An answer block is the passage directly under a heading that answers that heading, in two or three sentences, without depending on anything else on the page. It carries more weight than any other answer-readiness signal — 20 points of the 100 on GEOBee’s AEO lens — and it’s also the cheapest thing on the list to fix, because on most pages the answer is already written. It’s just sitting in the wrong position.
This is how to write one, and how to find the headings on your site that don’t have one.
What an answer block is
Three properties, all required. It sits first, immediately under the heading, before any setup. It answers the heading rather than introducing it. And it survives being read alone, with no heading above it and no paragraph before it.
The third property is the one most drafts fail. A passage can lead with its answer and still collapse the moment it’s separated from the page, because it says “this” or “the tool” or “as described above” and the referent lives two paragraphs up. Retrieval systems split pages into passages and can surface one without its neighbours, so a passage that needs its neighbours is a passage that breaks. What an answer engine reads covers how that splitting works.
Where 40–60 words comes from
It comes from two constraints meeting, not from any published rule.
The first is truncation. Google states that snippet length has no fixed value and that the snippet “is truncated in Google Search results as needed, typically to fit the device width” (Google Search Central, retrieved 2026-07-30). A passage quoted whole has to survive being cut short and still say something true.
The second is self-containment. The passage has to carry its own subject, which costs words — you can’t lean on the heading to supply the noun, because the heading may not travel with it.
Two or three sentences is where those meet. Under about 40 words you’re usually restating the heading rather than answering it. Over about 60 you’ve started explaining, which is fine — but the explanation belongs in sentence four onward, not in the block.
Step 1: Make the heading a question worth answering
By the end of this step every heading on the page either asks something a reader would type, or is honestly labelled as navigation.
Retrieval matches a question against a heading and the passage beneath it. A heading like “Overview,” “Our approach,” or “Rate limits” gives it almost nothing to match — those are labels, not questions. “How long does activation take?” and “What are the API rate limits?” are questions, and they tell you exactly what the passage below owes the reader.
The useful test runs backwards: if you can’t write a one-sentence answer to a heading, the heading is the problem, not the paragraph. Rewrite it until you can.
Not every heading needs this. “Related reading” and “Changelog” are navigation and should stay navigation. The rule applies to headings that make a claim or promise information.
Verification: every substantive heading on the page can be read as a question, and you can state its answer aloud in one sentence.
Step 2: Promote the answer you already wrote
By the end of this step the first sentence under each heading is the one that answers it.
On most pages the answer already exists, in sentence three or four, after the background. Writers put it there because that’s how good prose works for someone reading top to bottom — context, then conclusion. Extraction doesn’t read top to bottom. It reads the opening.
So move the sentence rather than writing a new one. The context you displaced isn’t lost; it goes after the answer, where it now reads as support instead of preamble. This is the reason the fix is cheap: you’re reordering, not drafting, and the page usually gets clearer for human readers too.
If you genuinely can’t find an answering sentence anywhere in the section, that’s a different finding — the section doesn’t answer its heading at all, and no reordering will fix it.
Verification: deleting everything except the heading and the first sentence still leaves a true, direct answer.
Step 3: Cut every dependency on the rest of the page
By the end of this step the block names its own subject and refers to nothing outside itself.
Work through the block and replace anything that points elsewhere. Sentence-initial “it,” “this,” and “that” get their nouns back. “The tool” and “our platform” become the actual product name, at least once. “As described above” becomes the thing described above, or gets deleted. A figure or table referenced as “the table below” needs its own restated context, since the table may not travel with the passage.
Naming the subject explicitly matters more than it looks. Titles, navigation, and headers are among the first elements discarded when a page is converted to the text a model reads — what AI crawlers extract has the full inventory — so the passage can’t borrow its subject from the page furniture.
Verification: no pronoun in the block lacks an antecedent inside the block, and the subject is named at least once.
Step 4: Read it with the page removed
By the end of this step you’ve confirmed the block works for someone who has never seen the page.
Copy the heading and the block into an empty document and read them together. Two questions decide it. Does this answer the question for a stranger? And if you had asked that question and received exactly this back, would you consider it answered?
A block that passes both is finished. One that passes the first and fails the second is usually too hedged — it’s technically responsive and commits to nothing. Add the figure, the named requirement, or the plain yes or no that makes it resolve.
Verification: the heading-plus-block pair reads as a complete question and answer with no other context available.
Before and after
The before is a real pattern, not a strawman — an accurate paragraph that answers its heading in the last sentence:
## Rate limits
Every API has to protect itself against bursts of traffic, and ours is no
exception. We spent considerable time tuning the thresholds so that normal
usage is never affected by them. The limit is 1,000 requests per minute per
API key, and requests over that return a 429 with a Retry-After header.The answer starts at word 32 of 52, the heading is a label rather than a question, and “them” and “that” both point backwards. The rewrite moves the answer up, makes the heading a question, and names what it’s talking about:
## What are the API rate limits?
The limit is 1,000 requests per minute per API key. Requests over the limit
return `429 Too Many Requests` with a `Retry-After` header giving the number
of seconds to wait. The ceiling is per key rather than per IP, so splitting
traffic across servers does not raise it.That’s 48 words, and every sentence resolves. The tuning-and-thresholds context isn’t deleted — it moves below, where it supports the answer instead of delaying it.
What isn’t an answer block
A summary box at the top of the page isn’t one. It answers the title, once, and does nothing for the eight headings below it.
An FAQ appended to the bottom isn’t one either. It may help readers, but it doesn’t fix a section whose own heading goes unanswered — and FAQ markup is entity description rather than a route into anything.
A sentence that restates the heading in different words isn’t one. “Rate limits are the limits placed on request rates” occupies the position without doing the job, and it’s the most common failure in drafts that know the rule.
Checking every heading on a site
Reading the first sentence under every heading works on one page and not on four hundred. What scales is measuring the opening passage under each heading and sorting by length, which surfaces both failure modes at once — the openers that are far too long to be answers, and the headings with no passage under them at all.
On markdown sources, that’s one line:
awk '/^##+ /{h=$0; want=1; next} want && NF {print NF"\t"h; want=0}' page.md | sort -rnIt prints the word count of the first paragraph under each heading, longest first. Anything well above 60 words is a candidate for step 2.
Run against this blog, the top row is a 137-word opener under “What never makes the trip” in what an answer engine reads — a section that spends its opening listing examples before it characterises them. Word count is a proxy and not a verdict: it finds candidates, and a person still decides whether the first sentence answers the heading. That judgement is why this check is a read-through job rather than a fully automated one.
For the diagnostic pass across all five common faults, of which this is the first, see five checks for a page that ranks but never gets cited. In GEOBee the answer block is the heaviest signal on the AEO lens, scored per URL with published weights, and findings and fixes shows how a failing heading becomes a finding with a fix attached.
Worth restating one limit. A page with an answer block under every heading is one a machine can read cleanly and lift from. Whether anything chooses to lift from it depends on competition, authority, freshness, and the question asked — none of which live on your page. Answer Engine Optimization: a page-level guide sets out the other seven signals this one sits alongside.