How to write a single extractable answer block for a cornerstone page
- An extractable answer block is a single, self-contained passage (usually one to three sentences) that states a complete answer to a page's core question, built to survive being lifted out of the page entirely and dropped into a generated AI answer.
- Placement matters as much as wording: the GEO paper's own visibility metric weights word position specifically, not just word count, which means the same correct sentence does more work in position one than position five.
- The test for whether a passage qualifies: delete everything else on the page and read it alone. If it still makes complete, specific sense with no missing antecedent ("this," "the above," "as mentioned"), it passes. If it needs the surrounding paragraph to resolve, it fails.
- This piece is a construction recipe for one specific block, not a general writing-craft guide. For the broader set of extractability patterns (original data, self-contained sections, specific quotation, structured formats), see how to write content that AI models can actually quote.
- Every worked example below is a real, currently-published passage from this site, not an invented before/after, so you can check the actual page each one comes from.
How to write content that AI models can actually quote covers five general patterns for extractable writing across a whole page. This is narrower and more mechanical: the specific recipe for the one block a cornerstone page needs most (the passage most likely to get lifted whole into a generated answer), with real, published examples instead of invented ones, and a checklist to audit whether a given passage actually qualifies.
What's the recipe, in order?
1. It states a complete claim, not a pointer to one. The block has to contain the actual answer, not a promise that the answer is coming ("there are several factors that determine this, read on to find out"). If a reader could stop after this passage and know the answer, it qualifies. If they need the next paragraph to actually learn anything, it doesn't.
2. It resolves with zero external context. No "this," "the above," "as discussed," or any pronoun whose antecedent lives outside the block. A model extracting this passage in isolation (which is exactly what happens when it gets lifted into a generated answer) has no access to the sentence before it unless that sentence is included. Every noun the passage depends on has to be named inside it.
3. It sits as close to the top as the page's structure allows. This one isn't about the words at all. It's about position. The GEO paper's own visibility metric (Position-Adjusted Word Count) explicitly weights how early and how prominently a passage sits, not just whether the right words exist somewhere on the page. The same well-written sentence does measurably more work as the first line of a section than as the fourth paragraph down.
What do four real examples look like?
Rather than invent before/after pairs, here are four passages already published and live, each one written to pass the checklist above, pulled from the actual opening lines of existing pages on this site.
From AI visibility score:
"An AI visibility score is a composite metric (usually scaled 0–100 or expressed as a percentage) built from three things: how often a brand is mentioned in AI-generated answers to relevant queries (citation or mention frequency), how favorably it's framed when it is mentioned (sentiment [...]), and how accurately the model represents your actual facts (information accuracy [...])."
This names the concept, states what it's made of, and needs nothing before or after it to be complete. Delete every other word on the page and this sentence alone still fully answers "what is an AI visibility score."
From what is entity consistency:
"It means your brand resolves as one confidently-disambiguated node across every source a model cross-references (your own site, Wikidata, third-party profiles, structured data) rather than as several partial, inconsistent, or contradictory records that a model can't confidently merge into one identity."
Notice this one is the direct answer to the FAQ question above it ("What does 'entity consistency' actually mean?"), restated as a standalone claim rather than assuming the question is still in view. That redundancy is deliberate, not wasteful: a model lifting the FAQ answer alone still gets a complete, self-contained passage.
From does llms.txt actually get read by anyone?:
"Ahrefs' June 2026 study of 137,210 domains found 97% of published llms.txt files received zero AI-crawler requests in a full month, and where retrieval bots did request the file, they were just 1.1% of total requests to it."
This is the pattern from Pattern 2 in the craft guide (original data with a named source and sample size) applied specifically as the page's lead answer block rather than buried mid-page. It's a complete, checkable claim with nothing left dangling.
From what does it mean to be "cited" by an AI model?:
"A model can fetch and use a page during retrieval without it ever appearing as a visible citation in the final answer. OpenAI's API returns a full list of consulted sources separately from the smaller, curated set that gets an actual citation annotation, and Anthropic's API attaches citations only to the specific pages the final text actually references, distinct from every page it fetched."
Longer than the others, because the concept genuinely needs the extra clause to be complete. The "test" isn't a word count, it's whether removing any part of it would leave the claim incomplete or ambiguous. Cutting this to one shorter sentence would lose the actual mechanism being described.
What's the self-audit checklist?
Before treating any passage as your page's answer block, run it through these four checks:
| Check | Pass | Fail |
|---|---|---|
| Does it state the complete answer, not a pointer to one? | "X is defined as..." | "There are several factors that explain X..." |
| Does every pronoun resolve inside the passage itself? | Names the subject directly each time | Relies on "this," "it," "the above" pointing outside the block |
| Does it survive being read with everything else on the page deleted? | Reads as a complete, specific claim alone | Needs the prior paragraph to make sense |
| Is it positioned at or near the top of its section? | First sentence of the H1 or the first H2 section | Buried after throat-clearing setup |
A passage that fails any one of these can still be good writing in context. The checklist isn't a quality bar for the whole page. It's a narrower test for whether this specific block is doing the specific job of being the thing a generative engine lifts out whole.
Frequently asked questions
What is an "extractable answer block"?
A single, self-contained passage (usually 1-3 sentences, sometimes a short block) that states a complete answer to the page's core question without requiring anything before or after it for context. It's built to be lifted whole into a generated AI answer, not just to read well as part of a longer page.
Where should the extractable answer block go on a page?
As close to the top as the page's own logic allows: ideally the first sentence after the H1, or the first sentence of the first section a reader lands in. The GEO paper's own visibility metric explicitly weights word position, not just word count. A correct, well-written answer buried three paragraphs down does less work than the same answer stated first.
How long should an extractable answer block be?
Long enough to be a complete claim, short enough to be lifted as one unit. In practice, one to three sentences for most pages, occasionally extending to a short block (a definition plus one supporting fact) when the concept genuinely needs it. The test isn't a word count, it's whether the passage still makes complete sense with everything else on the page deleted.
Does every page need one of these, or just cornerstone pages?
Every page benefits from one, but it matters most on cornerstone and pillar pages: the ones meant to be the canonical answer to a specific, well-defined question, where a generative engine has the clearest reason to go looking for exactly this kind of passage in the first place.
Sources referenced
- Princeton/Georgia Tech GEO paper (arXiv 2311.09735): source of the Position-Adjusted Word Count metric referenced above; full explanation in the GEO paper explained
This piece is a narrow construction recipe for one block. For the broader set of extractability patterns across a whole page, see how to write content that AI models can actually quote; for the research behind why position and specificity matter, see the GEO paper explained.
