win-win for reach and revenue. They've done similar deals in the US, UK, France, and Germany. Brazil is the latest market.
But the numbers here are wild. Brazil has more than 50 million monthly active ChatGPT users exchanging around 140 million messages per day. That's one of the largest markets globally—bigger penetration than you'd guess from GDP alone.
Why this matters beyond the press release
Content licensing deals like this are the new normal for frontier AI companies trying to train and serve models without getting sued into oblivion. OpenAI needs fresh, high-quality text—especially non-English content—and publishers need distribution and revenue as traditional traffic erodes.
The interesting wrinkle here is how OpenAI is framing it. Varun Shetty, OpenAI's VP of Media Partnerships, said they want to "bring more useful, timely, and locally relevant answers to ChatGPT, while supporting the broader news ecosystem." That's diplomatic, but the subtext is clear: ChatGPT is a discovery surface now, not just a chatbot.
Publishers are betting on this too. Carlos Ponce de Leon, Co-CEO of Folha de S.Paulo, said "AI will define the next era of the news industry and Folha wants to help shape that future." That's not just licensing revenue—it's existential hedging. If LLMs become the primary interface for information retrieval, being inside the answer is better than being a blue link nobody clicks.
The API access angle
Buried in the announcement: Grupo Folha and Grupo UOL also get access to Codex, ChatGPT Enterprise, and OpenAI's API. This is the sweetener that makes these deals interesting beyond simple content licensing.
Publishers can now:
- Experiment with AI-powered features for readers (personalized summaries, conversational search, translation)
- Optimize internal workflows (content tagging, CMS automation, research assistance)
- Build new products on top of OpenAI's models without waiting in line for API access
This is where the real optionality lives. A newsroom with GPT-4 API access and a decade of archives can prototype tools that would've required a dedicated ML team two years ago. Whether they will is another question—most legacy media orgs are not exactly shipping labs—but the access is meaningful.
Murilo Garavello, UOL's Content Director, framed it as distribution ubiquity: "We want our journalism in every environment Brazilians use—and that truthful information can be disseminated as widely as possible." That reads like someone who understands the platform shift happening.
The Brazil-specific context
Brazil's ChatGPT numbers are legitimately surprising. Fifty million monthly actives in a country of ~215 million people means roughly 23% penetration—higher than most markets. The 140 million daily messages figure works out to ~2.8 messages per active user per day, which tracks with heavy usage patterns.
Why is Brazil so ChatGPT-forward? A few educated guesses:
- Portuguese is a top-10 global language by speakers but underrepresented in high-quality training data relative to English. ChatGPT fills a gap.
- Brazil has high smartphone penetration but inconsistent broadband. Mobile-first AI interfaces work well in that environment.
- The country has a young, digitally native population comfortable with experimental tech.
This also makes Brazil strategically important for OpenAI. If you're training multimodal models and want diverse linguistic data at scale, partnering with major Brazilian publishers is smart. You get structured, edited text in Portuguese covering local politics, culture, and events—exactly the kind of grounded, real-world content that's hard to scrape at quality.
Attribution theater or real crediting?
OpenAI keeps emphasizing "attribution, transparency, and links back to original sources" in these partnerships. The question is: what does that actually look like in practice?
From what we've seen in other markets, ChatGPT will surface inline citations when summarizing news, usually with a clickable link to the original article. That's better than the black-box era, but it's not the same as traditional referral traffic. Users get the gist in-chat; most don't click through.
Sérgio Dávila, Editor-in-Chief of Folha de S.Paulo, said "The interest of an artificial intelligence giant like OpenAI in displaying content produced by Folha de S.Paulo and UOL only reinforces the importance of professional journalism." That's a careful non-answer. Publishers are taking the deal because they don't have leverage, not because they're thrilled about the traffic dynamics.
Paulo Samia, CEO of UOL, was blunter: "AI platforms benefit from reliable sources for news. It is only natural that they partner with the creators of high-quality content for that." Translation: we make the stuff you need, so you're going to pay for it. Fair.
What this signals for the next six months
OpenAI is moving fast on publisher deals. France and Germany partnerships came in late 2024, Brazil in early 2025. Expect more announcements in India, Japan, and probably Spanish-language markets (Mexico, Spain, Argentina) soon.
The playbook is consistent:
- License content for training and real-time retrieval
- Give the publisher API credits and enterprise tools
- Frame it as supporting journalism and local relevance
- Move on to the next market
This is smart for OpenAI because it boxes out competitors (Anthropic, Google, Meta) who are scrambling to do similar deals but don't have the same brand leverage yet. It's also a hedge against regulatory pressure—if you're paying publishers and citing sources, it's harder for governments to paint you as a parasitic scraper.
For publishers, the calculus is grim but rational. Traffic from Google is declining, social is a wasteland, and LLMs are eating search. Getting paid something and maintaining a presence inside the new interface is better than being disintermediated entirely.
Open questions
A few things I'm watching:
- Revenue splits: OpenAI isn't disclosing deal terms, but I'd bet they're paying low-seven-figures annually per major publisher. That's meaningful for a single newsroom but not game-changing for a media conglomerate.
- Training vs. retrieval: Are these deals just about showing news summaries in ChatGPT, or is OpenAI also training future models on licensed archives? The announcement doesn't specify, but that's where the long-term value (and risk) lives.
- User behavior: Will people actually click through to articles, or is ChatGPT the terminal interface? Early data from other markets suggests single-digit click-through rates, which would be devastating for ad-supported publishers.
- Exclusivity: Are these partnerships exclusive, or can Grupo Folha also license to Anthropic or Google? That's the real question for leverage.
Bottom line: this is a rational move by everyone involved, but it's also a symptom of a structural shift. When the interface layer captures most of the value and the content layer becomes a commodified input, everyone downstream is squeezed. OpenAI is playing this correctly—paying just enough to avoid legal/regulatory pain while building a moat. Publishers are hedging as best they can. And users? We get better answers in ChatGPT, at least for now.