Knowledge base

Answer-first page structure

Put the conclusion in the first paragraph so a model can quote it, then add evidence, limits and sources instead of opening with slogans.

Answer-first is a concrete instruction: state the answer in the first two sentences, then support it. Extraction looks for sentences that stand on their own, not for a tidy page skeleton. A reader should be able to stop after the opening and still hold the answer. Opening with a slogan, a carousel or a consent overlay hides the paragraph a model would otherwise quote.

A workable shape is answer, evidence, limits, source. The answer must read correctly without the surrounding page. Evidence names what was measured and how. Limits state when the claim applies. The source points to a page or dataset a reader can check. Keep each part short and single-purpose, and rewrite any sentence that needs the previous one to make sense.

Three patterns break this: answers stored in images, key parameters hidden in collapsed panels, and sentences that rely on pronouns from an earlier paragraph. A model does not expand an accordion or rebuild context. Anything meant to be cited must exist as plain text near the top of the page.

Here, answer-first work counts toward AI citability, the 25-point dimension. AI visibility, worth 30 points, requires live model answers. The evidence table lists fact, value and status, where status is pass, partial, fail or unavailable. Status values are written as words, so a reader with no styling still sees the difference.

Write the Chinese and English pages separately. The questions differ, so the opening sentence should differ too, even when the underlying facts are the same. After editing, open an example report and check that every claim has a paragraph behind it, and that the paragraph appears on the page before any interactive element.