Sign up for the daily CJR newsletter.

News consumers are changing, and journalism needs to change in order to meet them. At the moment, according to the Reuters Institute, only about 7 percent of adults use chatbots as a source of news, but that doubles to 15 percent for the under-twenty-five crowd. A day is coming, and soon, when people will turn to artificial intelligence for answers to all sorts of questions—despite the host of incredibly well-documented flaws of AI systems in providing accurate, credible, unbiased information. 

That is because AI systems are more attuned than news organizations to an individual’s needs. A small-business owner in the town of Gahanna, Ohio, is going to want a personalized AI-generated answer about how a change in the state’s tax policy might affect her, rather than plow through a long New York Times analysis written largely for a general audience. A parent in Fremont, Nebraska, interested in school board decisions probably wants to know how these affect her child and less about debates among members over diversity, equity, and inclusion. A revolution is coming, in which news is produced on the fly, by generative AI systems addressing the specific queries and needs of one person. The technology is already more or less mature enough to allow systems to do this reasonably well, with largely vetted and accurate information—even if, for the moment, they are doing it very badly.

Our AI-mediated near future is fraught with all sorts of dangers for democracy and civic engagement, not least the possibility that we will disintegrate into filter bubbles of one, and that there will be fewer shared facts and perspectives. But it’s also a world where communities that were never well served by one-size-fits-all news might find information that’s better tailored to their needs and told more from their point of view.

What can or should journalism do about this shift? To answer that, we need to dissect what value journalists bring. Historically, we’ve performed a number of key tasks in helping deliver information to the public: we’ve asked questions, uncovered facts, added analysis, brought in context, created narratives, distributed our work, and engaged with our readers. We’ve done some of these tasks better, and some worse; we’ve often not been great at engagement, and we sort of handed off distribution to tech giants some time ago. We put a lot of our effort—and our identity—into crafting narrative.

But it turns out that large language models are, in fact, not that bad at producing copy. They’re not fantastic tellers of lengthy reported features or gripping narrative essays. But most news isn’t that. Most news is what happened yesterday at the fire down the road, or the vote that the city council just took, or what the president said. Machines not only do a credible job on that front, they can personalize stories at scale and at nearly no marginal cost. And they are getting better at a frighteningly fast rate. 

In a world where the costs of assembling a story drop to close to zero, where is the value created? Most critically, it’s in understanding readers—who they are, what they care about, what they have read, where the news touches their lives. The organizations that know their readers best can personalize coverage for their audience, and that—in theory, at least—buys them not just loyalty, but also trust. Of course, an AI system’s value as a creator of news articles depends on the accuracy of its output. Asking the right questions, gathering the right facts, performing the right analysis, and factoring in the right context are the other parts of the value chain. 

Those are still tasks at which humans beat machines. But as AI takes on more of that work, news organizations will likely become smaller in scale, much more closely knit to the communities they serve. They will be less concerned about getting their work into as many hands as possible, and more focused on providing the most richly personal and useful information to each reader as possible. Outlets that are already tightly bound to their readerships and communities can get even more tightly bound; those that are less so should start investing in needle and thread soon. The APs and Reuterses of this world will probably have a niche in providing global information at scale. Medium-size news metros are likely most at risk—since they’re not big enough to have the scale to turn out great investigations or features regularly, nor small enough to be able to know their audiences intimately. 

A site focused on education in a city, for example, could leverage knowledge of its audience to personalize coverage of schools, classes, extracurricular activities, even teachers for every reader, in a way that larger news organizations—even tech platforms—can’t. I’m always drawn to the example of Homicide Watch DC, which fifteen years ago found a niche in covering every murder in the capital, rather than just the subset the city’s papers deemed “newsworthy.” Back then, it was doing what in theory all journalists can do in an AI age: serving an audience with highly personalized information. 

It’s unclear what a sustainable business model for news looks like in this world. But the existing model is already challenged, and simply turning out more content won’t help anything. Understanding and serving a community well with information they want and need creates a competitive advantage—and value—that offers a smart business team something to work with. To be sure, there are a number of alternative AI-mediated futures, some of them very different and very dark, both for journalism and for society. Where AI splinters shared realities, or governments and companies create misinformation and propaganda at scale. Or where platforms control the relationship with readers and use that to serve their ends, not the public’s. This is a vision for a future where journalism may, ultimately, undermine shared visions of the world, but do better at meeting the needs of every individual—including those who have been underserved by the way journalism has worked in the past. 

In any case, expecting readers to stay the same—or believing we can lecture them into abandoning AI-generated stories—is a fantasy we can’t afford to indulge. To paraphrase Donald Rumsfeld, we serve the audience we have, not the audience we would like to have.

This piece is part of Journalism 2050, a project from the Columbia Journalism Review and the Tow Center for Digital Journalism, with support from the Patrick J. McGovern Foundation.

Has America ever needed a media defender more than now? Help us by joining CJR today.