A recurring frustration: a company has invested seriously in content. The articles are researched, accurate and better than what competitors publish. And when an AI assistant answers a question the article covers thoroughly, it cites someone else.

This is rarely a quality problem. It is usually one of four structural problems.

1. The answer is buried

The article answers the question — in paragraph nine, after context, background and a discussion of why the question is more complicated than it appears.

An extraction system reads the top of the document with the most weight, and looks for a passage that answers the question directly. Nuance placed before the answer reads as absence of an answer.

Fix: answer in the first two sentences under a heading that states the question. Then qualify, contextualize and complicate at whatever length you like. You lose nothing with human readers, who also prefer knowing the answer before the discussion.

2. There is no extractable claim

The content is thoughtful and contains no sentence that can stand alone as a fact. Everything is "it depends," "in many cases," "organizations should consider."

This is often intellectually honest and it makes the content uncitable. A machine looking for something to quote finds nothing quotable.

Fix: commit to specifics somewhere. Give the number, the range, the typical case, the rule of thumb. You can qualify it in the next sentence. "Most implementations take six to ten weeks; complex integrations run longer" is citable. "Timelines vary considerably" is not.

3. The page is structurally opaque

One long flow of prose. Headings that are clever rather than descriptive. No lists, no definitions, no structure signalling what is where.

Extraction relies on structure to locate relevant sections. A well-written wall of text is hard to navigate mechanically.

Fix: descriptive headings phrased as the questions people ask. Definitions near the top. Lists where the content is genuinely a list. This is the same structure that helps skimming humans.

4. The source is not corroborated

The claims appear nowhere else. For a machine deciding whether to repeat something, a claim existing on exactly one site — a site with a commercial interest in the claim — is risky.

Fix: get cited elsewhere. Original research that others reference, contributions to industry publications, data others want to quote. Corroboration is slow to build and it is the difference between being a source and being a page.

The diagnostic

Take your best article and the question it answers. Ask three different assistants that question. If you are not cited, look at who is — then compare their page against yours on these four axes specifically.

In our experience the cited page is frequently worse content by any editorial standard. It is shorter, shallower and less accurate. It is also structured so that the answer can be lifted out in one piece, which on this surface beats being right at length.

That is not an argument for publishing worse content. It is an argument for putting the answer where it can be found, then being right at length underneath it.