Welcome to the world of machine audiences

Over the past 30 years, most recently in a former church in San Francisco, the Internet Archive has systematically preserved over a trillion snapshots of the web. Late last year it began archiving something different: how artificial-intelligence systems respond to questions about the world. The web’s great librarian has recognised something new worth preserving, and it is a signal that the centre of gravity in the information economy may be moving, from what is published to what is being asked.

The conversation about AI and the information economy is mostly about supply. Content is being commoditised. Journalism is dying. Intellectual property is being scraped without consent or compensation. The internet is being flooded with misinformation and slop. This is understandable, and it is not wrong, but it is radically incomplete, for a specific reason: the story has been largely shaped by those who represent the supply side. What’s missing here is that AI’s impact on the information ecosystem is also a demand-side shock, and arguably a demand-side expansion.

Every major information revolution, from printing to mobile, expanded the market for knowledge, often dramatically. AI is likely to do the same, but is also categorically different: the world is entering the age of machine audiences.

Consider what happens when someone asks an AI a question about the world. The system draws on vast amounts of knowledge and other content, synthesises what it deems relevant and gives an answer shaped to serve that person’s intentions. That answer is entirely new and may never be seen again. As the protocols for machine-to-machine communication—like Anthropic’s Model Context Protocol or Google’s Agent2Agent—begin to stabilise, it seems increasingly possible that information will soon pass through numerous AI systems before reaching a human at all. This is a new category of demand, operating at machine scale.

But AI does not only create machine demand. It simultaneously expands human demand too. The obscure corners of information demand—very specific needs that no article or broadcast could ever serve before—become addressable. Conversational AI brings latent demand to the surface too, helping people articulate needs they could not previously express. And AI lowers cognitive barriers: complex information, once comprehensible only to specialists, becomes digestible by anyone who can ask a question. ChatGPT has, by OpenAI’s own count, 900m weekly users. A separate NBER working paper published by OpenAI suggests that roughly a third of interactions are something closer to sense-making than information retrieval.

This is deeply uncomfortable for traditional knowledge producers, because the primary consumer in this new market is the machine, and the business models, ethos and culture those traditional producers have built over centuries were not designed for it.

To serve this new demand, it is crucial to understand the user: their accumulated context and their intent. Context is what a person already knows, the situation they are in, the history of what they have asked before. Intent is what they are asking and trying to achieve right now. Together, these form the demand signal, and this might be the defining asset of the AI-mediated information ecosystem. Not clicks. Not time spent. Those attention-economy currencies lose their worth in this paradigm.

By default, that asset accumulates with whoever holds the user relationship, and the more an AI system knows about a person, the better it can serve their needs, and the harder it becomes to switch to another. The AI systems that glean the most about users’ context become the gateway. Knowledge producers lose sight of the humans on the other side of the machines, because the intermediary absorbs the demand signal that once connected what was produced to what was needed. And the value of this vastly expanded market gets captured by whoever owns the interface. Left unchecked, this is a problem for everyone, even the AI companies themselves. Without the producers who can serve demand with reliable information, the knowledge base degrades, and AI systems have less and less of value to draw on.

Changing this requires three things. Producers need access to the demand signal, so they can build new products that serve real needs. Users need the ability to carry their accumulated context—the history of what they have asked, learned and decided—from one AI system to another, as one might take a relationship with a doctor to a new practice. And society needs common ground. When the answers are shaped to an individual’s context and intent, the shared realities that hold communities together begin to dissolve. A front page or a broadcast may be a crude technology but it produces something intent-driven personalisation does not: the experience of knowing what other people know.

This market has barely begun to form. There are no stable mechanisms to match and price the demand signal against the supply that could serve it. The information supply chain that could run from how knowledge is sourced, through how agents exchange it, to how AI delivers comprehension to humans, has no common rails, and no incentive structure to reward rigour over fluency. The knowledge products and services that could serve this expanded market have barely been imagined, let alone created.

For existing knowledge producers, the fight for fair attribution and compensation is necessary. But a new market is forming alongside it and will be shaped by those who build now. The disciplines that journalism and the scientific method developed over centuries—of truth-seeking, accountability and self-correction—are the very operating principles that market will require. ■