Publishers are no longer treating AI agents as ordinary web visitors. They are starting to build a separate way for machines to read, understand, and update their journalism.
People Inc. is developing custom APIs that give AI systems article text in Markdown, alongside details such as the author, publication date, update date, and other trust signals. The point is simple: an agent should not have to wade through a webpage designed for a human browser before it can find the information it needs.
That sounds technical, but it marks a meaningful change in the publisher-platform relationship. Publishers are beginning to treat machine access as a product surface of its own, with its own format, rules, and commercial questions.
HTML was built for people. AI agents need something else
People Inc. chief innovation officer Jon Roberts told delegates at Digiday’s Media Strategies London event that around 90% of a typical HTML page can be unnecessary for an AI agent. Headers, navigation, advertising code, related links, and other browser-facing elements all create extra material for a system to process.
That matters because processing is not free. Roberts argued that sending an entire HTML page to an agent means paying to process information that has little to do with the article itself. People Inc.’s proposed answer is a cleaner delivery format: the article in Markdown, plus the metadata that helps a machine judge what it is reading.
The metadata is the more important part. An AI system needs to know who wrote a piece, when it was published, whether it was updated, and which other trust signals surround it. Those details can help an agent distinguish current reporting from an outdated page, or an original article from a lightly changed copy.
People Inc. has already moved through several forms of machine access. It first allowed partners to crawl its webpages, then used Amazon S3 buckets to distribute content while retaining the ability to send updates and retract material. Custom APIs represent the next step because crawling gives publishers less control and makes machines do more work than necessary.
The publisher’s new product is controlled access
The strategic shift is not simply about making content easier for AI to consume. It is about deciding what AI systems receive, how they receive it, and which signals travel with the journalism.
That distinction becomes more important as AI agents and agentic browsers move from experimental tools into discovery interfaces. Digiday cites HUMAN’s 2026 State of AI Traffic report, which found that traffic from AI agents and agentic browsers grew 7,851% year over year. At the same time, publishers still need to separate useful, customer-led requests from automated scraping.
Other publishers are testing similar ideas in different ways. Time has converted pages from HTML to Markdown and has started serving ads to AI agents. Forbes is testing Markdown versions and an MCP layer for its Vetted product reviews, while also exploring how content might be presented differently to machines and people. The Economist is experimenting with making some material outside its paywall easier for agents to read.
These examples point to a new editorial layer between publishing and distribution. A publisher may still create one article, but it increasingly needs to decide whether that article is being read by a person in a browser, summarized by an assistant, used inside a recommendation, or retrieved by an agent completing a task.
People Inc. is reportedly not focused on selling advertising directly to AI agents. Its immediate priority is more basic: create a reliable, efficient, and controllable way for AI systems to access journalism. That is a practical response to a market where publishers are under pressure to find durable businesses, including after People Inc. laid off about 6% of its workforce.
The implication is bigger than Markdown. Publishers are moving toward machine-readable products that preserve context, freshness, and provenance instead of handing AI systems a page and hoping they interpret it correctly. The next fight over journalism may therefore happen before a reader sees a headline, at the access layer where publishers define what an AI agent is allowed to understand.