The Audience That Never Reads

Most discussions about artificial intelligence and journalism ask whether machines can write well enough to replace reporters. The question misses what is actually happening. Across the industry, publishers are no longer asking whether AI can produce a story. They are asking who the story is for — and increasingly, the answer is not a person.

That shift is the real story, and it carries a consequence that deserves more attention than it has received. When an outlet designs its work for an agent rather than a reader, it removes a human act of judgment from the process: the judgment of what matters, what can be cut, and what deserves to be said first. That judgment does not disappear. It is simply transferred to software that has no stake in the outcome.

A Briefing Written for One

The change begins quietly, in the morning routines of individual people. A public relations consultant in San Diego named Michelle Seo starts her day with a briefing assembled by her Muse agent, a tool she calls Chattiana. [1] The briefing lists her agenda and to-dos, then pulls conversational topics from her newsletters with links attached. Seo describes herself as an introvert working to start more conversations, and the summary gives her material for filling an awkward silence on a Zoom call. The detail is small, but it marks the direction of travel: news is being consumed through an intermediary that decides what is worth surfacing.

Seo is not unusual. What is unusual is that newsrooms have begun to notice her agent as a constituent. The trend has grown common enough that some outlets now produce material whose intended audience is not a human being at all, but the software that reads on a person’s behalf. That can take the form of feeds tuned for machine ingestion, or of articles structured so an agent can reassemble them into something personalized.

The Case for Structured Data

Axios has moved furthest in public. Its chief revenue officer, Jacquelyn Cameron frames the shift as a matter of following the audience wherever it goes, and the company’s diagnosis is blunt: the open web is deteriorating, search is shrinking, and hyperpersonalization is surging as large language models become more central to how people encounter the news. [1] The response is a product called Axios Direct, announced at the Digiday Publishing Summit. [1] The feeds are built for agents, not eyes. Cameron describes them as a more efficient route into Axios content, with narrative trimmed back and structured data given priority — closer in spirit to a Bloomberg terminal than to a stack of articles. [1]. The first feed, live since the fall, hands merger-and-acquisition material to the agents of Axios Pro Deals subscribers. [1]. A second, due next year, will serve as a training source for corporate in-house models. [1]. A third, planned for consumers, would route Axios reporting straight to an individual’s agent. [1]. None of the three is designed to be looked at. They arrive as standardized lines of text, unlovely by intention.

The sequencing is telling. Businesses come first because corporate budgets tolerate experiments that individual subscribers would not, particularly when an unproven concept might yield even a slight edge. Only after the business case is tested does the consumer version follow. The human reader, in this order of operations, is the last customer served.

What Gets Cut

Axios’s approach is subtractive: strip the prose, keep the facts, let the machine reassemble. Every, a six-year-old tech media and software company, goes the other way, though the effect on human judgment is similar. Its founder, Dan Shipper, explains that writers now embed prompts in articles wherever possible so readers can test ideas themselves. [1]. A piece about a new model might carry a series of prompts a reader can paste into an AI to watch how the model answers questions and weighs input. Guides carry buttons that summon Claude or ChatGPT, instructing the assistant to open with a tight summary and then offer to go deeper.

The underlying logic is that writers no longer have to compress. They can pack a story with detail without worrying about holding attention, because an agent will extract the relevant parts. Two years ago Every published its first “extendable” article, a long feature backed by a database of interview recordings, news articles, and videos that readers could query through a chatbot. Recently it published an 8,000-word piece on Codex that subscribers can consume in whatever form their agent constructs — a bulleted summary for one reader, a short film of interview clips for another.

Shipper names the trade-off with unusual candor. A writer must always choose what goes into a story, and must always wonder whether those choices are right — or right for a particular reader. [1]. With AI in the loop, the journalist stops choosing. What was once an editorial decision becomes a configuration.

The Audience That Never Reads (Image 1) — AI-generated image

The Slow Machinery

The institutional response has been slower and more varied than the technological one, and the gap between them is where the real friction sits. Many publishers, including WIRED’s parent company Condé Nast, have signed deals letting AI platforms surface their work. Others chase generative engine optimization, a discipline aimed at making sure articles appear when someone prompts a model. One WordPress plugin now generates LLM.txt files to help machines discover and interpret a site. The New York Times took the opposite road, suing OpenAI and Microsoft in 2023 for copyright infringement over the training of models on its reporting. That case is still running.

These are not trivial disputes, but they share a feature: each is a negotiation over access, not over authorship. None of them restores the moment when a person decides what a story is. The deals and the lawsuits settle who may read the work. They do not settle who the work is written for, and that is the question Axios and Every have already answered in practice.

The Economics of Being Unnecessary

Whether any of this pays is unresolved. Rasmus Kleis Nielsen, a professor of communication at the University of Copenhagen, describes the market for publishing to agents as emerging, immature, and extremely uneven. [1]. Price points remain unsettled, and so does the question of how agents will interact with content as the technology evolves.

Axios’s operators are more sanguine. Cameron and chief operating officer Allison Murphy argue that the outlet’s short, highly structured style already makes it a favorite source for models answering prompts. Murphy’s formulation is transactional: time is money for crawlers, and they do not want to travel three quarters down a page to find the main point. [1]. That preference, she suggests, already works in Axios’s favor.

Nielsen offers a gentler reading of the same dynamic. Even if readers never visit the homepage, brand visibility carries value, much as publishers accept that content made for TikTok and Instagram rarely sends viewers back to the mother ship. [1]. The comparison is apt and slightly deflating: the outlet becomes a supplier of raw material to a platform it does not control.

The pricing tells its own story about what remains scarce. An all-access Axios Pro Deals subscription runs $2,499 a year; an annual Every subscription costs $288. People will pay for information that other outlets and other models do not have — in-the-weeds guidance at Every, market-moving speed at Axios Pro Deals. Shipper treats the $288 as covering a human reader and that reader’s agent together. [1]. He may not want to read everything each month, but he wants his agent to know it, so the agent can tell him when something matters. The subscription now buys comprehension on his behalf.

The Contradiction That Remains

Murphy’s framing is the most revealing thing anyone says about the transition. She accepts that people already use AI to absorb information and that more will do so soon. [1]. She tries to make the prospect less existentially terrifying by recalling that media has been disrupted before — by social platforms, short-form video, podcasts. Her conclusion is that this is not really AI as the audience, but AI as a tool for reaching the audience. [1].

The distinction is worth holding onto, because it is the last line of defense for human judgment. If the agent is a tool, a person still stands behind it, still decides what to ask and what to trust. But the products themselves point the other way. A feed with the narrative removed, a database that lets an agent build a bespoke article, a button that hands the reading to a chatbot — each of these removes a decision that a person used to make. The publisher stops deciding what the story is. The reader stops deciding how to enter it. What remains is a transaction between machines, with a human somewhere at either end, nominally in charge.

That is the contradiction the industry has not resolved and may not be able to. The technology keeps improving, the feeds keep getting cleaner, the summaries keep getting tighter, and the role that once belonged to an editor — the person who says this matters, this can go, this is where the story starts — keeps migrating into software that has never read anything for pleasure and never will. The audience, in the end, may be the only participant with no opinion about what it is being told.

The Audience That Never Reads (Image 2) — AI-generated image

Sources

Wired — Quote source (original article)