Is Content King?

How will news publishers make money in an age of AI?

Or more pointedly, will news publishers make money in an age of AI?

Two Stanford MBA students, Alex Webb and Amrutha Nair, wrote an interesting paper back in May — that my colleague Adiel Kaplan helpfully pointed me to — that dives into that question.

It’s a clear-eyed look at the dynamics of the news business and the incentives for different types of news organizations to get into bed with the AI giants. (For example, newsrooms that have leaned heavily into subscription revenue have less reason to cut deals, while wires services, whose content is already widely available on the internet, might see licensing arrangements as extra revenue that doesn’t cut into their core business.)

It calls out the early, hurried deals back in the early internet days that set a (low) benchmark for future deals, and suggests we’re heading into a similar situation now. (The other Tow Center, at Columbia J-School, has a useful tracker on deals and lawsuits.) The paper quotes a media executive:

“The question isn’t whether AI will change how news works – it already has. The question is whether we get paid for it.”

Please dig into the paper; it’s well-worth reading.

I had a couple of main takeaways:

First: by and large, news stories aren’t useful as training data for AI models, which already know a lot about the world (thanks in part, ironically, to having scraped lots of news stories over the last few years.) If news stories have any value, it’s to power Retrieval Augmented Generation, or RAG systems — if you want to know what happened in Nigeria yesterday, simply asking a chatbot isn’t helpful; what you need it to do is tap into a source of news about Nigeria and have it use its language capabilities to summarize a story or stories for you. (Something we’ve also said before, even if that’s not the term I used back then).

News publishers produce information that is — well — new; and if readers (or AI systems) want to know what’s new, they need to go to the publishers.

Except that much of that information is already out there, for free. Everyone has covered the news in Nigeria, and if any given AI system can’t use the news from one newsroom, it can easily get it from another, for free or close to free. True, scoops or deeply reported investigations are exclusive information that only one publisher has — but for how long? Competitors will rush to match or summarize those stories; and in any case, few newsrooms have the resources to turn out world-beating scoops and investigations consistently and regularly.

Second: The numbers are dismal. The paper notes that clickthrough rates for AI search engines and chatbots are between 0.33% to 0.74%, compared to 8.6% for Google search. Add to that the fact that news sites convert around 0.05% to 0.1% of visitors to paying subscribers, and — well, you do the math. This does not look like a sustainable path, unless you’re getting a ton of money for the content you’re licensing.

Third: So what is a path forward? What about information that’s exclusive to a select audience, coupled with deep insights about that audience and its preferences and needs?

If you’re covering a school system, you might not need world-beating scoops; you might just need to know enough about the inner workings of those schools — which schools are underfunded, which principals are underperforming, which special programs are being expanded — to create personalized stories for parents with children in that school system: what the cut in budgets means for their child’s cherished art program, for example.

That gives them content they want, and can’t get anywhere else — not because the information isn’t available elsewhere, but because the publisher knows what each reader is looking for.

In other words, the real value for publishers may be in truly understanding and engaging with a community, and reporting news about things that matter to them, rather than competing for a broad audience. (In a recent Columbia Journalism Review piece, I flag the idea that the newsrooms need to be thinking about serving select audiences rather than aiming for mass appeal.)

The paper points to companies that are offering turnkey solutions to publishers to turn their content into RAGs — Dappier and Miso.ai, for example; a great start, although they seem to mostly focus on advertising revenue rather than driving subscriptions.

Meanwhile, though, depending on licensing news to AI companies may bring in revenue in the short-term, but they risk ceding the relationships with readers that make subscription businesses viable in the long term. As the paper notes:

The overwhelming challenge is discovery: how do publishers find new readers? AI is spearfishing – people search for a particular story or set of coverage. In that sense it differs from social media, where happening upon news you weren’t actively looking for was the name of the game.

In an AI-intermediated news world, content alone isn’t king; it needs to be married to knowledge about the audience. And newsrooms should be leveraging their content to build that knowledge, not licensing it for a short-term fix that undermines their best long-term opportunity

.