What is this?
This check asks whether a reader who stops after the first two sentences under a heading already has the answer. AccessKnight grades it from the page's own headings and body text and awards 0 to 8 points, passing at 7 — so a section that builds toward its point instead of leading with it forfeits most of the eight.
The structure this rewards is the inverted pyramid, which newsrooms have used for a century and search has rewarded for a decade. State the conclusion, then support it. What fails is the essay shape: a paragraph of context, a paragraph of history, and the actual answer buried in the third paragraph where a retrieval system chunking the page will not find it attached to the heading that promised it.
A direct answer is also not several things that resemble one. Restating the heading as a sentence is not an answer; promising one — "there are a few things to consider here" — is not an answer; nor is a link elsewhere. The test is whether the opening sentences still make sense quoted with nothing else around them, which is also why the answer belongs in an ordinary paragraph rather than a callout or pull-quote.
Who does this affect, and why does it matter?
Answer engines are affected first: ChatGPT search, Copilot, Perplexity, Google AI Overviews and AI Mode, and Claude all assemble a response from passages rather than whole pages. When the sentences under a heading open with throat-clearing, the passage an engine would lift holds no answer — the page is retrievable but not usable as one.
The human cost lands on anyone who does not read linearly. Screen reader users jump to a heading and hear what follows it, so a build-up paragraph is a wall between them and the information they navigated to. Skim-readers on a phone behave the same way with their eyes. Both are asking the page the same thing the engine is: answer the heading you just wrote.
Extractability is the 30-point pillar measuring how readily an engine can lift a clean, self-contained answer from a page, and this check takes 8 of those points — tied with question-style headings as the largest in it. The row passes at 7 or above, warns from 4 to 6, and fails below 4. It is not gating, so a weak score reduces the pillar without capping your headline score.
The mechanic is passage retrieval. An engine splits a page into chunks and ranks the chunks, which means the unit competing to be quoted is the paragraph under your heading, not the article as a whole. Leading with the answer is a precondition for that paragraph to stand on its own. It removes a blocker to extraction; it does not promise that any engine will use it.
The same edit pays outside the AI channel, which is the honest reason to make it even if the AI answer surfaces never send you anything. A self-contained opening paragraph is exactly what classic search has always drawn featured snippets from, and it is what a reader arriving cold from any source needs. You are not betting on one retrieval mechanism; you are writing the answer first, which was good practice before any of these systems existed.
How do I fix it?
Move the answer to the top of each section. Read only the first two sentences under a heading, ask whether they answer it with nothing else around them, and where they do not, promote the sentence that does into an opening block of roughly 40 to 60 plain words. Reorder rather than delete.
- Read only the first two sentences under each h2 and h3 on the page, ignoring everything below them — that is close to what the grader weighs for this check.
- For each section, ask whether those two sentences answer the heading with nothing else around them. If they need the paragraph below to make sense, they are not yet an answer.
- Move the answer to the top and reorder rather than delete: an opening block of roughly 40 to 60 words that leads with the heading's key phrase, then the context, history, and caveats underneath it.
- Cut the openers that only promise an answer — "there are several factors to consider", "let us take a look at" — since they consume the two sentences the check reads.
- Leave the answer as an ordinary paragraph rather than a callout, pull-quote, or card. AccessKnight's sampler flattens nested markup, so a styled container does not hide the text from this check — but retrieval pipelines differ, and a plain paragraph is the form that survives all of them.
- Re-scan with "Also grade AI extractability" ticked and confirm the "Direct answer in first 1–2 sentences" row reports 7 or 8 of 8.
<h2>Shipping and delivery</h2>
<p>Shipping is something we get asked about constantly, and over
the years we have worked hard to make it as painless as we can.
Below you will find everything you need to know about our
process, our carriers, and what happens if a parcel goes
missing in transit.</p>
<p>Standard orders leave the warehouse within one business day
and arrive in three to five business days.</p><h2>How long does shipping take?</h2>
<p>Standard orders arrive in three to five business days. We
dispatch within one business day of payment, ship with UPS
Ground, and email a tracking number at dispatch. Express
delivery arrives in one to two business days for $12.</p>
<p>Shipping is something we get asked about constantly, and over
the years we have worked hard to make it as painless as we can.
Here is our process, our carriers, and what happens if a parcel
goes missing in transit.</p>How does AccessKnight detect this?
This check is graded by a model, not matched by a rule. AccessKnight builds a content sample from the page — removing script, style, noscript, template, svg, nav, footer, and header, preferring the contents of main over body, and keeping the first 40 headings and 6,000 characters of text — then sends it to Claude in one call that scores all four Extractability prose criteria together. The criterion here: a direct, self-contained answer in the first one or two sentences under headings, scored 0 to 8. Temperature is 0, so the same page should score the same, but a model grade is a judgment and can move between runs. The row passes at 7 of 8 and warns from 4 to 6.
What strings mean this?
If one of these strings is in front of you right now — in a config file, an HTTP header, an AccessKnight report, or another checker’s output — it is describing the failure explained on this page.
- AccessKnight report
- “Direct answer in first 1–2 sentences”
- AccessKnight report
- “Lead each section with a direct, self-contained answer (inverted pyramid).”
- AccessKnight report
- “Graded from content under headings.”
- AccessKnight report
- “Prose extractability (question headings, direct answers, chunking, definitions)”
- AccessKnight dashboard
- “Also grade AI extractability”
Frequently asked questions
How long should the direct answer be?
Where does the sentence printed under my direct-answers score come from?
Why did my direct-answers score change between two scans of the same page?
Why does my whole Extractability pillar show as not applicable?
Does this check apply to my h1 and the paragraph under it?
Where does this come from?
Every external claim on this page traces to one of the documents below — the vendors’ and standards bodies’ own documentation. What AccessKnight adds is the scan, not the standard.