AI / GEO
內容分塊可提取性分析工具
AI answer engine同RAG系統唔會讀晒成頁——佢哋每次淨係擰一個chunk出嚟單獨顯示。呢個工具用同一個方法拆你嘅內容，同埋幫每個chunk評分，睇吓冇周邊上下文嘅時候仲講唔講得通。
These heuristics are tuned for English text; word counts and structure detection still work on other Latin-script languages, but the pronoun and phrase checks are English-specific.
What this tool checks
AI Overviews, ChatGPT, Perplexity and RAG-based search all work the same way underneath: they split a page into passages ("chunks"), store each one separately, and retrieve only the one or two chunks that answer a query — the rest of the page is never shown to the model. A chunk that relies on something said earlier on the page, opens mid-thought, or leans on an undefined acronym will confuse or get dropped by that retrieval step, even if the full page reads perfectly. This tool rebuilds that chunking step and grades each resulting passage for self-containment.
How chunks are built and scored
Nothing is guessed from a template — every chunk comes from your own pasted text, split the way a heading-aware ingestion pipeline would split it (roughly 150-240 words per chunk, breaking at headings first).
Heading-anchored chunking
Paragraphs are grouped under their nearest heading into passages of roughly 150-240 words each. A single paragraph longer than that is flagged on its own, since an automatic chunker has no natural break point inside it.
Self-containment checks
Each chunk's opening sentence is checked for a dangling pronoun ("This", "It", "They"...) with no antecedent inside that same chunk, and its full text is scanned for backward-looking phrases like "as mentioned above" that only make sense next to a chunk the reader will never see.
Cross-chunk acronym tracking
Every acronym or abbreviation is tracked across the whole document. If it is spelled out in one chunk but reused unexplained in another, that second chunk is flagged — a retrieval system will show it alone, without the definition.
Length and mid-thought cutoffs
Chunks under 40 words are usually too thin to carry a complete idea; chunks that end on a colon or a conjunction ("...and", "...because") read as cut off mid-sentence to anyone shown only that passage.
常見問題
呢個同易讀性或者關鍵字密度檢查工具係咪一樣㗎？
唔一樣。易讀性係量度人類讀得舒唔舒服；呢個工具係量度一段內容單獨顯示、冇成頁圍住嘅時候，仲講唔講得通——呢個係另一個問題，因為AI answer engine會擰出獨立嘅chunk嚟顯示先變得重要。
What counts as a "citation-ready" chunk?
A chunk scores 100 minus penalties for each issue found (a severe issue like a dangling pronoun costs more than a minor one like running long). It is marked citation-ready once it scores 80 or higher with no severe issues.
My content reads fine to a human — why are chunks flagged?
A full page reads fine because a human keeps context across paragraphs automatically. A retrieval system does not: it stores and shows one chunk in isolation, so anything that depends on an earlier sentence, an earlier chunk, or an earlier definition can fail silently even though the page itself is well written.
Does this tool send my text anywhere?
Your pasted text is sent once, over HTTPS, to our server to run the chunking and scoring (needed to keep the logic server-side and consistent), and is not stored beyond the request.