← All posts

Answer capsules: why AI cites some sections and skips others — a free checker

· 3 min read · By Mikhail Kuzmitskii

View as Markdown ↗

Spoiler for anyone who skims: an AI engine quotes the paragraph right after a heading if it stands on its own, and a free checker shows which sections do — ours scored 44%.

We shipped the checker for other people’s pages, then I pasted in our own GEO guide, expecting a pat on the back — we write about this daily, after all! It came back at 44%, which is why this post exists.

An AI engine doesn’t read a page top to bottom. Its RAG pipeline splits it into chunks by heading, and the section that makes it into the answer is the one that stands on its own — no “as mentioned above”. That self-contained first paragraph of ~40–75 words is what we call an answer capsule, and there’s a free tool for it: the answer-capsule checker.

Answer capsules: why AI cites some sections and skips others — a free checker

What the checker scores

Paste an article (or a URL — the text loads itself), and for every H2/H3 section the checker returns a verdict: capsule present (✓), short but punchy (≈), no capsule (✗) — plus an overall “capsule adherence” percentage. Everything runs in the browser, $0, and the text never leaves the tab, which seems the only honest way to build a tool people paste drafts into. Every failure comes with one of four reasons:

  • section opens with a list — a list needs an answer paragraph in front; on its own it usually doesn’t get cited.
  • opens with filler — “in this section we’ll cover…” instead of the answer (who is that for, honestly?).
  • answer buried — the first paragraph runs over 75 words, so it can’t be lifted whole.
  • unbound reference — “this works like…” (what is “this”? the chunk is out of context — a pointer to nowhere).

We ran it on our own articles — 44%

Hypothesis: our guides are fine; we know the rules. Check: paste What is GEO and the zero-click article into the checker. Result: 44% (4 capsules out of 9 sections) and 50% (3 of 6). We write about GEO daily — and still half our sections open with a list or bury the answer in an 80-word paragraph. Half! Both articles went straight into the capsule-refactor queue.

This post, by the way, is written in capsules — paste its URL into the checker and see.

How to fix a section without a capsule

The fix is one paragraph: put a direct answer to the heading’s question — 40–75 words, no references to neighbouring sections — before the list or table. A real example from our own guide, where the “How GEO differs from SEO” section opened straight with a list:

Before: ## How GEO differs from SEO → a five-item list, immediately.

After: same heading → a paragraph: “GEO optimizes not your ranking but the probability of entering the AI answer itself: SEO fights for the blue-link click, GEO for the engine quoting your paragraph and naming your brand. The metrics change (citation share instead of positions), and so does the unit of work — a section, not a page” → and only then the list.

When the checker isn’t enough

The checker shows WHERE capsules are missing — it doesn’t write them for you. That’s what the AEO Content Pack ($49) is for: given a URL, it generates a capsule draft for each of your H2s, ready-to-paste FAQPage JSON-LD and a starter llms.txt — you edit and ship. And the full citability picture of a page — from capsules to markup to speed — comes from the paid GEO/SEO audit.

Does the article you’re proudest of pass? Ours didn’t — probably the most useful thing the tool has told me.

→ Check your article’s capsules

FAQ

What is an answer capsule?

It's the first paragraph of a section (right after an H2/H3) that fully answers the heading's question without leaning on surrounding text, in ~40–75 words. AI engines' RAG pipelines split a page into chunks by heading and cite exactly these self-contained fragments.

Why 40–75 words specifically?

Under ~40 words the answer is usually incomplete and the engine has to pull context from elsewhere. Over ~75, the fragment stops fitting into a quotable block whole, and the answer gets buried. The range is empirical — based on which paragraphs AI Overviews and Perplexity actually lift.

Won't capsules hurt readability for humans?

No — it's the old journalistic "inverted pyramid" principle, answer first, details after. Humans scan the same way AI does. Lists and tables stay; the capsule just goes in front of them as one paragraph.

Was this helpful?
Full Check · $129 Launch SEO/GEO with seo·matrix → →

Comments

  • …

Comments are moderated.