Semantic chunking
Semantic chunking splits a page into pieces where the subject changes, so an AI system can read each piece by itself.
Semantic chunking is splitting a page into pieces where the subject changes, so each piece holds one idea and makes sense by itself. AI systems that search the web often break pages into chunks like this, then pick the chunks that best match a question.
For you as a writer, it means the section is the unit that gets found and quoted. A system struggles to use a section that only makes sense after reading the one before it.
How it works
A system could cut a page every 200 words, wherever that falls. That's quick, but it can split an answer in half, or glue two unrelated ideas together. Semantic chunking looks at the meaning instead. It keeps sentences together while they're about the same thing, and starts a new chunk when the subject moves on.
Your page's structure helps a lot. Headings, short paragraphs and clear topic sentences give the system obvious places to cut.
Writing for it
Write each section so it could be lifted out and still be understood:
- Give each section a heading that says what it answers
- Answer the question in the section's first sentence or two
- Name the thing you're talking about, instead of "it" or "this," at the start of a section
- Keep one subject per section, and start a new one when the subject changes
For example, Harrow & Pine's brewing guide has a section headed "Temperature" that starts "As mentioned above, it should be cooler." Lifted out, that tells nobody anything. Rewritten as "Brew green tea at 70 to 80°C, cooler than black tea, or it turns bitter," the section can answer a question on its own.
In AnswerOrbit
Page checks in Recommendations looks at whether a page has a title and headings, whether the prompt appears in a heading, and whether its opening answers the prompt. It runs the same checks on pages the platforms cite for that prompt, so you can compare.