People Inc. turns to custom APIs to serve journalism to AI agents

Publisher People Inc. wants AI agents to stop wading through full webpages to read its journalism, and is turning to custom APIs to make that happen. Chief innovation officer Jon Roberts told delegates at Media Strategies London, an event organised by Digiday, that the company is building the interfaces to serve its content directly to automated systems.

Through these interfaces, agents receive an article’s text in Markdown, accompanied by its author, its publication or update time and what Roberts described as other “trust signals that matter.”

“If you are an agent and you’re reading an HTML page, 90% of the content is junk meant for a browser to read, not an agent,” Roberts said.

That surplus is distracting as well as costly to process, he explained, since an agent pays for every token. “If you throw it the entire HTML header, that’s abuse,” he added.

The APIs mark the publisher’s third approach. Partners originally reached its content by crawling the site; People Inc. later switched to Amazon S3 buckets, a setup that let it send updates and pull back retracted material. In Roberts’ view, crawling remains inefficient for AI systems seeking information.

According to Roberts, AI firms are starting to value knowing when a publisher created or revised a piece. People Inc. is embedding such details in content used by its own AI tools and applying what it learns to its partner offering. Advertising directly to agents is not a priority, he said.

The change coincides with a wider reshuffle of investment priorities at the company, which let go of about 6% of its employees last week.

Data from HUMAN’s 2026 State of AI Traffic report puts year-on-year growth in visits by AI agents and agentic browsers at 7,851%.

Elsewhere in the industry, Time has switched its pages to Markdown and now sells advertising to agents that visit, while The Economist is testing ways to make its non-paywalled marketing and B2B sales content easier for agents to read. At Forbes, the Vetted product-review section now has Markdown versions and an “MCP layer.”

A ChatGPT plugin for product recommendations tied to that section is planned but not yet live.