August 21, 2026
Enterprise websites were built for people. The CMS assembles a page, navigation and search help someone find it, and the visitor uses the branding, layout and surrounding content to work out what it means and whether to trust it.
Agents do not use a website in quite the same way. They may browse a page, retrieve a fragment or query an API, but the page is rarely the destination. It is evidence used to answer a question, compare options or complete a task.
Most enterprise platforms - Optimizely, Adobe, Sitecore, Salesforce, SAP - can now expose content through an API. But that does not make the content agent-ready.
The problem is that most enterprise content was never structured as knowledge. It was structured as landing pages, campaign pages, articles, PDFs and regional variants, often written and managed separately. An API simply gives a machine faster access to the same duplication, gaps and contradictions.
An API over page-shaped content is still page-shaped content.
For an agent to use an organisation as a reliable source, it needs more than access.
Entity clarity: It should be obvious who the organisation is, what it offers and how its products, services, locations, people and policies relate. Years of campaigns, acquisitions and regional variation often make this surprisingly hard to establish.
Canonical truth: If a product page, support article, PDF and regional site give different answers, which one should the agent trust? Important claims need a clear owner, scope and status.
Useful paths: Human navigation is organised around menus and page hierarchies. An agent needs relationships: which policy applies to which market, which product replaces another, what evidence supports a claim and where to look next.
Capabilities: Agents increasingly need to do something with the information. Comparing options, checking eligibility, configuring a product or requesting a quote all require clear rules and interfaces, not another page of copy.
This is why llms.txt has so far made little difference. It offers agents a curated map of a website, but most implementations point to the same pages they could already find. If the underlying content has not been organised around meaning, the map adds very little. The evidence that it improves citation in consumer AI search is thin, and Google has said it is not needed for AI Overviews or AI Mode.
A more useful approach would be closer to an AGENTS.md for the organisation’s digital estate: what it knows, where the authoritative version lives, how the pieces relate and what an agent can do. It might use HTML, structured data, an API, a manifest or an emerging agent protocol. It does not need to be a text file at all.
The format is secondary. The real work is restructuring content as organisational knowledge rather than material for assembling pages. The website can still render that knowledge for people, while agents get the context and routes they need to use it reliably.
Your CMS might have an API now. Your content still isn’t agent-ready.
Next in this series, Richard Trigg, our Chief Design Officer, looks at what happens to the brand itself once those structured facts get pulled out and shown without any of the page around them.
