One of the hallmarks of the early AI is that it would tell you something but not credit what it had trained on in order to tell you that it was all based on real people's work, but those people were never credited. You can't have AI without source material. And you can't have the source material be public without traffic and compensation.
Hello and welcome to AI Curious. My name is Jeff Wilser. I'm a journalist. I'm a human and I am curious about AI. And let's start with a blunt question. What happens when AI can read the whole internet but the internet stops volunteering? Because right now we are in a strange new phase of the web. For the last twoish decades, there is this informal bargain. Google crawls your site and in return you get traffic. Not perfect, not always fair, but it was an exchange of value. With generative AI, a lot of publishers and a lot of writers feel like that bargain got broken. The AI systems trained on mountains of public writing, often without asking, often without crediting, and often without compensating. And now, as AI summaries become a default layer between readers and the open web, the fear is that the incentive to publish in public gets weaker and weaker. So, what do we do about that? Do we just accept that? Do we do licensing deals one by one? Do we block the crawlers? Do we move our best writing into private spaces — Discords, group chats, newsletters where it's harder to scrape? That's at heart what today's episode is about. And we're joined by someone who is not just theorizing from the sidelines. He is in the arena dealing with this in real time. And he is one of the most important voices in this battle. My guest is Tony Stubblebine, the CEO of Medium. Now, Medium is obviously one of the most important writing platforms on the internet, and Tony has been one of the more candid voices on what he calls AI's broken social contract with the web. In this conversation, Tony lays out a simple framework he calls the three C's: consent, credit, and compensation. We talk about why publishers are pushing back, what a realistic internet standard for AI licensing could look like — something that doesn't just reward the biggest players — and why the future might not be dead internet, but something more specific, a dead public internet where the real human stuff retreats into smaller, more private communities. And then we end somewhere I did not fully expect, a kind of pragmatic optimism. Tony argues that all of this pressure may force a new equilibrium because AI needs human source material and humans need incentives to keep producing it. If you're a writer, an editor, publisher, founder, creative, or just someone trying to understand what the internet turns into next, I think you'll find this one relevant and refreshingly non-binary, not good or evil. So, with that, please enjoy my conversation with Tony Stubblebine, CEO of Medium. Tony, welcome to AI Curious.
I believe you have said that AI has broken a social contract with the internet. Let's start there. What does that mean? What is the contract that we have with the internet? And how is AI breaking that contract?
I mean, it means get your popcorn because there's a bunch of fighting going on between big companies and small companies and medium-sized companies like the one I run. Sometimes on the internet companies can be really collaborative because there's a natural exchange of value. And in this case there was no exchange of value — actually something worse than that, where the AI companies kind of came out of the blue having scraped and consumed and trained on a lot of content from every media site including Medium, where millions of people write, without offering anything in exchange. So what we're used to, which is kind of the basic social contract of Western civilization, is that if you get something you give something, right? And we are specifically in that relationship with Google where they crawl our sites, in exchange we get traffic from people who search on Google. That wasn't happening with the first version of OpenAI or whatnot. In fact, our view is it was actually quite a bit worse — that they were making money and in exchange we were facing an onslaught of spam and slop that cost us money to fight back. And so what happens? What does a company like Medium or the New York Times or Reddit or Quora — what do they do in that situation? And that's I think kind of the drama of the internet right now — that we're not going to do nothing. And so we have to respond to that.
Can you kind of give the blunt version of what OpenAI, Anthropic, and Google — those companies — what did they do wrong in your view? What was the chief problem and crime?
I think the simplest way to frame it is what we call the three C's, which actually had been used in other places before we started using it. First is consent — ask for permission, allow us to say no. The other is credit. One of the hallmarks of the early AI is that it would tell you something but not credit what it had trained on in order to tell you that the generative AI was all based on real people's work, but those people were never credited. And then the third is compensation, which could be money, but I think Google Search has shown that most of the internet would rather just get free traffic — we know what to do with free traffic — but some sort of compensation in there. And if you're not doing that, then I think the AI companies should expect to be cut off. And in fact, they have. There's been a retreat from the public internet that had allowed them to do this. And so I think they're not going to be able to do it as much in the future. And then there's tricky stuff where the AI crawlers that come to your site, sometimes they get blocked. But if one of the big services that a lot of us use, Cloudflare, they allow us to — if we catch that it's a crawler — to just start poisoning the results with weird gobbledygook, right? And that's the kind of world where — that's why I say get your popcorn.
Like if we don't come to a collaborative solution it's going to be a fight. A point you've raised before — that given the incentives we have now, which are questionable at best, to write on the internet — you predicted we would see a retreat of writers writing in more private forms. Can you elaborate a little more on that? And are we starting to see that? What evidence do we have that's actually happening?
And no credit to me for saying this — we had a dinner of what we call interesting people in New York where we arranged all the AI people we could find — people building AI products, people from media, journalists, journalists covering culture. Everyone agreed there's a retreat from the public sphere. And it started before AI. It started because Twitter just got so toxic that people realized I'd rather be somewhere else. And so if you start with Twitter as the trend, there's an exodus from Twitter and a lot of people think that exodus went to Bluesky or went to Threads. I would point out that a lot of it went to group text messaging and Discord — safer, more private spaces. And so I think that's accelerating because Google is sending everyone about half as much traffic is what I say. They say something different, but the click-through rate dropped in half and it dropped in half when the AI search summaries showed up. So essentially half of the traffic we used to get now goes to an AI search summary.
Wow. Half the traffic. So you're saying half the traffic from Google now?
Yeah. Well, the way we defended our traffic — but the click-through rate dropped in half and it dropped in half when the AI search summaries showed up. So essentially half of the traffic we used to get now goes to an AI search summary.
So, okay. So half of the incentive that they were providing to prop up this global town square — what some people call these social media platforms — this idea that I want to contribute to some global collective intelligence, one of the original inspiring ideas behind the internet — well, the incentive got a lot weaker for that. And so now people are thinking: what about something else? What about trust? What about connection? What about real community? And those things are kind of better done in private spaces.
And look, that's a kind of natural reaction to the changing incentives and it's also a real threat to the emerging AI world because what we see from generative AI is that it is relying on access to real information from somewhere else, right? The way that AI search works is that there's an LLM that does all the generative language, but it pulls source material — sort of a RAG approach is what it's called — and then summarizes all of that. Well, if it can't pull the right source material, if the product review doesn't exist or if the product reviews are themselves all AI generated and the real product reviews are hidden in some private Discord, then it becomes suddenly a lot less effective.
A brief digression here. What does that internet look like if we follow that to its logical conclusion and we become more and more kind of cloistered off in private conversations? Sketch that out. What does that world look like online?
I think I've switched to be more optimistic and I think this might even be jarring for this podcast where we started out being like these people are crooks, they've broken the social contract, F them. But actually now that you think about it and now that we're in it, there was already something kind of unhealthy about the internet before AI. It's alien — it's dehumanizing to perform in public but not for other humans. You're performing for the algorithm. How much do you change what you say so that it goes viral, so that it does well in search results? You lose track of your audience in a way that is often unhealthy. And so what we're kind of rediscovering is that smaller spaces are much more connected. They're much more communal. They're much more — you can see the person in some way. And there's a person that you're speaking to rather than an algorithm.
And we've had these — I mean this is like a pendulum of the internet. This is maybe Usenet in the early days. This is maybe like what AOL was for people.
I even remember BBS boards, right? I mean public but it feels very kind of private and communal.
I'm of the age where I was on a BBS.
Oh, me too. I still know my names.
That's right. And I met those people in person, right? And that's not true of — you know, I've had, I think my biggest hit on Medium had two million views, right? It's like I don't know any of those people.
And we didn't meet at the park, right?
So, to what extent is Medium shifting or evolving to adapt to that potential future for a more private world?
I mean, so many ways. Although I think we're probably actually more strongly responding to something else — like I don't know, the way I put it just a few minutes ago to someone I was talking to at work is: if we don't champion writing, who's going to champion writing? Every time there's always some trend in media to pivot to video and here we are people that just love the value of writing. This isn't just like we're championing writers, we're championing the idea that everyone should be writing as much as possible. And so we had been looking at Medium and we've been like, wow, we've really set the bar really high. To do well on Medium, it means you have to have written something that you feel comfortable putting out to the entire public, that you think it's going to withstand scrutiny, and that it's worthy of being read by millions of people, right? Well, there's a lot of things I have to say and that you have to say that aren't that. Or maybe 17 people should read, not two million people.
Maybe 17 people should read, not two million people in my case.
That too. And the way it comes up at work is actually the founder of Medium, Evan Williams, who I just call my boss — he's the chairman of the board and I'm the replacement who took over when he stepped down as CEO — we will say to each other a lot of our best writing was just to each other. And so that idea that yeah, how do we make it possible to write to two people or three people or 17, or to write without that public scrutiny, or to work something out, or to say something that is not safe to say in public? How can we do that? So we're already essentially saying this what you see on Medium right now is the right edge of where we want to be and we're not going to go further than that. We're not about the creator economy. We're about the expert economy and everything amateur to the left of that. Because just like I see point one percent of the world could make a living as a writer and 99.9 percent of the world should be writing. That's who I want to talk to. That's who I want to help express themselves. That's how I want to get their life experiences out into the world to connect and help some collective intelligence that could be global or it could be just communal. So this is our right edge and then how can we make it possible for more and more people to write. For us it means publications — which are kind of like, think of it like a subreddit, a group of people with a shared interest — and then we have essentially a private social network, private publishing in the works. But you know, I hate to give vaporware, say on Monday the beta testers started using it for the first time. So imminent feels like too strong of a word.
You heard it here first folks on Tuesday. It's a hard rollout. It's happening.
So we started with kind of the core problem that you articulated with AI and how they train these models and the three C's. Let's pivot a bit to solutions. One obvious possible solution out there is the kind of licensing deals that the New York Times, AP, and Reddit have used. As I understand it, you have a different framework in mind. Talk a bit about why you might not feel those solutions are the right way to go and what you'd prefer to see instead.
Yeah, I mean I feel like we're doing the AI companies a big favor by just handing them the solution. And I say I resent that they didn't lead with this. But the solution is that there has to be an internet standard that governs how to pay for content. And if the AI companies are going to consume it and use it, they have to give some form of compensation. And this should not be done through ad hoc negotiations and it should not be run by a single company. So right now I think Cloudflare is a company I really respect for what they're doing in AI.
Can you quickly, for listeners who aren't as plugged into that, describe what Cloudflare is doing with AI right now?
This is one of those companies that if you host a website you know what they are and if you don't you would never know. Essentially, they're one of our service providers that sit in between us and every person in the world, kind of filtering what can get through and sometimes caching it so that it gets delivered to you faster. But as a result, they often are in the place where they're best able to block the AI crawlers for us. And they've done this thing which I think is the right thing to do — they've said the AI companies will not come to the table unless we first force them. And so they first made it really easy to block all of them all at once at a really massive scale because Cloudflare is a company that we all use. And I totally agree with that sentiment. And then they said, now that we've done that, we will negotiate for payment on behalf of our customers. And that's where I think that's the best current available option. But it doesn't make sense that we would have vendor lock-in for this type of negotiation. And so more recently, there's finally an internet standard called the Really Simple Licensing standard that attempts to have a standardized way of communicating: here are our license rules and here's how to contact us. And eventually what that standard is working up to is essentially a licensing body that handles all of the financial transactions in the way that if you've ever heard of ASCAP on the music side — like if your barbershop is playing music that musician is making money and the organization that kind of makes that happen is ASCAP. And so we want the same thing for AI.
