What is chunking in AI search?
Chunking is how retrieval-based AI systems split a page into small passages, then fetch only the few that match a question and write the answer from those. A chunk that names its subject and states its fact in one place is quotable; a chunk that says "as mentioned above" without saying what "it" is gets skipped or misquoted.
The short version
Search-backed AI assistants do not read a page the way a person does. They split it into small passages — chunks — and retrieve only the few that match the question. The answer you see was written mostly from those passages.
So the real question for anyone publishing content is not "is my page good?" but "does each passage of my page still make sense when an AI pulls it out alone?"
Why AI search chunks at all
Every AI answer has a budget: context window, retrieval cost, and relevance. Passing a whole page for every question wastes all three.
Instead, retrieval systems do this:
- Split the page into chunks (by headings, paragraphs, tokens, or meaning).
- Match the question against the chunks.
- Retrieve the best few chunks.
- Answer from those chunks, ideally citing them.
That is why a 2,000-word page can be represented in an answer by a single sentence — the sentence that came from the retrieved chunk.
How content is usually split
Real systems differ, but the common approaches are:
- Heading-based: each section becomes its own chunk, sometimes with the heading repeated as context.
- Fixed-size: the text is cut every N tokens or words, often with overlap.
- Semantic: the text is split where meaning shifts, using an embedding model.
Prechunk uses a simple, common approach — split at headings, then at paragraph boundaries past a size you choose — because the goal is to show weak spots, not to copy any one vendor (how the checks work).
Why chunking decides what gets quoted
A retrieval system can only quote what a chunk contains. Three failure patterns show up constantly:
- Dangling references. A chunk says "as mentioned above, it costs $89" — the system did not retrieve the "above" part, so it skips the fact or misquotes it.
- Subjectless openings. A chunk starts mid-thought ("Because of that, we recommend…") without naming the thing it is about.
- Split facts. The price is in one chunk and the service name in another; the answer may repeat only half.
NRLC.ai's studies of 395 AI answers found the passages that actually got quoted were self-contained: a published price, a rating, a service in a named place, a fair comparison (what AI assistants quote).
How to make your chunks quotable
- Name the subject in every section. Do not rely on the title three screens up.
- Open each section with a direct statement someone could quote without editing.
- Keep the fact with its context. "Furnace repair in Pasadena: $189–$340" beats "$189–$340" in one place and "furnace repair" in another.
- Avoid "as mentioned above" and "this means" across chunk boundaries.
- Check it, don't assume it. Paste the page into Prechunk, and read the chunks the way an AI would: alone.
What chunking is not
Chunking is not a ranking trick. Changing chunks does not make a page more truthful, more authoritative, or more relevant — it makes whatever the page already says survive being cut into pieces. A bad page in good chunks is still a bad page; a good page in bad chunks gets misquoted.
Try it now: paste your page into the free chunking checker and see which chunks would confuse an AI assistant.
Frequently asked questions
- What is chunking in AI search?
- Chunking is the step where a retrieval-based AI system splits a page into small passages, or chunks, so it can fetch only the passages that match a question instead of the whole page.
- Why do AI assistants chunk content?
- Passing an entire page into a model for every question is expensive and dilutes the answer with irrelevant text. Chunks let the system retrieve just the relevant parts and answer from those.
- What makes a chunk easy for AI to quote?
- It names its subject, opens with a direct statement, includes the specific fact (price, rating, service and place), and does not depend on text outside the chunk.
- How can I check my own content's chunks?
- Use the free Prechunk tool. Paste your page, pick a chunk size, and it shows how the content splits and flags the chunks that will not make sense on their own.