What Is Passage Extraction?

In one sentence

Passage extraction is when a retrieval system selects a specific chunk of your page rather than the whole document — which means individual sections have to stand alone.

What Passage Extraction means

Retrieval systems split pages into chunks and evaluate each independently. The chunk that best matches the query gets passed to the model, not the page as a whole.

So the unit of competition isn't the page. It's the passage — and a brilliant page whose best answer depends on three paragraphs of earlier context loses to a mediocre page with one self-contained section.

Why it matters for AI search

This is why answer-first structure matters mechanically rather than stylistically. Chunking happens whether or not you designed for it.

It also means a single page can win multiple different queries through different sections, if each section is genuinely self-contained.

See it in action

The same page, chunked. Watch which passages survive.

Chunk-level evaluation

Retrieval simulation
user asks → how much does AI SEO cost?

Chunk 1 — intro

Sets up the topic. No answer present. Not a candidate.

not cited

Chunk 2 — pricing section

States ranges directly in the first sentence. Self-contained.

not cited

Chunk 3 — case study

Mentions a figure but needs the client context above it to make sense.

not cited

One chunk of three was usable.

Chunk 3 contained relevant information and was still dropped, because quoting it without the preceding paragraph would misrepresent it.

How to get it right

Writing for passage extraction

  • Treat every H2 section as a standalone answer to a specific question
  • State the answer in the first sentence of the section, then support it
  • Avoid pronouns and references that depend on earlier sections
  • Repeat the key noun rather than saying 'it' or 'this approach'
  • Test by reading each section in isolation — if it doesn't make sense, a model won't use it

Common questions

How large is a chunk?

It varies by system, typically a few hundred words or a logical section. You don't control the boundaries.

Which is why designing sections that happen to be self-contained is the practical approach.

Can one page win several queries?

Yes, and that's the upside of good structure. Different sections can be extracted for different questions.

Pages with clear question-shaped headings tend to do this well.

Does this conflict with narrative writing?

Somewhat, and it's a real tradeoff. Content that builds an argument across sections reads well and extracts badly.

The compromise is usually to lead each section with its conclusion, then develop it.

These come up alongside Passage Extraction constantly.

Free 12-point check

Send Ron your site

Two fields. We’ll crawl your site as GPTBot, ClaudeBot and PerplexityBot, benchmark a sample of your category’s prompts, and send you what we find.

  • What each AI crawler actually receives from your site
  • Whether models resolve your brand as a real entity
  • A sample of category prompts and who gets cited
  • The three fixes we’d prioritise first

No commitment, no sales sequence. If we’re not a fit, we’ll say so.

We’ll pull your favicon so you know we found the right site.

We reply from a real address. No sequence, no newsletter.