AI Is Transforming Media Forever, Here’s How

I run a media company, and I love writing. So it’s bittersweet to write this, but here it is: AI is going to fundamentally change media in all sorts of ways over the next five years. It’ll make running some newsletters like running a buggy whip business, but it will also open up new opportunities for content creation that we can’t imagine today. 

Given this state of affairs, I’ve been thinking a lot about how AI affects what we do at Every—what are its opportunities, and what are its risks? We want to build for the future, while also respecting the fundamental thing that we created Every to do: create more high-quality business writing in the world. 

In my opinion, the best way to think about things like this is to experiment. So that’s what we’ve been doing. We’ve been building chatbots that make it easier to ask questions of written content. We’ve also been building little tools and workflows, some public and some not, to see how AI fits into our writing process and our strategy. My co-founder Nathan has even incubated an AI writing app. I want to lay out a few things that I’ve learned, and some predictions for what it might mean for the future. 

Transitions are hard, and this one is no exception. I think it’s right to be apprehensive about the impact this will have on creatives of all stripes, and in particular on those who work at larger media organizations that were built for a pre-AI world. For example, screenwriters are currently on strike and one of their demands is a guarantee that AI won’t be used in the scriptwriting process. 

I think these are important issues, and I'm hopeful that Hollywood and its screenwriters can find an equilibrium that uses AI to augment writers and compensate them better—rather than ban it entirely.

Because far from sucking the marrow out of the creative act, I’m optimistic that these tools will become platforms upon which more people can make better creative work, and where smaller teams of people will be able to have a larger impact.

Here’s a short rundown of the ways I predict that AI will fundamentally change the piece of media that I know best: written media.

De-risked content is going to get automated

Media companies monetize creativity, but the creative act is inherently unpredictable. Sometimes an individual or group sits down at a computer or in front of a canvas or behind a camera and produces something earth-shattering. Mostly, they produce crap. Usually, they do it behind schedule.

This makes running a media company hard. If the quality of your product is inherently unpredictable and changes every day, creating stable, growing cash flows is like throwing a leash on a feral alligator and trying to take it on a walk in Central Park: ill-advised.

Because of the inherent unpredictability of the creative act, most of the media business is about looking for ways to de-risk creativity. De-risking strategies are like SSRIs for media companies. Usually, they flatten the upside of your creative work a bit, but they seriously dampen the downsides. De-risking makes it easier to stack the bricks of your business and provide a quality product day after day. There are many strategies that can make this work:

• Summarize other people’s work. It’s easier to summarize something interesting than it is to create it from scratch. Morning Brew, which provides short voice-y summaries of business news, is a great example of this. 
• Find a content arbitrage. Take a piece of content that did well somewhere else and reformat it for your audience. Early Buzzfeed found that if they took popular posts on Reddit and rewrote them as listicles they’d reliably go viral. This is also why Hollywood turns books into movies. Finding arbitrages is easier than finding your own ideas.
• Create a format. Formats give structure to your creativity, which makes creating something new a little more like filling out a Mad Libs and a little less like alchemy. Matt Levine’s format is a few mini-essays reacting to finance news. Ben Thompson’s format is taking a piece of news about a large-cap tech company and then quoting old pieces he’s written about aggregation theory to analyze it. Axios’s format is smart brevity. Every has experimented with this a lot, too, with 3 Shorts and more.
• Make it shorter and shallower. Complexity increases exponentially with the length and depth of your creative output. The shorter and shallower your piece is, the easier it is to make. Not calling anyone out here in particular. 😆
• Get a beat. If you want to get a piece out every day, you have to play in a pond where you can effortlessly react to what’s going on without the need for too much additional research. A beat solves this by creating a box around what you need to know to make something interesting. For example, Ben Thompson sticks to large-cap tech companies, and I stick to what’s going on in AI.
• Make sequels. You don’t have to risk coming up with something new if you’re remixing something old. This is why Hollywood does sequel after sequel. Each sequel might cost a lot of money to produce, but they’d have to make 10 to 100 movies with net-new characters in order to find one that does as well—so it’s relatively cheaper to make a sequel.
• Interview people. It’s easier to have an interesting conversation than to think up new ideas yourself. Have a conversation with someone and then summarize it. I’ve done this a lot for Every.

Interestingly, while each of these strategies de-risks a human’s ability to make pretty good content—they also make it easier for AI to automate. Summary-based media products are easier to automate. Shorter pieces are easier to automate. Content arbitrages are easier to automate. Formats are easier to automate. Interviews, which are primarily summaries, are easier to automate.

I know this is true firsthand because we’ve already done it at Every. Each Sunday, we send out a digest email to Every readers linking to the articles we published that week with little one-paragraph summaries of each one. I’ve written these summaries off and on every Saturday for the past few years—much to my girlfriend’s chagrin. We’ve also had a rotating cast of writers who’ve tried their hand at it.  

Now it’s automated. Lucas Crespo, who runs our ad and course operations, built a little app that does it. You stick a headline, author, and article text into the app and it outputs a summary using GPT-4. The summaries it generates are modeled on previous summaries that we’ve written in the Digest, so it comes closer to our style. Oh, and did I mention that Lucas built this app in Python and deployed on Heroku in two days—and that he has zero prior programming experience? All he had to do was tell GPT-4 what he wanted to build, follow its programming instructions, and ask it to help him when he got stuck.

Even though we have this summarizer tool now, I still spend time editing the summaries we put in the Digest. Occasionally I’ll rewrite them if I don’t think the ones the tool produces are good enough. GPT-4’s ability to summarize is not perfect—yet. But I wouldn’t bet against it over the next year or so, especially as tools for fine-tuning and reinforcement learning through human feedback become more widely available. 

What does this mean? Individual writers will have a lot more leverage to create more media products on their own. The cost structure of creating these media products will look more like software and less like media. There will be some R&D cost upfront to encode your taste and sensibilities into a summarizer. But once that’s done, you’ll be able to let it loose on the world without having to write every single day. The same thing goes for interviewing—you’ll still have to spend time asking good questions (though AI might help even with that). But turning a conversation into a transcript into a readable interview might soon be a far less time-consuming project than it is today. It’s also possible that readers will prefer to own the summarizer themselves and curate it according to their own preferences—à la my co-founder Nathan’s Infinite Article idea.

This means that writers will be able to devote more time to gathering new facts and doing research. They’ll also be able to spend more time writing longer, more interesting articles that are less amenable to automation.

There’s creative upside here, but there’s also a lot of risk for writers who are currently writing more de-risked types of editorial products as part of their day job at larger companies. I can imagine that type of job changing significantly—and soon. If that’s what your job consists of, the best option, in my opinion, is to learn to use these tools today to get better leverage and create good work. It’s a skill in and of itself to get good summaries or interviews or formatted news hits from AI, and your value will go up tremendously if you know how to do it.

Another smart move for writers who are part of unionized workplaces is to apply collective pressure in the same way screenwriters are doing in Hollywood. Media compensation was designed for a world without AI—and it’s clear that compensation frameworks need to change as technology changes. 

The implications of AI in the media landscape extend beyond the realm of de-risking content in existing formats. It will also lead to the creation of new content formats that were previously impossible.

AI unbundles research and narrative

Writers, especially in journalistic organizations, do two things: research and write. Any given news piece in The New York Times is the sum of research, in the form of one or multiple interviews, plus the summarization of those interviews into one cohesive story. 

This bundling of research and narrative is a product of a past where news was printed, summarizing was a rare skill, and there was limited space for stories. 

It comes with real upsides: There’s a significant amount of good judgment, craft, and experience needed to create the right story from a mess of underlying data. It’s also nice that, filter bubbles aside, everyone in your audience reads the same version of the story so there’s a consistent source of truth, and stories can spread more easily. 

But this also comes with real downsides: There are always going to be complaints (real or imagined) that important details were left out and that the narrative tilts too much in favor of one party or another.

This bundling of research and narrative is no longer strictly necessary. Publications should still write their own version of the narrative given the facts they’ve gathered. But they could also expose the research underlying those facts, and allow readers to interrogate that research with AI. 

Imagine a world in which you can click a button next to a news story and see all of the interview transcripts, Google searches, editorial conversations, and more that were used to create the story. Based on this source information, news organizations could make chatbots available that could rewrite the canonical story for you with more details—or answer key questions you might have that went unaddressed in the original. These experiences, when well-created, might constitute new content formats in and of themselves and will require craft to create.

To be clear, this unbundling of research and narrative still allows publications to use their judgment in telling the story they believe is important. But it could help respond to questions of bias in how stories by letting readers know: You can see for yourself how the sausage is made.

It might also make some parts of news organizations look more and more like wire services. As my colleague Evan Armstrong argued, the job of these organizations might shift to uncovering and corroborating new and important facts, as they leave the basic summarization of facts to bots that they run, or to bots that are run by individual users. Writers at these organizations may shift toward more research and investigation, or toward more creative and ambitious editorial projects that require original thinking. 

AI might change how media organizations present narratives. It might also broaden the notion of who is considered a writer.

AI means unwritten stories will be written 

AI will change what writers do at large and small organizations, but it may help more people write. This is an old dream, but it’s getting a new moment. 

Medium, for example, was launched with the idea that any person with one good idea should be able to write a viral blog post. The idea was that smart people shouldn’t need to blog in obscurity for years before they got noticed, and they certainly shouldn’t need a team of editors to gatekeep whether or not their work was high-quality. Just put your idea in a text box, and Medium would figure out who should see it.

The problem is that writing good blog posts is a complex skill. There are actually very few people in the world who can do it well. There are even fewer who can do it consistently. There are even fewer who are crazy enough to want to do it. Writers like me revel in harvesting weird ideas into letters on a page. These are the only kind of people that tend to put in enough hours to actually learn how to make writing that spreads. 

When your supply of writing consists of one-off posts from non-professionals, you tend to get bad-quality writing with surface-level PR-fluff marketing content.

AI might close the skill gap here because it could get good enough to turn people into writers without forcing them to enter anything into a text box at all. 

In particular, it could even evolve enough to turn an incredibly interesting conversation into a good tweet, or essay, or maybe even a book. In turn, it may bring Medium’s original dream a lot closer to reality. 

This will be incredibly important because it will open up new vistas for writing that were never previously available. For example, take the business world. From the beginning of Every, we’ve noticed that the best ideas in business are never written down—they’re locked up in people’s heads. You’ll hear them on a Zoom call with a famous investor, or at an impromptu dinner with a few founders who happen to be in town. These are things said by people with a twinkle in their eye, and that might make you see the world differently or change the trajectory of your business. But these ideas don’t spread because whoever said them doesn’t necessarily have the skills to write about them.

Part of the mission of Every is to bring these ideas to life on the page—but for now, it’s been slow, difficult, and expensive. We do a lot of interviews with people for this purpose, and we’ve also dabbled in ghostwriting. But it’s a fundamentally talent-constrained endeavor. Good ghostwriters are incredibly rare, and they can only work with a few people at a time. 

AI changes this equation significantly. We’re already starting to see how AI makes interviewing easier to do: It’s brought down the cost of producing a basic transcript significantly. We’re starting to tinker with doing more on this front, using it to take a basic transcript and polish it into a more final form. It’s still early, but it looks like it could work. If it does, it might be able to help with the even more complex task of ghostwriting. This will dramatically expand the number of high-quality posts we can publish from people who don’t consider themselves writers but who have incredible ideas about business. I can see this holding true in a lot of other subject areas. 

We’ve covered a lot in this article, so let’s pull back and take stock. If AI is able to automate certain types of content creation, unbundle research from narrative, and help to write unwritten stories…what happens then?

The robots are coming, but will we like them?

The idea to use machines to automate parts of the content creation process is not new. 

For example, in BuzzFeed’s early days, CEO Jonah Peretti created a bot called LilyBoo that would help him trawl through Pinterest posts looking for maximally cute pictures of dogs, cats, and more. The pictures were then gathered into a list format and published. 

According to the excellent new book Traffic, by BuzzFeed’s former editor-in-chief Ben Smith, the idea for LilyBoo was, “machines, with a little human help, feeding culture back to itself, scaling to infinity.” LilyBoo, though, never quite took off. The problem was that there was “something sterile about simply, clinically, feeding people’s own social media posts back to them.”

This new generation of AI will face the same problem. GPT-4’s summarizing capabilities might be quite good—but they still lack the zing and voice of truly great human writing. Great writing—even when it’s de-risked—is still a complex activity where what it means to be “good” is changing all of the time. AI is still far off from being able to do that on its own. There’s a good chance it will get there over some long, undefined period of time, but anyone who uses these tools day-to-day can see they’ll need a lot of help for the foreseeable future.

I’ve argued before that AI will probably change the set of skills you need in order to be a writer, but that it won’t get rid of writers. That’s why I’m optimistic about approaches that intelligently blend what AI can do and what humans are great at. In an ideal world, these technologies give humans more time to do satisfying creative work at a higher level of ambition and scale and also make media businesses significantly easier to run predictably.

We’ll see how it goes!