AI agent audience is the phrase publishers now use for a reader that never opens a web page. On 27 October 2026 WIRED’s Kate Taylor reported that some news outlets “are starting to produce content that treats AI agents—not human readers—as the intended audience.” Her examples run from Axios, which is building feeds for software agents, to Every, which now writes articles that a subscriber’s agent can rebuild into a summary, a podcast or a short video.
The story opens with Michelle Seo, a public relations consultant in San Diego. Every morning her Muse agent, which she calls Chattiana, sends her a briefing. It lists her agenda and to-do list, then pulls talking points from the newsletters she subscribes to, with links back to the sources. The agent does the skimming, and the publishers who wrote those newsletters are learning to treat that kind of agent as a customer.
This guide sets out what the AI agent audience shift actually involves. It covers the three Axios Direct feeds, Every’s agent-ready articles, Time’s ads written for bots, the data on how many people get news through chatbots, and the hard question of who pays. It ends with practical steps for any business that publishes content online. For the newsroom side of the same trend, see our earlier piece, AI Agents in Journalism: 7 Shocking Challenges for Newsrooms.
Table of contents
- What the AI Agent Audience Shift Means
- How Axios Is Building for the AI Agent Audience
- How Every Writes for the AI Agent Audience
- Time’s Ads Written for an AI Agent Audience
- How Many People Already Read the News Through AI?
- The Tools Behind the AI Agent Audience
- Comparing the Main AI Agent Audience Strategies
- Can Publishers Make Money From the AI Agent Audience?
- What the AI Agent Audience Means for Businesses That Publish Online
- Our Verdict on the AI Agent Audience
- AI Agent Audience FAQs
- References
What the AI Agent Audience Shift Means
For twenty years, digital publishing has had one basic loop. A person searches or scrolls, clicks a link, lands on a page and sees an advert or a paywall. Every part of that loop assumes a human at the end of it. The AI agent audience breaks the loop, because the reader that arrives is software working on someone’s behalf.
From human readers to an AI agent audience
An agent does not browse. It is given a job, such as “brief me on today’s news” or “track deals in my sector”, and it fetches whatever it needs to finish that job. It reads quickly, skips design and adverts, and hands its owner a short answer. The person still benefits from the journalism, but they may never see the publisher’s page.
That is why Axios chief revenue officer Jacquelyn Cameron describes the change in such broad terms. “We believe that the media ecosystem is changing,” she told WIRED. “We see this with open web deteriorating. We see that search is shrinking. We see that hyperpersonalization is surging, that LLMs are becoming more important in the way that people are consuming” their news.
Her point about the open web echoes Google’s own words. In a 2025 court filing in its advertising case, Google said that “the open web is already in rapid decline”, as Fortune reported, before clarifying that it meant open-web display advertising. Either way, the old route from search result to publisher page carries less traffic than it did.
Why WIRED calls it a foundational change
Publishers have already tried several responses to AI. WIRED lists the main ones. Many outlets, including its own parent company Condé Nast, have signed licensing deals that let AI platforms show their content. Others work on generative engine optimisation, or GEO, so that their articles appear in AI answers. And some have gone to court: The New York Times sued OpenAI and Microsoft in 2023, and that case is still going.
Writing for an AI agent audience is different from all three. A licensing deal sells content to an AI company. GEO tries to get noticed by AI search. But building products that agents consume directly changes who the story is for. In WIRED’s words, it “represents a more foundational shift in how the media operates—and who its stories are for.”
Cameron’s own summary is simpler: “We have always said that we want to be where the audience is.”
How Axios Is Building for the AI Agent Audience
Axios announced its plan at the Digiday Publishing Summit in Key Biscayne, Florida, on 16 September 2026. Digiday’s Tim Peterson reported the details two days later. The product is called Axios Direct, and Cameron described it as taking “the concepts of a Bloomberg terminal and an RSS feed” and updating them for agents.
Three feeds for three kinds of customer
Axios Direct is not one product. It is a plan for three feeds, each aimed at a different buyer in the AI agent audience. The table below combines what Cameron told Digiday in September with what she and chief operating officer Allison Murphy told WIRED in October.
| Feed | Who it is for | What it carries | Timing |
|---|---|---|---|
| First feed | Investment firms, mostly existing Axios Pro Deals subscribers | Merger, acquisition and investment reporting for their agents | Launched this autumn |
| Second feed | Companies running their own in-house models | Axios content as a source for internal systems | Next year (WIRED); no set date (Digiday) |
| Third feed | Individual subscribers | Axios reporting sent straight to a personal agent | To be explored in 2027 |
The two reports differ slightly on the second feed. WIRED says it will launch next year, while Digiday’s September article said Axios had not set a timeframe. Both agree that the consumer feed comes last.
Structured lines of text, not articles
The feeds will not look like Axios stories. Cameron told WIRED they cut back on narrative to put structured data first, and that all three “will look similar to RSS feeds”. They are “not aesthetically pleasing, as they aren’t intended for the human eye”, taking the form of standardised lines of text that an agent can parse quickly.
This is an AI agent audience product in its purest form. Axios is already known for short, bulleted stories, and its executives say that style has made it a favourite source for AI crawlers. “Time is money for crawlers,” Murphy said. “They don’t want to have to go three quarters down the page to find the main point.”
How the deals are priced
Digiday reported that companies will pay an annual fee for the first feed. The fee will vary with the size of the company and the value of the assets it manages. Axios is signing these first deals for two years, so that it can adjust the product with each client before renewal.
Cameron said the feed is “an important upsell” for firms that already pay for Axios Pro Deals. According to WIRED, an all-access Pro Deals subscription costs $2,499 a year. She also said that about 52% to 54% of Axios revenue this year will come directly from clients, and that the company prefers to sell new products directly rather than through AI content marketplaces. She did not rule out marketplaces altogether, and she mentioned talks with Snowflake about its marketplace.
Why Axios can afford to experiment
Axios is not making this bet from weakness. Cameron told the Digiday audience that the company hit its 2026 revenue goal in early September, compared with the end of October in 2025. She would not give figures, saying only that growth was somewhere between 10% and 50%. Starting with business clients also makes sense, WIRED notes, because corporate budgets “tend to have more tolerance than individual subscribers for experiments”, especially when the content could move markets.
How Every Writes for the AI Agent Audience
Every takes a very different route to the AI agent audience. Axios is trimming its already brief content into data. Every, a six-year-old media and software company founded by Dan Shipper, is making its writing longer and richer, on the assumption that a reader’s agent will pick out what matters. WIRED says Every “expects its subscribers to treat AI as almost an extension of themselves.”
Prompts written for the AI agent audience
The most common difference, Shipper told WIRED, is that Every’s writers include prompts in their pieces wherever they can. An article about a new model, for example, comes with prompts the reader can paste into a chatbot to see how the model answers and evaluates input for themselves. The article becomes something to run, not just something to read.
Some guides also carry buttons marked “Read With Claude” or “Read With ChatGPT”. Pressing one opens a prompt that tells the assistant to “Start with a tight summary: one paragraph, bulleted. Then offer to go deeper.” Because an agent will do the filtering, writers can pack stories with detail without worrying about losing a human reader’s attention.
Extendable articles and the Codex story
Two years ago, Every published its first “extendable” article: a long feature with a database of the reporting behind it, from interview recordings to news articles and YouTube videos. Readers could ask a chatbot questions, and it would pull extra details from that database.
The newest example is Shipper’s own feature “How Codex Happened”, published on 15 October 2026 and described by WIRED as an 8,000-word article. It draws on interviews with six people who built OpenAI’s coding product, including Greg Brockman and Alexander Embiricos. The page offers the full interview transcripts, inviting readers to “explore them with your agent”. Using that material, a subscriber’s agent can produce a new version of the story, whether a bulleted summary, a podcast or a short film made from interview clips.
When the reader makes the editorial choices
This is the part of the AI agent audience trend that most challenges journalism. “As a writer, you have to make choices about what goes into the story you’re telling,” Shipper told WIRED. “And there’s always a question about, are those the right choices? Or, are those the right choices for a particular reader?” With an agent in the middle, WIRED observes, journalists no longer make that choice alone.
Shipper sounds excited rather than worried. “People are gonna do crazy, crazy stuff with this,” he said.
Time's Ads Written for an AI Agent Audience
If agents are readers, they can also be shown adverts. Time tested that idea before Axios Direct was announced. Digiday’s Sara Guaglione reported on 30 July 2026 that Time had begun serving ads to AI agents, with Ally Bank and the Project Management Institute among the first buyers.
Markdown pages and FAQ-style ads
The ads sit on special versions of Time’s pages built for the AI agent audience. In June, Time began converting its web pages from HTML into markdown: plain-text copies without design or images, which machines read more easily. It works with the ad technology firm Mobian, which converts the pages and writes the ads. Each ad is formatted as a set of questions and answers about the brand, and carries a “sponsored content” label at the top, even though no rule yet requires one.
Time sells one agent ad per markdown page and charges a premium for it. Its chief operating officer, Mark Howard, said Time sees more bot traffic than human traffic on most days, with big spikes when it publishes a Time100 list. Mobian’s chief executive, Jonah Goodhart, put the pitch bluntly: “When you influence ChatGPT, you’re influencing potentially all of ChatGPT.”
The cloaking risk for AI agent audience ads
Not everyone is sure this will last. Rob Derow of BCG X warned Digiday that AI systems could one day treat markdown ads as cloaking, the old search trick of showing crawlers different content from what people see. If that happens, pages carrying such ads could lose visibility. For now, nobody knows whether AI systems weigh sponsored answers differently from editorial ones. Time’s own executives say they are “paving the first path forward”.
How Many People Already Read the News Through AI?
The AI agent audience is still small compared with search or social media, but it is growing. The best measure is the Reuters Institute’s Digital News Report 2026, published in June and based on a YouGov survey across dozens of markets.
Weekly use is up, but still a minority
According to the report, as summarised by The Decoder, 10% of people now use an AI chatbot for news each week, up from 7% a year earlier. Just 1% call a chatbot their main source of news. Use is highest among the young: 17% of 18 to 24-year-olds, compared with 5% of the oldest group. Most growth is coming from Asia, Africa, Latin America and southern and eastern Europe.
The report also asked chatbot news users what they do with these tools. The chart below shows the five most common answers.
Bars are scaled to the largest answer: 35 is 83% of 42, 34 is 81%, 33 is 79% and 30 is 71%. Summaries and simplified news are exactly the jobs Seo’s morning briefing does, which is why publishers see the AI agent audience as a coming market rather than a novelty.
Clicks back to publishers are rare
The more worrying number for publishers is how rarely people click through. Across 27 markets, only 4% of all respondents said they always or often click from an AI chatbot to the original source. For search engines the figure was 19%, and for social media 17%.
Bars are scaled to search: 17 is 89% of 19, and 4 is 21%. Part of the gap reflects the smaller number of people using chatbots for news, but the direction is clear. An answer replaces the click. That is the economic problem the AI agent audience creates, and it is why Axios and Every are trying to charge agents directly instead of waiting for visits.
Trust is still low
Trust is the other limit. The report found that only 20% of the general public trust news from AI chatbots, against 37% who trust most news. Among people who actually use chatbots for news, though, trust rises to 44%. For the AI agent audience model, that gap matters: people who rely on these tools trust them, but most people do not yet. The Reuters Institute’s advice to publishers is not to compete with AI platforms on their own ground, but to focus on what chatbots cannot supply: original reporting and credibility.
The Tools Behind the AI Agent Audience
Writing for agents is partly an editorial choice and partly a technical one. WIRED mentions one small but telling example: in 2025 AIOSEO, a search plug-in for WordPress sites, added a feature that creates llms.txt files for websites.
llms.txt and markdown copies
An llms.txt file is a plain markdown file at the root of a website. It gives agents a short description of the site and links to the most useful pages, ideally in clean markdown versions. The idea was proposed in 2024 by Jeremy Howard of Answer.AI. Version two of the proposal says that “thousands of sites” now publish one, that OpenAI, Anthropic and Google’s Gemini publish them for their own developer documentation, and that Chrome’s Lighthouse checks for one in its agentic browsing audits.
The proposal also suggests a markdown copy of each important page at the same address with “.md” added. That is the same approach Time took when it converted its pages to markdown. Turning a web page back into clean text is a classic natural language processing problem, and every wasted token costs an agent time and money, so publishers that do the conversion themselves get read more reliably.
Structured feeds and licensing
Above the file level sit the commercial pipes. Axios Direct is a direct feed. Microsoft and Snowflake run marketplaces where publishers can offer content to companies building their own systems. Licensing deals such as Condé Nast’s agreement with OpenAI let AI platforms show content inside their answers. Each of these is a different way of reaching the AI agent audience, and they can be combined.
If you want the wider background on how AI systems collect web content in the first place, our guide AI Is Eating Website Traffic, Websites Are Blocking AI explains the crawler side, and Accel-Backed Keenable Is Indexing the Web for AI Agents covers a start-up building a search index just for machines.
Comparing the Main AI Agent Audience Strategies
The publishers in this story are making different bets. The table below sets them side by side, using only what the sources report.
| Publisher or tool | Approach | How it makes money | Main risk |
|---|---|---|---|
| Axios Direct | Short, structured feeds for agents | Annual fees, two-year business deals | Unproven demand outside finance |
| Every | Rich articles with prompts and reporting databases | Subscriptions ($288 a year, per WIRED) | Readers’ agents rewrite the story |
| Time | Markdown pages with FAQ-style agent ads | Premium-priced sponsorship | AI systems may treat it as cloaking |
| Licensing deals (e.g. Condé Nast) | Let AI platforms show content | Fees from AI companies | Dependence on a few platforms |
| Lawsuits (e.g. The New York Times) | Challenge AI training in court | Damages or settlements, if they win | Years of uncertainty |
| llms.txt and GEO | Make content easy for agents to find and read | Indirect, through visibility | Visibility without visits |
Trimming versus expanding
The sharpest contrast is between Axios and Every. Axios assumes the AI agent audience wants less: the main point first, in a fixed format, with no narrative to wade through. Every assumes it wants more: every interview and source, so that the agent can build exactly what its owner wants.
Both can be right, because they serve different jobs. A fund’s agent tracking deals needs speed and precision. A curious reader’s agent preparing a weekend podcast wants depth. The lesson for other publishers is to decide which job their content does before choosing a format.
Can Publishers Make Money From the AI Agent Audience?
This is the question WIRED leaves open, and the honest answer is “not yet at scale”. Rasmus Kleis Nielsen, a professor of communication at the University of Copenhagen, told WIRED that publishing for agents is an “emerging” market that remains immature and “extremely uneven”. Price points and the way agents will use content in future are both still unsettled.
Visibility as a return on the AI agent audience
Nielsen sees some value even without visits. Brand visibility inside AI answers can help a publisher, he says, much as making videos for TikTok and Instagram can, even though those rarely send viewers back to a website. Axios already benefits here, because its structured style makes it easy for AI systems to quote.
Paying for what AI cannot give away
The deeper problem is that a chatbot can hand people common facts for free. People will only pay for information that other outlets, and AI models, do not have. WIRED gives two examples. At Every, that means detailed, practical guidance on using AI. At Axios Pro Deals, it means fast access to information that could move a market.
| Subscription | Annual price (WIRED) | Per month | What the buyer pays for |
|---|---|---|---|
| Axios Pro Deals, all access | $2,499 | About $208 | Speed on market-moving deal news |
| Every | $288 | $24 | In-depth guidance on AI and how to use it |
The monthly figures are simple division by 12: $2,499 / 12 = $208.25, and $288 / 12 = $24. The Axios product costs about 8.7 times as much ($2,499 / $288 = 8.68), which shows how much more a business will pay for speed than an individual will pay for insight.
One subscription for a person and their agent
Shipper argues that a subscription should cover both the reader and their agent. “I may not want to read all of your stuff every month,” he told WIRED. But if the subscription includes his agent, “I may want my agent to know it so that it can tell me, ‘Hey, this thing is important.'” In that model the AI agent audience does not replace paying readers. It becomes a reason for them to keep paying.
What the AI Agent Audience Means for Businesses That Publish Online
This trend is not only about newspapers. Any organisation with a website now has two kinds of visitor: people and software acting for people. Time’s sales team is already pitching to brands that have converted their own sites to markdown, and Mobian estimates that about 15% of brands now do so.
Make your content easy for the AI agent audience to read
The first step is technical and cheap. Publish an llms.txt file that describes your site and links to your key pages. Offer clean markdown or plain-text versions of important pages. Put the main point at the top, as Axios does, and use clear headings and structured data. If you need help with this kind of work, our GEO services team works on visibility in AI answers alongside traditional search.
Decide what you will give away and what you will charge for
The Axios and Every examples show that agents happily take common information for free. Think about which parts of your content are unique, such as your data, your expertise or your speed, and whether they belong behind a login, in a feed or in a paid product. Remember that AI agents will summarise whatever they can reach.
Measure bot traffic, not just people
Time knows it gets more bot traffic than human traffic because it measures it. Check your server logs or analytics for AI crawler and agent requests, and track how often your brand appears in AI answers. Without that data, you cannot tell whether the AI agent audience is reading you, ignoring you or misquoting you.
Keep humans in the loop
Finally, label sponsored material clearly and keep a human review step, as Time does with its agent ads. Rules on how AI systems treat promotional content do not exist yet, so transparency is the safest position. The wider economics of the open web are being rewritten too; our article Can Cloudflare CEO Matthew Prince Save the Web From AI? looks at one company’s attempt to give publishers more control.
Our Verdict on the AI Agent Audience
The AI agent audience is real, but it is early. Only one person in ten uses a chatbot for news each week, and the products built for agents are mostly aimed at businesses for now. Yet the direction is hard to argue with. Agents such as Seo’s Muse assistant already read newsletters on their owners’ behalf, and our hands-on review Instinct Was the Buzziest AI Agent Around — Can It Survive Muse? shows how fast those assistants are spreading.
The publishers in WIRED’s story are not giving up on people. Murphy put it best: “It isn’t really AI as the audience. It’s AI as a tool that helps us reach our audience.” She also tries to make the change “a little less existentially terrifying” by recalling that media has adapted before, to social media, short video and podcasts. Serving the AI agent audience well may simply be the next version of being where the readers are.
AI Agent Audience FAQs
What does “AI agent audience” mean?
It describes software agents that read content on behalf of people, such as an assistant that gathers news into a morning briefing. Some publishers now design content and products for these agents directly, rather than only for human readers.
What is Axios Direct?
Axios Direct is a set of planned content feeds for AI agents. The first, for investment firms that subscribe to Axios Pro Deals, launched this autumn. A second feed for companies’ in-house models and a third for individual subscribers’ personal agents are due later.
How is Every writing for agents?
Every puts prompts inside articles, adds “Read With Claude” and “Read With ChatGPT” buttons, and publishes long features with databases of interviews and sources. Subscribers’ agents can use that material to create a new article, podcast or video.
Is Time really serving ads to AI agents?
Yes. Since July 2026, Time has placed FAQ-style ads, labelled as sponsored content, on markdown versions of its pages. Ally Bank and the Project Management Institute were among the first buyers.
How many people use AI chatbots for news?
The Reuters Institute’s Digital News Report 2026 found that 10% of people use an AI chatbot for news each week, up from 7% a year before. Only 1% say a chatbot is their main news source.
Should my business create an llms.txt file?
For most businesses it is a quick, low-risk step. An llms.txt file helps the AI agent audience find and understand your key pages. Pair it with clear, well-structured content and a way to measure AI traffic.
References
AI Agents Are Media’s New Favorite Audience (WIRED)
Axios preps new Axios Direct feeds for AI models, agents as revenue tops 2026 goal (Digiday)
Time has started serving ads to AI agents (Digiday)
Digital News Report 2026 (Reuters Institute for the Study of Journalism)
More people get news from AI chatbots, but trust remains low (The Decoder)
The /llms.txt file proposal (llms-txt)
What Is LLMs.txt? Plus, Why You Need It On Your Site (AIOSEO)
Google got caught saying the open web is in ‘rapid decline’ (Fortune)
Condé Nast signs deal with OpenAI (WIRED)
Meta’s Muse is better at surveilling than helping me (WIRED)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
