API

An API is an agreed format that lets one piece of software request something from another and get a predictable answer back. It sets out what can be asked for, how to phrase the request and what the reply will look like, so the systems on either side don’t need to know how the other works inside.

Web APIs became common in the 2000s, when publishers and platforms began opening their content and data to outside developers.

An endpoint is the address a request goes to, such as one for articles and another for authors. Authentication, usually a key or token, controls who can ask and what they can see. Rate limits cap how many requests a system can make in a given time. In a headless CMS, the content API is how content leaves the system: the website, the app and every other channel request it from there.

Why it matters for content

APIs turn content from something fixed to one page into something other systems can use. A website can pull the current product description, an app can show the same help article and a support tool can check a draft against the one maintained version of the guidance. Every channel draws on the same source, so a correction made once reaches all of them.

That only works if the content behind the API is structured. An API passes on whatever it’s given: if a “summary” field sometimes holds a summary and sometimes holds the opening paragraph, every system that requests it inherits the confusion. Field names, required fields and content types become an agreement with other teams, and changing them can break things those teams have built. Content model changes start to need the same care as code changes, with versioning and notice.

Publishing through an API also changes who controls the content. Once it leaves, the publisher no longer controls every place it appears. When the Guardian opened its content API in 2009, developers could store articles for 24 hours at most, and were warned that keeping them longer risked republishing a story withdrawn for legal reasons.

How AI changes the picture

Assistants and agents reach content through APIs. A chatbot that answers from approved guidance, a writing tool that checks drafts against current rules and an agent that updates product records all depend on a controlled way to request and change content. MCP, the Model Context Protocol, now supported by many AI tools, gives models a standard way to discover those connections and decide when to use them.

That raises the stakes on what an API exposes. Draft, expired and approved content need to be distinguishable in the response, or a model will treat them as equally usable. Permissions matter more when the caller is software acting for a person, because an agent with write access can make a lot of changes very quickly. For content teams, “can AI use this?” usually means “is there a structured, governed way for a system to reach it?”