Ten years ago, the data moat thesis was simple: more data means a better model. Companies scraped everything they could reach, hoarded user activity logs, and built data lakes the size of small oceans. The assumption was that scale was the only axis that mattered.
That thesis has inverted. General data is now a commodity. The marginal value of another terabyte of web text is approaching zero. What has become scarce is the specific: your customer support transcripts, your dental practice's scheduling patterns, your institution's forty-year history of maintenance logs, the weird way your team names things internally.
This is the specificity premium. As AI gets better at general skills, writing code, generating text, analyzing data, the premium shifts to the inputs that cannot be averaged or generalized. The things that make your situation different from everyone else's.
Consider the captcha arms race. Text captchas were solved by OCR models. Image captchas were solved by object recognition. Behavior-based captchas were solved by synthetic human behavior. Each round, AI commoditized the previous discriminator, pushing the next one toward something more specific to real human context. The pattern is a miniature of what is happening across the economy: the general gets cheap, the specific gets valuable.
Two companies buy the same model API on the same day. One plugs it in as a generic Q&A bot. The other wires it to their support ticket database, their internal wiki with the handwritten notes, their customer history going back a decade. Same model, completely different outcomes. The model is a commodity. The specificity is the moat.
This explains why the most interesting AI tools are getting narrower, not broader. Bitsy's tiny game engine forces creativity through constraint. A tool built for one specific workflow beats a general platform every time, because specificity embeds in usage patterns that are hard to replicate. Julia Evans writes blog posts for one specific person and they resonate with thousands. The mechanism is the same: tuning for a narrow context produces output that feels real.
The specificity premium creates a strange tension with how venture capital works. The most defensible businesses are often inherently uninvestable, too niche, too tied to one institution's quirks, too weird. A company that owns the data pipeline for dental claims processing has a real moat, but it will never be a billion-dollar SaaS company. It is too specific. Most specificity-rich businesses will never fit the VC scale model, which means they will be built by incumbents or by small teams who do not need permission to be boring.
There is a dark version of this story. When governments and large institutions realize specificity is valuable, they tend to reach for theater first. Rio de Janeiro announced a "homegrown LLM" that was mostly a fine-tuned open model with some municipal data. The real moat would have been their proprietary city data, traffic patterns, permit histories, utility usage, but building those pipelines is hard and unglamorous. The easy path is to buy the model and call it sovereign. The specificity premium punishes shortcuts.
The content strategy implication is direct. In a world where everyone can generate broadly competent text, the only remaining moat is being weird, niche, and specific. The best general strategy is to stop optimizing for reach and start optimizing for the one person who will actually use what you make. Writing for an audience produces undifferentiated noise. Writing for one specific person produces something someone will actually read.
This is the inversion. For a decade, the mantra was that data moats were about scale. Data is no longer scarce, but context is. Your institutional history, your production edge cases, your team's accumulated knowledge of what breaks and why, these are the things that cannot be bought from an API provider. They cannot be averaged across customers. They cannot be generalized.
The companies that will win the next phase are not the ones with the most data. They are the ones who best understand what makes their situation different from everyone else's, and who have the discipline to invest in the boring work of making their specific context usable.