How Prechunk checks your content
Prechunk splits content at headings, then at paragraph boundaries once a chunk reaches the size you choose. It then checks whether each chunk names its subject, opens with a direct statement, avoids references to text outside the chunk, and is a usable length.
Why chunks matter
Search-backed AI assistants retrieve passages rather than whole pages, then write an answer from what they retrieved. If the passage that matches a question says "As mentioned above, it starts at $89" without saying what "it" is, the assistant either skips it or has to guess. A chunk that stands on its own is much easier to quote correctly.
Real systems differ. Some split by tokens, some by headings, some overlap chunks. Prechunk uses a simple, common approach so you can see the weak spots; it isn't a copy of any one vendor's system.
How content is split
- Every heading (
#to####, or HTMLh1toh4) starts a new section. - Each section is cut into chunks at paragraph breaks once it passes the size you chose (150, 250 or 400 words).
- The heading is kept with each chunk, because many systems carry it along as context.
The checks
| Check | What it looks for | Why |
|---|---|---|
| Names the subject | The main subject you entered appears in the chunk or its heading | A chunk that never names the business or topic can't be attributed to it |
| Dangling reference | Opens with words like "This", "It", "They", "As mentioned", "See above" | Points at text the retriever didn't fetch |
| Direct opening | First sentence is 30 words or fewer | Short, direct first sentences are easier to quote as an answer |
| Has a heading | The chunk sits under a heading | Headings phrased as questions match how people ask |
| Specifics | Contains numbers, prices, dates or places | Concrete facts are what answers quote |
| Length | Under 40 words is flagged as thin | Very short chunks rarely carry a complete answer |
Limits
These are heuristics. A chunk can pass every check and still be unhelpful, or fail one for a good reason. Use the flags as prompts to reread, not as a score to chase.
Frequently asked questions
- What is content chunking in AI search?
- Retrieval-based AI systems split pages into smaller passages, called chunks, and retrieve only the few that match a question. The answer is written from those chunks.
- What makes a chunk easy for AI to quote?
- It names its subject, opens with a direct statement, includes specific facts, and doesn't depend on text outside the chunk.
- Does Prechunk send my content anywhere?
- No. The analysis runs entirely in your browser.