The scientific "study story" has a questionable future. That may be a good thing

The way we do science journalism is going to change

Display of food science journals from a library at the University of the Basque Country. By Vmenkov - Own work, CC BY-SA 3.0, via Wikimedia Commons.

A lot of journalists hate how AI is going to change journalism, but it is going to happen. I may not approve, but I am also not naïve.

Here, I want to write about one of the changes that I see as a possible, or perhaps even likely, outcome. The following represents my somewhat delayed — but thorough! — reaction to recent news about AI filtering directly into journalistic writing, including use of AI in the writing process at the Cleveland Plain Dealer and a story in the Wall Street Journal about a Fortune journalist named Nick Lichtenberg, who according to the story, is using AI to rapidly generate stories, sometimes multiple ones per day.

Okay, so: Prediction. On the climate beat, and on the science beat generally, I think we’re going to see a decline of the so-called “study story,” or at least, its human-authored version.

When I say “study story,” what I mean is a journalistic instance in which a reporter writes a story largely or solely about one new piece of scientific research, almost always timed to come out right when the research comes out. This in turn often means that the research was embargoed and the journalist had prior access, with agreement not to publish before a certain date and time. Which, in turn, means that study stories are mostly written about studies appearing in major journals that have their own embargo systems and publicity machines, like Science, Nature, etc.

Some may say the demise of this story form would be a good thing, which we can debate. I’ll provide plenty of fodder for this debate below. But the overarching point is, the form is AI-vulnerable, in the sense that it is getting too easy for robots to replicate; or perhaps rather, for humans to produce rapidly with heavy amounts of AI assistance.

(Not that anyone should be writing journalistic stories with AI, but I now fear it may be unstoppable, in the sense that at least some outlets will decide to go this way.)

To show as much, let’s first go over in more detail what this story type is, and how it can be problematic, before thinking about what may happen to it in the future.

The scientific study as a news event

Scientific study stories generally unfold along the lines of, “New study finds X,” quoting the researchers who produced the work and, at least in the more credible versions, some more neutral experts, who were able to review the research as well. Such stories are often written in the (itself rather robotic) “inverted pyramid” format, with the finding of the new research presented immediately at the top and treated as news. Then implications of, and perspectives on, the research appear lower down, along with whatever other context is deemed helpful for better understanding the new research.

In its best version, the study story serves as an opportunity to explain, with strong writing, how some important aspect of the world (or, of ourselves) works, and how a new investigation illuminates that.

“I still assign one to my students every semester, because I think there’s really value in digging through and learning what statistics there are, and how to parse that, how to find good sources,” said doctor and medical and science journalist Ivan Oransky, the co-founder of Retraction Watch and editor-in-chief of The Transmitter, and a journalist in residence at the Carter Journalism Institute at New York University. “So as an exercise I think it’s still very valuable.”

However, Oransky has his concerns about the model or how it is often implemented, as do many scientists and science journalists. The critiques include: 1) that new studies that seem very novel and striking (and thus garner media attention) are often wrong; 2) that journalistic coverage tends to hype individual studies, and rely too heavily on self-interested press releases in doing so; 3) that scientific understanding depends on consilience, not any one study; and so on.

These issues were well summarized by Brian Resnick, writing in Vox a number of years back:

The problem is that each headline gives an incomplete glimpse of how science works. One lab produces a result. Another lab — ideally — tries to replicate that result. Rinse and repeat. Eventually someone needs to do a meta-review of the totality of the evidence on the question to reach a conclusion. That meta-review, rather than any one study in isolation, is likely to get closer to the true answer.

So, yeah: You often need multiple studies, getting at something from different angles and with different methods, to really know if it is true. You need meta-analyses and reviews and assessments of the evidence for knowledge to truly be logged. That means single study stories, especially in the health and biomedical arena, can be just misleading.

‘“You can always get traffic by contradicting the previous headline that you wrote,” said science writer and editor Corey Powell, former editor-in-chief at both Discover and American Scientist. “From a commercial standpoint, there’s a value to it; there may even be a certain ethical value in showing how science is this continuous process. But a lot of the effect is, people conclude, these researchers just change their minds all the time. I think there’s a low-level cynicism that that kind of coverage generates.”

I have written many single study stories myself. I have done so being very aware of these criticisms, but also feeling that a study story done very well is a study story worth doing. I still mostly think that. As time wore on, though, and I became increasingly focused on data journalism, I began to develop an additional concern about the story model, which I’ll unpack in the next section, before getting to the new AI problem.

The big bias in which studies get covered

Scientific journals at the University of North Carolina Health Science Library. By Selena N.B.H., via Flickr, CC BY 2.0.

To dig a little deeper here, I think it helps to ask the question: What does the science writer think they are doing in the first place?

It usually boils down to some combination of personally loving the subject and wanting to fascinate others in the way you yourself are fascinated; plus a vague sense that the “public” needs to know more about science, for lofty Enlightenment-style reasons. (I would note that the latter view may reflect problematic deficit model assumptions.) And so the science writer starts producing lots of words about science. And some writers learn (especially if they are web traffic conscious, as they will certainly be if employed full time at a newspaper or magazine) that scientific studies produce an endless stream of potential content, providing the opportunity to explain research but also to frame it in interesting ways that draw in readers.

Deficit model issues aside, I don’t think it is wrong for a writer to want to write about science, out of passion and interest, or for that writer to latch onto scientific studies as one way of doing so. The problem, however, is how this can lead you to become tied to a particularly selective and limited information stream.

Look, this is global science, and it is massive:

Not only has science become increasingly international in the past several decades, leading to a very large increase in the volume of publications; but scientific journals have also proliferated rapidly. From less than 100 in the year 1900, the number of “mainstream” journals rose to just under 4,000 in 1980 and now number over 9,500, according to researchers David Baker and Justin Powell, authors of the recent book Global Mega-Science, and their colleagues.

“There are more and more studies coming out, there is more and more access to studies, so the arbitrariness of which get covered, it’s always been there but it is more and more evident,” said Powell (the former Discover editor).

Let’s just take the climate change beat, and science publications therein. I published this chart early on in the annals of this Substack, but have just updated it with one more year of data (2025, the highest year yet):

By this metric, annual climate science studies now number over 15,000. And the chart above is the result of a fairly conservative method for searching the Web of Science database for studies. You can get a much larger total if you search through more detailed article abstracts instead of just article titles, for instance (title searches comprise a key part of the approach shown above).

The more you write about a field of science, then, the more you realize there’s an enormous amount out of research out there, and you’re missing most of it. You also realize that the journals that have embargo systems, and thus tend to get your attention, only publish certain kinds of things. And their contents do not span across all that researchers read or care about. In the climate field, for instance, I soon learned that journals like Geophysical Research Letters or The Cryosphere or Earth System Science Data were packed with fascinating studies that rarely get media coverage.

This bias, in which journalists only cover a small fraction of all published studies — and generally choose to cover studies in the big name journals — was formally documented in this paper by Marie-Elodie Perga and colleagues in Global Environmental Change. They looked at over 50,000 climate science studies published in the year 2020 alone (almost certainly finding many more studies than the above chart shows thanks to the use of a less restrictive search approach). Then, the researchers used Altmetric, a service that measures scientific impact based on online and media discussion (rather than the traditional measure, citations) to gauge how much media attention each study got.

The paper found that only a small fraction (2 percent) of studies received “extensive media attention” and that these were from a “restricted subset” of journals (about 13 percent of all journals publishing in the climate space). Forty-one percent of the coverage focused on studies in just 6 journals: Three Nature journals, two AAAS journals (Science and Science Advances), and Proceedings of the National Academy of Sciences.

(Wait, did I just write about a single scientific study?)

The point is, there is a bias in the study story genre. It does not cover all research, nor is there much evidence of systematic thinking about which research really, really matters to the audience, or world. Rather, the genre is tied deeply to the biggest scientific journal publishers.

So: I’m not sure there is anything wrong with writing about a study when it fascinates you. But if you’re so fascinated, you should read widely and write about studies in all types of journals relevant to your beat. This also has the benefit, of course, of saving you from writing about the same research that everybody else is writing about. (At least until your editor shows up and says, “Why don’t we have that story about the study in Science that everyone else is covering?”)

Too easy for AI to replicate

But now, the study story is facing a bigger issue, perhaps, to its viability. AI.

Journalists wrote lots of study stories because they would often go viral and get lots of readership. And, they could be done on a roughly daily basis. After all, there was only one text to read (or, I fear in many cases, skim) and then you made a few phone calls or sent a few emails for outside perspective. Boom, 800 words.

This is exactly the kind of journalism that AI can largely replicate, and it can produce more study stories, faster, than any human can. The one thing it can’t replicate are the outside expert quotations, of course. As long as those are freshly gotten by a journalist actually interviewing someone directly.

But honestly, in many cases, with the bigger studies that get released, there are now embargoed press conferences with transcripts or recordings made available. So at least for the main authors, these quotes can be fed into the machine too. Also, some press offices or services provide outside expert reaction to new research that everybody knows is coming, and is going to get a lot of attention…so, yes, AI can get those too. Especially if some journalist feeds them in.

It all suggests a world in which some journalists may become very good at grabbing potentially newsworthy texts (like, studies) and using AI to very rapidly frame and craft stories about them. I don’t like this world, but the barriers to entry are not high. It may already be here. (Although whether such “reporting” can be considered copyrighted work is, at minimum, problematic.)

In such a world, while there would likely still be some journalists out there doing study stories well, and in an all-human way, their work could find itself afloat in a sea of AI generated content.

“There’s more and more outlets that churn content out of press releases, instant coverage of studies, and now instant coverage of preprints that aren’t even peer reviewed,” said Powell. “The slop factor far predates AI slop, AI slop is just one more amplification.”

Going beyond the “archive”

I certainly do not mean to say that all study stories will become moot, not worth doing, not worth existing, etc.

There are certain biomedical study stories that report on large, multi-year, fundamental trials that do indeed deliver to us big news about choices affecting our health. There are certain studies in other fields that represent undisputable news, or true breakthroughs.

And there are many other nuances needed to the argument. In climate science, for instance, there are certain large, multi-author publications of new datasets compiled over many years, or new compilations of large amounts of evidence, that might appear in a single “study” but at the same time, carry extra scientific heft. Also, there is coverage of mega-reports, like those from the U.N.’s Intergovernmental Panel on Climate Change, that can read like a study story but are really much more than that. That’s because, while the story covers a single text, that text in effect synthesizes enormous numbers of studies.

After this story first published, I heard from another long time science journalist — John Schwartz, formerly with the New York Times, now at the University of Texas at Austin teaching the field’s future — who had another important point. He wrote, and I’m quoting with permission:

The less obvious problem here is that study stories help new science journalists climb the learning curves of science writing — it's the preparation for the deeper explorations to come. If young journalists aren't picking up those basics through this important practice, it will be hard to move on to bigger things, and harder to maintain a pipeline of promising science journalists.

I agree with this!

So, for a variety of reasons, this approach, the single study story, is not going to entirely go away. But going forward, it probably won’t be the way that a writer distinguishes oneself, either. If AI-generated study stories proliferate, I certainly imagine this would dissuade a lot of the best science writers from just piling on.

So what then, as writers, should we do?

Well, we write different, more inventive, more human-driven stories. We use narrative approaches. Most of all, we write stories that contain unique, new information, rather than those reliant on some tokenizable text that many journalists — and, perhaps, many LLMs — have access to.

My onetime professor and Princeton historian D. Graham Burnett wrote a striking story in The New Yorker last year about how AI is transforming thought in the classroom, and what it means for the humanities. A lot of academia has been dismissive of AI; not Burnett, who shows how its capabilities startle both himself and his students. Along the way, Burnett memorably refers to what the models contain in their training data as the “archive”: They have consumed all or virtually all of existing thought and learning. That is how they do their magic.

But let us also remember that the “archive” is all the models have. The writings and records of the past. No matter how dazzling their combinations of elements from this archive, they are never the sources of entirely new information. This means that any original human reporting, any dusty scholarship involving actual records, anything else at all that is not in the “archive,” cannot be matched. In other words, journalism in this traditional form of reporting will remain continually valuable.

“Just be a journalist,” advised Oransky. “Just actually be a journalist.”