What does AI do to the help-centre content itself?

When the AI is the primary reader of your help-centre content, the content's quality determines the AI's quality, which means the help-centre team's job shifts. Writing for humans alone is no longer enough. The content has to be unambiguous to a model, self-contained per article, and structured so retrieval can find the right piece. The shift is editorial, not technical.

A help-centre article was written five years ago for a customer to read. It says "if you have a Pro account, this works differently; see the related article." A human reader follows the link. The AI reading that article does not always know which one to follow, or whether the customer is Pro. The article was correct for the old reader and confusing for the new one.

What people in the field are saying

kdschemin's "Reliability is the product" argues that the reliability of the AI depends on the data layer underneath, and the help centre is part of that data layer now, not a separate concern.

What changes in how help-centre articles are written?

Four shifts. Self-containment: each article includes the context needed to answer the question, not "see the linked article." Disambiguation up front: "this applies to Pro accounts only" or "this is for orders not yet shipped" stated at the top. Structured information: prices, timelines, policy thresholds in a format the model can extract. Consistent vocabulary: the same thing is called the same name across articles.

What about the SEO version of the same article?

Often it conflicts. The SEO version is written to rank, with keywords and long-form structure that helps Google. The AI version is written to be quoted, which is shorter and more direct. Some teams maintain two versions; some change the SEO version too. Either is defensible; both is more work than most teams expected.

Where does this go wrong?

The team that owns the help centre is separate from the team that owns the AI. The two do not coordinate. Articles get written for humans and the AI is left to make sense of them. The AI's answers degrade over months as the gap between content and AI-fit accumulates.

What does the new editorial discipline look like?

Each new or updated article goes through a brief AI-fit check: is the context self-contained, is the policy stated unambiguously, can the model extract the structured facts. Articles that fail the check get rewritten before they ship. The team responsible for AI quality has visibility into help-centre changes the same way it has visibility into the AI's prompts.

What is the smallest first step?

Audit the top ten articles by AI traffic. Read them with the question: would the AI quote this correctly to a customer who only got the AI's reply? Rewrite the ones where the answer is no. The pattern that emerges from that audit tells you what the rest of the help centre needs.