Credit: Adobe Firefly
OpenAI and Axel Springer just announced a partnership to allow ChatGPT access to news stories from some publications. With the agreement in place, ChatGPT users will get answers that tap news from Politico and Business Insider as well as German titles Bilt and Welt — including paywalled content — along with links to the original articles.
As is the norm with this sort of thing, the exact terms of the agreement weren't disclosed. But any money changing hands is going from OpenAI to Axel Springer to grant access to its news. And it's a given that only ChatGPT Plus users will be able to access the feature since the free version can't even browse the web.
The feature wasn't available as of this writing. When I queried what's happening in the U.S. today, ChatGPT Plus spat back the answer in its standard list format, citing mainly Reuters and Al-Jazeera as sources. Even when I asked directly what's on Politico today, it told me it couldn't access the site.
How this partnership is structured from a technical standpoint is relevant here. From the sound of it, this isn't simply an enhanced version of Browse with Bing, the existing feature that lets ChatGPT access recent news. Today, when you ask something about current events, ChatGPT looks for recent articles and summarizes them as best it can — as long as the sources haven't blocked it (just try asking ChatGPT what's on the front page of The New York Times).
Without knowing exactly what OpenAI has put in place, I suspect the feature relies on retrieval-augmented generation or RAG. With RAG, a large language model is enabled (and even required) to query specific datasets for the most up-to-date information before returning an answer. Not only would that allow ChatGPT to bypass Bing and directly access a publication's depth of content on a specific topic, but the throughline to the original information is much clearer than when the answer is based solely on training data, meaning it will always be able to cite a source.
Another open question is whether OpenAI can now use Axel Springer's content archives as training data. That feels like giving away both the cow and the milk, so I suspect Axel Springer is only allowing ChatGPT to use its content for RAG queries as long as the agreement lasts, but is forbidding the training of future models on it. We have queries in to Axel Springer for answers to these questions.
A bigger question for the media industry is how do we measure success here. Although this isn't the very first media deal that OpenAI has made (it partnered with the Associated Press back in July), this one feels like it has more potential to set a precedent. With many publishers' social and search traffic dwindling, ChatGPT's claimed 100 million weekly users provide a tempting alternative.
What OpenAI is doing for Axel Springer is going to be table stakes in the future. Any organization worth its salt is already researching how to ensure its content becomes fodder for future generative AI. Further, Axel Springer giving away what amounts to a dragon’s horde of content is deeply shortsighted. The company, with a fairly small investment (about $10 million according to one developer we spoke with), could have built their version of whatever chatbot that will come out of this partnership while OpenAI will have a million new data points to fold into its already deep bench of content.
Think about what’s happened: Axel Springer will get a chatbot that nobody uses that brings their content to the fore. This is basically like Google but without certainty.
If referrals are the idea — or even part of the idea — that strikes me as misguided. Clicking through to the source from an answer that's engineered to negate the need for that click is likely something only researchers and journalists will do. The only way this makes sense from the media side of things is if the deal is lucrative enough. Whatever OpenAI is paying Axel Springer, will it be enough to sustain, and perhaps even grow, the original journalism that's supplying those answers?
That's the big question that the future of news media hinges on. Journalists can adapt to writing for chatbot users instead of readers. But if they don't get paid, eventually ChatGPT won't have a lot to say about what's going on today.
More from the podcast:
Like this post? Why not check out The Media Copilot podcast (listen on your favorite platform here, or via Apple Podcasts below)