Self-contained content

Self-contained content means each passage stands alone. It matters because a model usually retrieves a fragment, not your page.

What breaks self-containment?

Backward references. Phrases like “as mentioned above”, “as we discussed”, or “this approach” without naming the approach all assume the reader arrived through the preceding text. A retrieved fragment has no preceding text.

How do I fix it without repeating myself endlessly?

Name the subject again at the start of each section rather than relying on a pronoun. Restate the minimum context a reader would need if this paragraph were the only thing they saw.

Does this hurt readability for humans?

Slightly, and it is worth it. Naming the subject reads a little more formally, but it also helps readers who arrive mid-page from a search result, which is most of them.

What else breaks a passage apart from backward references?

Four things, all of which read perfectly in place:

• Unexpanded acronyms. The expansion is in the introduction, which the fragment does not carry.
• Relative time. “Last year”, “recently” and “currently” are meaningless in a passage retrieved eighteen months later, and dangerous when they are quoted as current.
• References to page furniture. “The table above”, “see the sidebar”, “as the chart shows” — none of it travels.
• Figures without their unit or their date. A number lifted out of context is the single most quotable and most dangerous thing on a page.

How do I test whether a passage stands alone?

Copy one paragraph into an empty document and read it as though you had never seen the page. If you cannot tell what product, company, version or period it is about, neither can a retrieval system, and it will pass over the passage in favour of one that identifies itself. This takes about thirty seconds per section and catches almost everything.

Where does the context actually go?

Into the first sentence under each heading, and into the heading itself. A heading phrased as the question a reader would ask carries its own subject and gives the passage beneath it something to attach to. Naming the subject once per section costs a few words and is the difference between a quotable fragment and an orphan.

Is this the same thing as writing for chunking?

They are two halves of one problem approached from opposite ends. Chunking is the mechanism that cuts your page up; self-containment is the editorial discipline that makes the pieces survive being cut. You cannot control where the cut lands, so the only available strategy is to make every likely piece independently intelligible.