Paywalled Creativity: What Happens When New Knowledge Stops Being Free

How the economics of AI search could quietly put a price tag on every new idea, and leave the rest of us reading scams. Here's a thought experiment. You discover something. A genuinely new thing: a research finding, a clever open-source library, an essay that reframes a problem, a creative technique nobody's tried. In 2015, you'd have done the obvious thing and published it. A blog post, a GitHub repo, an arXiv preprint, a tweet thread. You'd have given it to the world, because giving it to the world was how you got credit, reputation, citations, a job, a following. The web ran on a simple bargain: publish freely, get attention, convert attention into something valuable later. Now run the same thought experiment in 2026. You discover the same thing. Except now you know that very soon, an AI training will ingest it, and within days, the model will be reselling your insight, unattributed, unlinked, stripped of the path back to you, to millions of people through a $20/month subscription. You did the work. Someone else monetizes it. Your reward is that you got to feel briefly useful. So you pause. And you think: maybe I should sell this instead. That pause is the whole article. Because if enough people pause, and the incentives say they will, the open web as a place where human knowledge accumulates for free starts to die. And what replaces it is worse than most people realize. This is a sequel, of sorts, to my piece on following the money in AI. Follow it here too. The money explains everything. The Bargain That Built the Web Is Breaking For thirty years, publishing knowledge freely was rational self-interest, not charity. You wrote the blog post because it built your reputation. You released the open-source library because it got you contributors, a job offer, a consulting pipeline. You published the paper because citations were currency. The system worked because attention flowed back to the creator. Free publication and personal benefit were aligned. AI search breaks that alignment at the root. When a model answers a question by synthesizing your work into a tidy paragraph, the user never visits your site. They never learn your name. They never become a contributor, a customer, or a citation. The attention that used to flow back to you is intercepted and absorbed by the model. You still did the work. You just don't get the thing the work was supposed to earn you. The numbers are already moving. By 2024, automated traffic crossed 51% of all web activity , the first time in a decade that bots outnumbered humans online. An SEO firm that analyzed 65,000 articles found that by mid-2025, roughly half of all newly published articles were AI-generated . Publisher referral traffic from search is collapsing as AI overviews answer the question on the results page and the click never happens. And the people who produce genuinely valuable knowledge have noticed. There's now a name for their response: the Human Silo effect , high-value human insight retreating behind paywalls specifically to escape the crawlers. The useful web is "darkening" to public access, not because anyone decided it should, but because the incentive to keep it open evaporated. The New Default Is "Sell It" Watch what's already happening at the top of the knowledge economy, where the actors are sophisticated and the money is real. Academic publishers, the people sitting on humanity's research output, did the math first. Taylor & Francis was expected to make around $75 million from AI licensing deals for academic publications in a single year. Wiley booked roughly $44 million. Oxford University Press is reportedly working toward the same. These aren't rogue actors. These are the institutions that control the gate to peer-reviewed human knowledge, and they have discovered that the knowledge is worth more sold to a model than given to a reader. The news industry is doing the same thing at speed. OpenAI alone has signed something like eighteen licensing deals with publishers. Condé Nast, Vox Media, the Financial Times, the Associated Press, Axios: the content, including paywalled content, is now an asset to be licensed, not a public good to be read. Where publishers aren't licensing, they're suing. The New York Times, the Chicago Tribune, and Penske Media are all in court over exactly this. Here's the part that matters. Those deals and lawsuits are the visible tip. They're the actors big enough to have lawyers. The real shift is happening one rung down, in the heads of millions of individual creators, the blogger, the indie researcher, the open-source maintainer, the essayist, who are quietly arriving at the same conclusion the academic publishers reached: if my work is going to be monetized by a model regardless, I should be the one monetizing it. The default action for new knowledge is shifting from "publish it" to "price it." And defaults, once they flip, are very hard to flip back. The Two-Tier Internet Now extend the trend line. If the rational move for anyone with valuable new knowledge is to put a price on it, the AI labs face a fork: buy the knowledge, or get stuck. The well-funded frontier labs will buy. They already are. They'll license the journals, the news archives, the premium data, and their flagship models will stay current and sharp. But those models are not what most of the world uses. Most of the world uses whatever is free: the cheap tier, the ad-supported tier, the default assistant baked into a phone or a search bar. And the free tier is precisely the one that can't afford to keep buying fresh, high-value, paywalled knowledge at scale. So you get a two-tier information system. The expensive models, used by a minority who can pay, stay connected to the frontier of human knowledge. The free models, used by the 90% who can't or won't pay, slowly fall behind, trained and re-trained on an aging, picked-over, increasingly synthetic public web, because the good new stuff went behind walls they can't reach. We have spent thirty years living inside an extraordinary historical anomaly: the near-total democratization of the world's knowledge, available to anyone with a connection, for free. Wikipedia, open-access papers, Stack Overflow, a billion blog posts. We treat it as the natural state of the internet. It was never natural. It was a temporary equilibrium held up by an incentive structure that AI is now dismantling. The luxury of public knowledge could turn out to be exactly that, a luxury, and a brief one. And here's the cruel arithmetic of the paywalled version: even for those who can pay, nobody can pay for everything . When every journal, every archive, every quality publication sits behind its own subscription, the practical result isn't "knowledge for those who pay." It's "a thin slice of knowledge for those who pay, and no one, however rich, can read across the whole field anymore." The cost doesn't just exclude the poor. It fragments the rich. The Deeper Problem: AI Is Derivative. Creativity Is Not. Now we get to the part that actually keeps me up, because it's not an economic problem. It's an epistemic one. AI is, by construction, backward-looking. A model is a compression of everything that has already been written, said, and coded. It is extraordinary at recombining the past, at giving you the median of all prior human thought on a topic, fluently and instantly. But it produces, fundamentally, a derivative of what already exists. That's not an insult; it's the architecture. A model cannot tell you what no one has discovered yet, because there is nothing in the training data to compress. Creativity is the opposite. Creativity is forward-looking. A genuinely new idea, a hypothesis no one has tested, a connection no one has drawn, an artistic move no one has made, is precisely the thing that isn't in the corpus. The entire value of a discovery is that it didn't exist before. New knowledge is, by definition, the stuff the model doesn't have. So watch the loop this creates. The forward-looking, creative, net-new knowledge gets locked behind paywalls. The backward-looking, derivative, free AI tools become the primary way most people interface with information. The majority of the population, the 90% on the free tier, gets fed an endless, fluent, confident stream of the recombined past, and is systematically cut off from the forward edge where new thinking actually happens. An entire population could be quietly optimized toward the median of what's already known, trained out of the habits of critical thinking and original creativity, because the tools they use all day are structurally incapable of either. You don't get more creative by consuming derivatives of derivatives. You get more creative by friction, by encountering the genuinely new, the unresolved, the contested, the thing that doesn't fit. If the free, default, mass-market information layer is one that smooths all of that away into a confident average, the long-term cognitive effect on the people who live inside it is not neutral. It's a slow drift toward mediocrity, dressed up as convenience. The Other Side of the Coin: Who Fills the Vacuum? Here's the question almost nobody is asking. If valuable knowledge retreats behind walls, and the public web becomes a low-value zone that the good creators have abandoned, who is still motivated to publish there? Think about it as a simple incentive problem. The legitimate expert now has every reason to either paywall their work or not publish it at all. So the open web's signal-to-noise ratio drops. But the web doesn't go empty. Nature abhors a vacuum, and so does the attention economy. Someone will keep publishing freely and at volume into the public web. Who? The people for whom polluting the public knowledge pool is the business model. Scammers. Content farms. Influence operations. SEO-spam operators. The actors whose goal was never to inform you but to manipulate you: to rank, to phish, to mislead, to sell you something or steal from you. For them, a public web that legitimate experts have vacated isn't a tragedy. It's a land grab. This isn't hypothetical. We're already watching the early version. The "AI slop" wave of 2024 and 2025 flooded Facebook, YouTube, and search with mass-produced synthetic junk engineered for engagement, not truth. A Stanford-Georgetown study traced entire networks of pages pumping out unlabeled AI images to farm audiences and then monetize them. Search crawlers now reportedly burn something like 40% of their indexing budget wading through low-value synthetic noise. The pollution is already here. The only question is what happens when the good content stops showing up to dilute it. And the pollution doesn't stay in the public web. The free models that train on it ingest it. There's already a name for the failure mode, model collapse , or model autophagy: AI trained increasingly on AI output, degrading toward incoherence, hallucinating more on niche topics as the human signal thins out. I wrote about this risk in The Golden Age of AI Agents . The paywall dynamic pours fuel on it: the cleanest, most reliable human signal is exactly the knowledge that's being walled off, leaving the free models to feed on an increasingly contaminated public diet. So the worst-case version of this isn't just "knowledge gets expensive." It's a shared pool of public knowledge drained of its best contributors and flooded by its worst, read mainly by the cheap AI tools most of the planet relies on. A misinformation amplifier with a friendly chat interface. To Be Clear: This Is a Risk, Not a Prophecy I want to be honest about the epistemic status of all this, because the AI conversation is already drowning in people stating speculation as certainty, which was the entire argument of my piece on how AI marketing is burying the actual technology . I'm not going to commit the same sin in reverse. This might not happen. There are real countervailing forces. Open-access movements are pushing hard in the opposite direction, and some argue convincingly that the right response to AI licensing is more open access, not less. Plenty of people will keep publishing freely out of genuine conviction. The open-source ethic is not dead, and there will always be those whose moral compass points toward sharing regardless of incentive. New models for compensating creators directly could emerge; revenue-sharing standards are being drafted, even if no major lab has signed them yet. Regulation could force attribution and payment. The doomsday loop is not guaranteed. But I don't think you need a conspiracy or a catastrophe for the bad version to take hold. You just need ordinary economic incentives, applied to ordinary self-interested people, at scale. That's the unsettling part. Nothing here requires anyone to be evil. It just requires everyone to be rational. What Actually Determines Which Future We Get If the mechanism is incentives, then the fix has to be incentives. A few things genuinely move the needle, and they're worth naming so this doesn't end on a shrug: Attribution that pays. The single most important variable is whether the creator gets something back when a model uses their work. If AI systems reliably attribute, link, and, crucially, compensate, the bargain that built the open web can be rewritten rather than broken. Revenue-sharing standards exist on paper. Whether the major labs adopt them is maybe the most consequential open question in this whole space. Provenance and the value of the verifiably human. As the public web fills with synthetic noise, verifiably human, verifiably original work becomes more valuable, not less, the way handmade goods gained cachet after mass production. Systems that can prove "a real person made this, and made it first" could become the foundation of a healthier model than blanket paywalls. Reading past the median. For you as an individual, the fix is straightforward. The free, default tool gives you a tidy average of the past, nothing more. So make a habit of going where it can't: primary sources, unsettled debates, people pushing into new territory. That's where the thinking it can't do actually lives. The One Thing I Want You to Take From This The open web was never the natural state of the internet. It was a bargain, publish freely and get rewarded indirectly, and AI is quietly rewriting the terms of that bargain in real time. For thirty years we got to live as though the world's knowledge was a public park: free to enter, free to wander, owned by no one and open to everyone. That park was always being paid for by an invisible exchange, creators gave their work away because attention flowed back to them. AI intercepts the attention. And once the creators figure that out, the rational move is to put up a fence and a turnstile. Maybe we rewrite the bargain in time, with attribution and payment and provenance, and the park stays mostly open. Maybe we don't, and we wake up in a world where the good knowledge costs money no single person can fully afford, the free knowledge is increasingly written by people trying to scam you, and the AI tools most of humanity uses every day are confidently feeding them a slightly stale, slightly polluted average of a world that has already moved on. I don't know which one we get. But I know the deciding factor isn't the technology. It's whether we fix the incentive before the default finishes flipping. And defaults, once flipped, are very hard to flip back. \n \