Plan requirement
| Subscription | Suite Team or higher, Knowledge Professional or higher |
| Access | Agent |
Seven habits that make an article usable by an AI agent, all of which make it better for people too. One article, one question, answered at the top.
1. One question per article
The most important. An article that answers one thing can be quoted; one that answers five is summarised into vagueness.
2. Answer in the first paragraph
Then explain. The opening carries weight twice over: search engines build the article's description from its first 140 characters, and that snippet is what a reader decides on. An article that opens with three paragraphs of context shows context where the answer should be.
3. Title it as the question
"Why can I not sign in?" rather than "Authentication troubleshooting". It helps matching, and it tells the AI what the article is for.
4. Say the conditions out loud
If something applies only on a particular plan, or only to business customers, write that in the article. An AI cannot infer it, and it will happily tell a customer they can do something they cannot.
5. Use real structure, in the article body
Headings, numbered steps, lists. Structure is meaning, and both a screen reader and an AI use it to work out what belongs together.
It has to be in the body. Generative answers and generative search cannot read attachments uploaded in your articles, so a procedure that lives in an attached PDF is invisible to them, and so is anything explained only in a video. The remedy is the obvious one: put the content of the attachment in the article. Web crawlers, by contrast, do read content that is in the page source even when a UI element such as an accordion hides it, so those two behave differently, and which you rely on changes the outcome.
6. Remove contradictions
Two articles disagreeing is worse than one being missing. Find them by asking your AI agent the questions where it gives inconsistent answers.
7. Mind what the agent is allowed to read
Writing well is not enough if the article is out of scope. An AI agent answers from the knowledge sources connected to it, and for a signed-in customer that can include restricted articles, while a visitor gets public ones only. So the same article can be usable for one person and absent for another.
More sources is not the fix either. Zendesk puts it carefully: too many can in some cases lead to reduced accuracy and increased latency, and it recommends keeping the number within a reasonable limit without naming one. So this is a judgement to make deliberately, not a threshold to stay under.
What not to do
Do not write for the AI. Articles stuffed with phrasings, or written in stilted language to be machine-friendly, read badly to people and are not better for AI either.
Write clearly for a person with a problem. That is what works.
Comments
0 comments
Article is closed for comments.