AI and retrieval-augmented-generation (RAG) systems do not read whole pages; they break content into passages and retrieve the chunks most relevant to a query. If your passages are too long, too short, or only make sense with surrounding context, they retrieve poorly and are less likely to be cited. This tool shows how your page would split and how self-contained each chunk is.

It parses the HTML and segments content along headings and paragraph boundaries, grouping text under each heading into discrete passages — much like how an indexer or RAG pipeline would chunk it. Each resulting passage is then scored individually, so you can see exactly which sections are retrieval-friendly and which need work.

Each passage is scored with rule-based heuristics: length (enough words to be meaningful but not bloated), presence of context (a heading or topic anchor so it stands alone), and standalone readability (sentence length and whether it opens with dangling references like 'this' or 'as mentioned above'). The result is a per-chunk score plus an overall page score.

A self-contained passage answers or explains one idea without requiring the reader to have seen earlier text. It has its own topic anchor, avoids opening with unresolved pronouns ('it', 'this', 'they'), and packs its key claim into the first sentence or two. These passages are far easier for AI systems to lift and attribute correctly.

Roughly 40 to 120 words is a good target for most informational content. Shorter chunks often lack enough substance to answer a query, while very long passages mix multiple ideas and dilute relevance. The analyzer flags passages that fall well outside this range so you can split or expand them.

No. The page is fetched and parsed on our server only to extract and score passages with deterministic heuristics — no LLM or external AI API is called. Your content is not stored or used for training; the tool simply returns the chunk analysis.

Add a clear heading above context-light passages, rewrite opening sentences so they do not depend on previous text, split long mixed passages into focused ones, and merge fragments that are too thin to stand alone. Re-run the analyzer to confirm each chunk now reads as a standalone, retrievable answer.