Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
When it comes to the rise of generative AI, many in the media industry seem to have learned from their painful experience with online gatekeepers such as Google and Facebook.
This time, they are acting earlier to keep control of their content. But given the allure of chatbots like ChatGPT, do they have any better chance of holding on to their audiences and online revenues than they did before?
A flurry of recent deals — and lawsuits — shows they are at least acting early. This week, Perplexity, a search engine start-up that uses generative AI to produce enhanced results, announced revenue-sharing deals with an eclectic group of publishers including Automattic (owner of Medium), Der Spiegel and Time.
This followed accusations that it was surreptitiously scraping media websites to feed into its search service. That practice is not illegal, though it is a breach of the online etiquette that has governed how information on public-facing websites should be used.
OpenAI has also reached a series of deals with media companies, including with the Financial Times. But along with partner Microsoft, it also faces the biggest legal challenge from news publishers, in the shape of a copyright infringement claim from the New York Times.
In many ways, the economic challenges and legal issues raised by generative AI are familiar from the early days of the internet. Many publishers have long complained that services such as Google News steal their audience — while at the same time relying heavily on them to generate traffic. Courts have given search engines a legal pass, and what direct financial benefit the media companies have been able to extract has come through the political system, with countries such as Australia and Canada passing laws to force the biggest internet companies to pay up.
To some extent, after a free-for-all that apparently led to much news content being sucked up to train large language models, some order has been restored. OpenAI, for instance, has said it will abide by publishers’ requests not to crawl their websites. By the end of last year, most big publishers were already blocking AI crawlers, according to the Reuters Institute for the Study of Journalism.
Simply opting out of the next big technology revolution, however, hardly looks like a durable strategy for media companies. Perplexity’s deal with the publishers amounted to an admission that the AI companies may struggle to defend themselves against copyright claims.
For now, though, much of this debate is purely theoretical. ChatGPT stunned the tech world, but it has yet to show that chatbots can match other mass-market information platforms. The “productisation” of LLMs is still in its infancy. What form these new services take — and how the economic models built around them work — is still up for grabs. That gives the media industry an important opening.
Perplexity, for instance, has agreed to give publishers a share of any advertising tied directly to results that depend on their content. This won’t pay many journalists’ salaries in the short term: Perplexity hasn’t even come up with the new advertising formats yet. And like many search engine start-ups that came before it, it faces a steep uphill battle in carving out a business in a market dominated by Google. But for publishers, it at least establishes an economic model they could promote more widely.
A key question is how much bargaining power they have. Like the internet before it, AI exposes the commodity nature of much online content. It will also be hard to extract much value, if any, for the use of their material in the general training of LLMs. According to OpenAI, the entire news business only makes up a “tiny slice” of the data these models are trained on. The unspoken threat to publishers is: if you don’t play by our terms, we’ll be happy to cut you out.
On the other hand, once trained, LLMs are static and the information they produce can quickly look stale. Techniques that combine fresh and relevant data with the output of LLMs to produce tailored results could fill the gap. For this, access to up-to-date sources of information becomes vital.
How services like these will work is still unclear. Will they produce only snippets as Google News does? How prominently will they show the sources they draw on, and how much traffic will they drive back to publishers’ websites? And, importantly, what extra revenue will they produce and how will it be shared? For publishers, trying to shape the outcome to questions like these seems well worth the effort.
richard.waters@ft.com
