What it means
Chunking is how a page gets split into smaller passages before it is embedded and stored for retrieval. In AI search, the chunk is the retrieval unit, not the page. A retriever pulls individual chunks that match a query, so the model sees fragments of your content rather than the whole article. How your page breaks into chunks decides what can actually be retrieved from it.
Why it matters
A long pillar page does not compete as one document; it competes as many chunks, and each one is judged on its own. That cuts both ways. A tightly written passage can get cited even if the rest of the page is mediocre, and a vague or off-topic paragraph can be the chunk that gets retrieved and drags down how the model reads you. This is the logic behind Passage Ranking, where Google scores different parts of a page independently. Headings matter here because they give chunks their boundaries and tell the system what each passage is about.
Imagine a 4,000-word guide from an HR-tech brand, Onboardly, on remote onboarding. The section on "equipment shipping" is sharp and specific, so it gets chunked cleanly and cited for shipping queries. But the intro rambles across 5 loosely related topics, so it embeds as a muddy chunk that surfaces for the wrong prompts and never earns a citation. Same page, opposite outcomes, decided at the chunk level.
How to use this knowledge
Write self-contained sections. Make each section answer one question fully so it stands on its own as a retrievable chunk.
Use descriptive headings. Clear H-tags give chunks clean boundaries and signal the topic of the passage beneath them.
Fix your weakest passages. Audit thin or rambling paragraphs, since the weakest chunk can be the one that gets retrieved and misrepresents the page.
Front-load facts in each chunk. Put the key claim or number early in a section so the passage is liftable on its own.
Growth Memo guidance
It's just us getting better at more granularly understanding the content of a page, and being able to score different parts of a page independently.
LLMs extract and cite specific passages. Clear structure makes pages easier to parse and excerpt.
By better understanding the relevancy of specific passages, not just the overall page, we can find that needle-in-a-haystack information you're looking for.
Passage Ranking — Google scoring parts of a page independently, the search-side analog of chunk-level retrieval.
Embeddings — the meaning vectors each chunk is turned into before storage.
Vector database — the index that stores chunks as vectors for retrieval.
Retriever — the component that pulls individual chunks, not whole pages.
Content structure — the semantic HTML and formatting that determine how cleanly a page chunks.