This is the third post in a short series I am writing about AI literacy for technical writers. The first two covered what AI is good at and where it falls short. This one is about writing content that works for AI as well as humans.
It turns out that the qualities that make documentation work well for human readers are largely the same qualities that make it usable by AI. Which, if you have spent any time thinking about what good technical writing actually is, should come as no great surprise.
What do we mean by “structured for AI”?
"Structured content" is one of those terms that means different things depending on who you're talking to and when. If you've been around technical communication for a while, you probably associate it with XML-based authoring like DITA and DocBook, where structure is enforced by schemas and tags. That's still valid, and XML does lend itself well to AI because the structure is unambiguous and machine-readable. But when people talk about structuring content for AI, they usually mean something less technical than that.
Structured content for AI is more about clear headings, consistent terminology, explicit language, and chunks of information. Good authoring practice, essentially. You don't need to be writing XML to produce content that AI can work with effectively. But if you are, your content will already be in a good place for AI.
What does your content need for AI?
AI processes text sequentially and doesn't browse, skim, or read between the lines the way a human does. That’s why it works best with content that is explicit, consistent, and well-organised. And that’s what is usually meant by well-structured content for AI.
Sound familiar? Yep, that’s what good technical communication already is.
There are also more technical approaches, such as semantic markup and metadata schemas, but in practice those tend to be the responsibility of developers and content architects rather than the writers themselves.
Clear headings that provide context
Clear and consistently structured headings are important as they tell AI what an article or section is about. AI also uses headings to understand how sections relate to each other.
So it is important to use meaningful and specific headings that give context. Vague headings like “More information” or “Next steps” should be replaced with more exact headings, such as “Where to find further support” and “How to set your preferences”.
Metadata and context at the top
AI processes what comes first most reliably. A document that opens with a clear statement of what it covers, who it's for, and what version it applies to gives AI the frame it needs to interpret everything that follows. A document that buries that information (or has information missing) leaves AI guessing.
One idea per chunk of information
AI works best when each chunk of content covers one idea. This means we should make our articles, and the sections inside them, focus on one idea or process. No combining different ideas and processes into the same piece.
This approach has been common in technical writing for decades, especially when using XML. It’s an important part of making content reusable when producing content to scale.
Information has to be explicit
AI is not as good as humans at filling in the gaps in communication. It can’t read between the lines and figure out the context as well as a real user. So we need to make sure our content is clear and provides the full context:
State conditions before the instruction, not after
If a step only applies in certain circumstances, state it at the beginning of the step rather than weaving it into the text. This works for humans too, as users may also complete the first part of the instruction before reading the rest.Define terms the first time you use them
Again, another thing that’s best practice in technical communication. If there are specialist terms or acronyms in the text, make sure to define them the first time they appear in each article. Or link to a definition. Don’t leave AI to try and figure it out, as it may misinterpret or “hallucinate” the meaning.Use annotations that are separate from the main paragraphs.
Don’t weave in important notices, warnings, and information about different versions into the main paragraphs. They should be called out in separate structures like notes, cautions, and warnings, so that AI and humans can quickly identify them as exceptions.
Consistent terminology
With AI, consistent terminology matters more than you might think. If your docs refer to the same thing as a "workspace" in one place, a "project" in another, and an "environment" somewhere else, a human reader will probably figure it out from context. AI may treat them as three separate things or confidently merge them when it shouldn't.
Consistent vocabulary is a best practice staple in tech comm, and now it is even more important as AI can depend on it.
Structured format over narrative prose
Use tables, numbered steps, bullet lists, and similar “content structures” in your article to help AI process the information. These structures all work better than traditional paragraphs, as they communicate relationships too:
Table: Things that share some of the same properties or attributes
Bullet list: Things that are related, but their order is unimportant
Number list: Steps that need to be completed in order.
It’s not wrong to provide the information in prose; it’s just less effective. Humans can’t scan the information quickly, so have to read it in full (and can be deterred by large blocks of text). AI isn’t good at interpreting relationships and sequences unless given specific instructions and information. Using the appropriate content structures makes life easier for everyone.
Real-life projects and AI
In my own experience, making my content work for AI has not required much extra effort. The technical teams were able to get the AI bots working well without any significant changes to the content. I suspect that’s partly because my background in XML-based authoring has stuck with me. Even when I am working in Markdown or plain text, I still tend to think about content in the same structured way. Old habits, it turns out, can be quite useful.
Does your existing content measure up?
The good news is that if you already follow technical writing best practices, your content is probably in reasonable shape for AI. Specific headings, consistent terminology, explicit language, structured formats -- these are not new ideas. They are the foundations of good technical communication. AI just gives you another reason to care about them.
That said, it is always worth casting a critical eye over your back catalogue. We all have those weeks where deadlines are tight, resources are stretched, and a few corners get quietly cut. Content that was good enough for a human reader in a hurry may need a bit of attention before you would trust AI to do something useful with it.
