Most people who use online forums do not think about money when they type a reply. They are answering a question, settling a debate, or sharing a fix that took them three hours to figure out. The post goes up. They move on. The conversation sits there, quietly, for years.
Something changed about what that sitting means.
The companies building the most powerful AI systems in the world need one thing above almost everything else: real human language, written by real people, about real problems.
And a very specific corner of the internet turns out to hold more of it than almost anywhere else.
Why AI companies started paying serious attention to forum posts
For years, AI labs scraped the open web for free. Forum threads, product reviews, how-to answers, all of it fed into training runs worth hundreds of millions of dollars in compute alone.
Then the platform that owns the biggest trove of those conversations decided the arrangement was worth reconsidering. The math shifted fast.
Reddit is especially valuable because its forums capture real conversations in a way polished articles never do.
That rawness, the misspellings, the corrections, the arguments, turns out to be exactly what large language models need to learn how people actually think and speak.
The quiet asset that millions of people created without realizing it
Every niche subreddit, every long comment chain, every thread where someone explains how to fix a leaking pipe or argues about a movie ending, is a data point that AI systems prize.
Reddit’s forums offer hyper-specific, real-time insights into everything from niche hobbies to emerging trends, ideal for refining large language models in areas like sentiment analysis and contextual understanding.
It is not just volume. It is texture. Structured opinion at scale, the kind that does not exist in encyclopedia entries or news articles, is a genuinely scarce resource in AI development.
That scarcity did not go unnoticed, and it quietly set the stage for a negotiation that most users had no idea was happening on their behalf.
How the deals started, and why the original terms no longer hold
The first wave of AI content deals was simple: a flat annual fee, negotiated once, paid to the platform. No dynamic pricing. No usage tracking. A fixed number in exchange for access.
The initial wave of AI content licensing deals in 2023 and 2024 focused on flat-rate annual fees, mostly for training. That structure is already evolving.
Those early terms were set when the buyers had far better information about how much the data was worth than the sellers did.
That information gap is now closing, and the platforms that held the data are pushing back hard.
For anyone tracking where this is heading, the contrast with how AI companies are managing their own model costs is striking, just as open AI models are being priced against closed ones, the same pressure is landing on the data that feeds them.
The $550 million figure analysts put on everyday posts
Reddit signed a content licensing deal worth about $60 million per year with Google to access Reddit’s Data API for use in Google’s AI models, an agreement reached in early 2024, just before Reddit’s IPO filing.
In its IPO prospectus, Reddit disclosed data licensing arrangements with an aggregate contract value of $203 million and terms of two to three years, according to TechCrunch.
But those original figures are already looking conservative. The renegotiations are live, and the numbers being floated are a different order of magnitude.
On June 5, 2026, Wells Fargo turned more constructive on the renewals, forecasting that combined annual revenue from the two deals could surge to $550 million on renegotiation, more than quadrupling a current run rate of roughly $130 million.
Reddit is discussing with Google and OpenAI a future deal structure that could enable dynamic pricing, where the platform earns more as its data becomes more valuable to AI-generated answers.
And the leverage is real. According to Profound, which analyzed more than 4 billion AI citations across major platforms including ChatGPT, Google AI Overviews, and Perplexity, Reddit is the number-one most-cited source in AI-generated answers aggregated across those platforms. Profound’s data put Reddit at 3.11% of all citations, more than twice the share of Wikipedia, which sits at 1.35%.
The posts ordinary users wrote for free are, by that measure, the internet’s most consulted resource for the systems that hundreds of millions of people now treat as a first stop for answers.
The full picture of how AI content tools are being woven into everyday platforms makes this trend land differently: the data being licensed is not abstract. It is the specific things people typed, from specific lived experiences, on a platform they used for free.
What the new pricing model means for everyone who ever posted anything online
Publishers now get paid when AI systems fetch or ground their content in real time, and Reddit anticipated this second wave of usage-based pricing.
That shift matters beyond Reddit. These numbers map a market that barely existed three years ago.
The global AI training dataset market was valued at $3.59 billion in 2025 and is projected to grow to $4.44 billion in 2026, according to Fortune Business Insights.
Separate estimates from market research firms project the broader dataset licensing segment to grow substantially over the next decade, though figures vary widely depending on methodology and scope.
The underlying question, who owns the value in user-generated content, is now one of technology’s most consequential open disputes.
What started as a forum full of arguments about TV shows and home repair is turning into a legal and economic framework for how the AI era handles human knowledge.
The posts are still sitting there. They are just worth a lot more than anyone thought when they hit send.
