My first major newspaper internship was on the editorial desk of the St. Petersburg Times in the early aughts. I got paid to question state leaders who came by to woo the board, to participate in lively debates, and to write staff editorials that could actually change things in a big, influential state. And when I was done each day, I would drive 20 minutes to one of several world-class beaches and read until sunset.
It was a dream internship by any measure.
And yet my most visceral memory of that summer is sitting in an over air-conditioned office staring at yet another blank page with dread. I’d research and write—haltingly—for hours, only to decide it was all garbage and start again. Sometimes this went on for days. Sometimes the arguments just never quite worked and my editors scrapped the piece.
To be sure, this was partly due to youth and inexperience. But it also reflected the reality that writing good editorials is hard. The ideas have to fit the facts, offer timely or novel insight, and be worthy of the space. You have to wrestle with what you want to say, and often you just can’t know whether an argument works until you hash it out on the page. That’s true for op-eds too.
The Big Idea: That process—that struggle—is how inklings turn into good ideas and how good ideas develop into defensible arguments. It’s also how some ideas end up getting trashed.
That happens because writing an op-ed or essay isn’t just about crafting sentences or giving a piece structure; it’s about drawing on your bank of experience, making sense of research and reporting, and getting feedback from other knowledgeable people. In that way, parts of the process are not all that different from speaking at a conference or debating with colleagues. The writing process helps refine your thinking and ensure your argument is both grounded and worth other people’s time.
But now, our AI co-workers are turning every thought that pops into our heads into finished copy. And we’re all at the beach.
Fewer and fewer people are actually wrestling with what they want to say about complex issues, and the result is now flooding my inbox everyday. And I’m certainly not alone. Some of it is obviously AI slop; some of it is pretty well done at first blush, but the arguments are a bit too tidy, the clever language outweighing the meat.
Often these are opinion pieces pitched by people with impressive credentials, no doubt with interesting things to say. But their perspective gets dulled, or worse, their AI-assisted writing just recycles other people’s perspectives. The Atlantic recently highlighted new research by doctoral candidates at the University of Maryland on “argument collapse,” the idea that essays generated by LLMs have a tendency to converge on a small set of main arguments, sub-points, and paragraph-level structures. As an editor, those findings certainly pass the sniff test.
The authors conclude: “As LLMs are increasingly used to draft public-facing arguments, they may flatten public debate by repeatedly introducing the same polished, plausible arguments.”
That should be cause for concern, especially for readers of this publication. Work Shift is all about how education, training, and work combine to impact economic mobility—and the specific issues we cover are complex. For example, there aren’t a lot of clearcut answers to how we should overhaul job training in the age of AI or to who should pay for credentials that are required for essential but low-paying jobs in healthcare and education. We need all the good ideas we can get.
What we don’t need is more check-the-box punditry.
How We Got Here
We’ve been on that road for a while. Before starting Work Shift, I did a stint as a communications consultant, and a lot of my work was ghostwriting. It’s easy enough to think that having AI write for you is not any different. With the practice at its best though, I don’t think that’s true. Good ghostwriters not only know the author’s voice and viewpoint, but also have their own expertise on the subject at hand. The best op-eds I wrote always involved a bylined author who came in with an interesting perspective and was willing to do a lot of back-and-forth to refine the argument. It was work I was proud of.
Of course, the practice is not always at its best. I can think of times when I gave in and let the author treat me like a human AI, and the result was prose that was fine but arguments that were simplistic or derivative.
Those op-eds probably shouldn’t have gotten published, though most did. And that wasn’t uncommon. Now though AI has turned what was a regular flow of mediocre commentary into a flood, even as it cannibalizes the very traffic all that content is meant to create for news outlets. Perhaps in some perverse way, the technology is doing us all a favor by pushing the status quo past the breaking point.
Polished writing never should have been a substitute for substance.
What We Want
For Work Shift, maximizing traffic has never been the focus. From day one, we set out to publish news and carefully-curated opinion that might not have mass appeal—but that would have an influential and engaged audience. That approach comes with limitations, but it also gives us the freedom to stay focused on what we think truly serves our readers.
That includes publishing commentary that presents fresh ideas, written by people qualified to have an opinion, and done in their voice.
Maybe because of our orientation, we hadn’t been pitched a lot of obviously robot-written op-eds until the past six months or so when it became a deluge. I’m not sure what exactly changed—maybe the loosening standards at places like The Wall Street Journal, evolving norms in the PR industry, or people’s writing habits just catching up to the tech’s capability. But whatever the case, we now need to be clearer:
We expect op-eds, essays, and other contributed pieces that run in Work Shift to be fully drafted by the bylined author(s) or in real partnership with a human ghostwriter or editor, for all the reasons above. And in either case, the bylined author should be the source of the main idea.
That said, we’ve also worked with generative AI for long enough to believe that it can have a limited role in thoughtful writing. We know policy wonks and researchers increasingly use it in data analysis. And turning to AI in the information-gathering phase of a writing project is generally fine; we sometimes tap AI-powered research alongside standard search in the early stages of our reporting. But contributors need to be able to exercise judgment about what’s worthy, facts need to be drawn from the original source, and summaries should be their own.
We’re also okay with contributors using AI as a souped-up version of spell check, prompting it to point out factual errors like misspelled names, as well as grammatical and minor style issues, in otherwise final copy. That doesn’t mean letting it rewrite text. We’re open to hearing other ways people think AI can be appropriately used in the commentary writing process, but lean heavily toward limited use.
And to be clear, we will reject any op-ed or essay that appears to be drafted or edited by AI. We don’t need to be able to prove that AI was used in order to make that judgment, and we also know that we’ll sometimes let things through mistakenly. That’s okay. The point isn’t about policing the technology so much as doubling down on substance. If a piece reads like it was produced by AI, it lacks an authentic voice and viewpoint no matter who or what actually wrote it.
Readers will tune that out.
With the topics we cover at Work Shift, we need people to tune in and weigh in thoughtfully. We’re looking to publish more of those voices. We’ll especially be leaning into pieces that contain original research or draw on on-the-ground experiences. Here are a couple recent examples of what we mean. We’re also happy to provide feedback on a commentary pitch before it’s fully written. But if you aren’t sure if you should even start, two questions are useful:
Is this idea so important that I’d take the time to write about it myself? And why should someone listen to me?