Chunking is the practice of splitting a long piece of content into smaller, self-contained passages. The word is used in two connected ways: AI search systems chunk web pages so they can store and retrieve individual passages, and writers chunk their content so each section makes sense when read, or quoted, on its own.
How chunking works
AI answer tools such as ChatGPT search, Perplexity and Google’s AI features do not usually feed a whole website into a model at once. A common approach, called retrieval-augmented generation, breaks each page into chunks of a few hundred words or less, converts each chunk into an embedding (a numerical representation of its meaning), and stores them. When someone asks a question, the system finds the chunks whose meaning is closest to the question and hands those passages to the model to write the answer.
How a system draws the boundaries varies. Some split at a fixed number of words, some at headings or paragraphs, and some overlap neighbouring chunks so context is not lost. The exact methods used by each AI product are not published, so nobody outside those companies can say precisely where your page will be cut. What is clear is that a passage which only makes sense alongside the paragraph before it is a weaker candidate once it has been separated.
Google has worked at passage level for traditional search too. Its passage ranking lets a single relevant section of a long page rank for a specific question, even if the rest of the page covers other things.
Why it matters
If a page is retrieved and cited in pieces, each piece has to carry its own meaning. Consider a London accountant’s page about Making Tax Digital. A section that opens “As mentioned above, this applies from the date we discussed” is close to useless when lifted out. A section that opens “Making Tax Digital for Income Tax applies to sole traders and landlords with qualifying income above the threshold set by HMRC” can be quoted and understood with no other context.
Well-chunked content is also easier for people. Most visitors scan, jump to the heading that matches their question and read that part. Writing for retrieval and writing for a busy reader turn out to be largely the same discipline.
Common mistakes
- Leaning on earlier context. Pronouns and phrases like “this”, “the above” or “as we said” lose their meaning when a passage stands alone.
- Vague headings. “Overview” or “Things to consider” tell neither a reader nor a retrieval system what the section answers.
- Chopping content into fragments. Chunking does not mean one-sentence sections. A passage still needs enough substance to answer something properly.
- Burying the answer. A section that spends three paragraphs on background before stating its point is less likely to be the passage selected.
- Treating it as a trick. Restructuring thin content into neat sections does not make it worth citing.
How to act on it
Take one of your most important pages and read each section in isolation, as if it were the only text someone would see. Ask whether it names its subject, answers one clear question and makes sense without the rest of the page. Rewrite the openings that fail, so the first sentence of each section states the point plainly and names the thing it is about.
Use descriptive heading tags phrased the way people ask, keep one idea per section, and put facts such as prices, dates and eligibility rules in full sentences rather than relying on a table elsewhere on the page. This is a large part of how I approach generative engine optimisation for UK businesses: making each passage of a page clear enough to be found, understood and quoted.