本文为原文前 6,000 字符的节选翻译，完整内容请查看原文。

AI Agents Are Media’s New Favorite Audience

Every morning, Michelle Seo receives a briefing from Chattiana, her Muse artificial intelligence agent. Chattiana’s briefing lists Seo’s agenda and to-do list, and plucks water-cooler topics from the headlines in her various newsletters along with links to the sources.

每天早上，米歇尔·徐（Michelle Seo）都会收到她的 Muse 人工智能代理 Chattiana 发来的简报。Chattiana 的简报列出了徐的日程安排和待办事项，并从她订阅的各种时事通讯头条中挑选出适合茶余饭后谈论的话题，同时附上来源链接。

“I’m kind of challenging myself as an introvert to start more conversations,” Seo, a San Diego-based public relations consultant, tells WIRED. The newsletter, she says, gives her a sense of the daily news and options for filling an awkward silence on a Zoom call.

“作为一名内向者，我正在挑战自己去开启更多的对话，”这位居住在圣地亚哥的公关顾问告诉《连线》（WIRED）杂志。她说，这份时事通讯让她能够了解每日新闻，并为填补 Zoom 通话中尴尬的沉默提供了选择。

Seo is far from the first person to consume her news with an AI assist. In fact, the trend has become so common that some news outlets are starting to produce content that treats AI agents—not human readers—as the intended audience. This can look like anything from feeds optimized for AI to pieces structured to allow agents to create personalized articles.

徐绝不是第一个借助人工智能获取新闻的人。事实上，这种趋势已经变得非常普遍，以至于一些新闻媒体开始制作将人工智能代理（而非人类读者）作为目标受众的内容。这可以表现为针对人工智能优化的信息流，也可以是旨在让代理能够创建个性化文章的结构化内容。

“We believe that the media ecosystem is changing,” says Axios chief revenue officer Jacquelyn Cameron. “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.

“我们相信媒体生态系统正在发生变化，”Axios 首席营收官杰奎琳·卡梅伦（Jacquelyn Cameron）表示。“我们看到开放网络正在恶化，搜索正在萎缩。我们看到超个性化趋势正在激增，大语言模型（LLM）在人们消费新闻的方式中正变得越来越重要。”

Major publishers have responded to that change in a variety of ways. Many, including WIRED’s parent company Condé Nast, have signed deals to allow AI platforms to surface their content. Outlets are grappling with “generative engine optimization,” or GEO, strategies to ensure articles appear in responses to AI prompts. Last year, AIOSEO, a SEO plug-in for WordPress sites, launched a feature that provides websites with LLM.txt files, which help AI discover and interpret sites’ content.

各大出版商以各种方式应对这一变化。包括《连线》母公司康泰纳仕（Condé Nast）在内的许多机构已经签署协议，允许人工智能平台展示其内容。各媒体正在努力研究“生成式引擎优化”（GEO）策略，以确保文章出现在人工智能的回答中。去年，WordPress 网站的 SEO 插件 AIOSEO 推出了一项功能，为网站提供 LLM.txt 文件，帮助人工智能发现并解读网站内容。

Others have taken a more defensive posture: The New York Times sued OpenAI and Microsoft in 2023 for copyright infringement, alleging that the companies illegally trained models on reporting. That case is still ongoing.

其他机构则采取了更具防御性的姿态：《纽约时报》于 2023 年起诉 OpenAI 和微软侵犯版权，指控这些公司非法利用其报道训练模型。该案件目前仍在审理中。

What Axios and a handful of others are trying, though, represents a more foundational shift in how the media operates—and who its stories are for. “We have always said,” Cameron says, “that we want to be where the audience is.”

然而，Axios 和少数几家公司所尝试的，代表了媒体运作方式及其受众定位方面更根本的转变。“我们一直说，”卡梅伦表示，“我们希望出现在受众所在的地方。”

At the Digiday Publishing Summit in September, Axios announced the upcoming launch of Axios Direct: feeds designed for AI agents. “We do believe that a significant portion of news consumption will move into AI systems, LLMs, and devices,” Cameron says. She describes the feeds as a more efficient way for agents to access Axios content, cutting back on narrative to prioritize structured data—more similar to a Bloomberg terminal than a series of articles.

在 9 月份的 Digiday 出版峰会上，Axios 宣布即将推出 Axios Direct：专为人工智能代理设计的信息流。“我们确实相信，很大一部分新闻消费将转移到人工智能系统、大语言模型和设备中，”卡梅伦说。她将这些信息流描述为代理访问 Axios 内容的一种更有效方式，通过减少叙述内容来优先处理结构化数据——这更类似于彭博终端，而不是一系列文章。

The two initial Direct feeds are aimed at corporate clients. The first, which launched this fall, gives the AI agents of Axios Pro Deals subscribers access to merger and acquisition content. The second feed will act as a source to train in-house large language models, and will launch next year.

最初的两项 Direct 信息流面向企业客户。第一项于今年秋季推出，允许 Axios Pro Deals 订阅者的人工智能代理访问并购内容。第二项信息流将作为训练内部大语言模型的来源，并将于明年推出。

Starting with feeds aimed at businesses makes sense. Corporate budgets tend to have more tolerance than individual subscribers for experiments, especially if these unproven concepts could offer even a slight business advantage. But the third Axios Direct feed, which the company plans to develop next year, will be consumer-facing—a way for individual subscribers to have Axios’ reporting sent directly to their agent. All three feeds will look similar to RSS feeds—not aesthetically pleasing, as they aren’t intended for the human eye, but instead taking the form of standardized lines of text.

从面向企业的业务开始是有道理的。与个人订阅者相比，企业预算对实验的容忍度往往更高，尤其是当这些未经证实的理念可能带来哪怕微小的商业优势时。但 Axios 计划明年开发的第三项 Direct 信息流将面向消费者——让个人订阅者能够将 Axios 的报道直接发送给他们的人工智能代理。这三项信息流看起来都类似于 RSS 订阅——并不美观，因为它们并非为人类阅读而设计，而是以标准化的文本行形式呈现。

While Axios is trimming its (already brief) content to satisfy AI, Every is taking a different approach. The six-year-old tech media and software company, which announced a $2 million funding round from investors including LinkedIn founder Reid Hoffman last year, expects its subscribers to treat AI as almost an extension of themselves.

当 Axios 正在精简其（本已简短的）内容以满足人工智能时，Every 采取了不同的方法。这家成立六年的科技媒体和软件公司去年宣布获得包括 LinkedIn 创始人里德·霍夫曼（Reid Hoffman）在内的投资者 200 万美元的融资，它希望其订阅者将人工智能视为自身的一种延伸。

Every’s founder Dan Shipper says the most common way Every’s content differs from a typical article is that its writers include prompts in pieces whenever possible, allowing readers to test concepts themselves. An article about a new model, for example, includes a series of prompts the reader can copy-paste into AI, demonstrating how the model answers questions and evaluates input.

Every 的创始人丹·希珀（Dan Shipper）表示，Every 的内容与典型文章最常见的区别在于，作者会尽可能在文章中加入提示词（prompts），让读者能够亲自测试这些概念。例如，一篇关于新模型的文章包含了一系列读者可以复制并粘贴到人工智能中的提示词，展示了该模型如何回答问题和评估输入。

Certain types of articles—such as guides—also include buttons that readers can press to “Read With Claude” or “Read With ChatGPT.” Clicking pulls up a prompt, instructing AI to “Start with a tight summary: one paragraph, bulleted. Then offer to go deeper.” Writers pack stories with details without worrying about losing readers’ attention, knowing agents will pick out the most relevant bits.

某些类型的文章（如指南）还包含按钮，读者可以点击“用 Claude 阅读”或“用 ChatGPT 阅读”。点击后会弹出一个提示，指示人工智能：“先给出一个精炼的总结：一段话，带项目符号。然后提供深入阅读的选项。”作者在文章中填充了大量细节，而不必担心失去读者的注意力，因为他们知道人工智能代理会挑选出最相关的内容。

Two years ago, the outlet launched its first “extendable” article: a long-form feature accompanied by a database of reporting, from interview recordings to news articles to YouTube videos. Readers could query a chatbot, which would fetch additional details from the database.

两年前，该媒体推出了第一篇“可扩展”文章：一篇长篇专题报道，并附带一个报道数据库，内容涵盖采访录音、新闻文章和 YouTube 视频。读者可以查询聊天机器人，机器人会从数据库中提取更多细节。

Recently, Every published a 8,000-word article about Codex that subscribers can consume in even more forms. Drawing from the database, subscribers’ agents can tailor a new piece—be it an article, podcast, or video—to suit their interests. One reader might want a bulleted summary, while another could request a short film of interview bits.

最近，Every 发表了一篇关于 Codex 的 8000 字文章，订阅者可以以更多形式消费该内容。利用数据库，订阅者的人工智能代理可以根据他们的兴趣定制一篇新作品——无论是文章、播客还是视频。一位读者可能想要一个带项目符号的总结，而另一位读者则可能要求制作一段采访片段的短片。

“As a writer, you have to make choices about what goes into the story you’re telling,” Shipper says. “And there’s always a question about, are those the right choices? Or, are those the right choices for a particular reader?” With AI, journalists no longer choose.

“作为一名作家，你必须对故事中包含的内容做出选择，”希珀说。“而且总会有一个问题：这些选择正确吗？或者，对于特定的读者来说，这些选择是正确的吗？”有了人工智能，记者们不再需要做这些选择了。

“People are gonna do crazy, crazy stuff with this,” Shipper adds. As outlets eye AI agents as a new audience, another question looms: Can publishers actually make money off of this?

“人们会用它做出非常疯狂的事情，”希珀补充道。随着各媒体将人工智能代理视为新的受众，另一个问题随之而来：出版商真的能从中赚到钱吗？

Rasmus Kleis Nielsen, a professor of communication at the University of Copenhagen, says that while the market of publishing for agents is “emerging,” it remains immature and “extremely uneven.” Major questions remain, he says, about price points and how AI agents will interact with content in the future.

哥本哈根大学传播学教授拉斯穆斯·克莱斯·尼尔森（Rasmus Kleis Nielsen）表示，虽然面向代理的出版市场正在“兴起”，但它仍然不成熟且“极其不均衡”。他说，关于定价以及未来人工智能代理将如何与内容互动，仍存在重大疑问。