Free tool No account Experimental

Vectorscope

Semantic chunking, made visible: see what survives the trip into vector space.

Vectorscope is a free semantic chunking visualizer: it cuts one URL into the passages a retrieval system would store, groups them by meaning, and shows how much of the page never reaches the index at all.

e.g.

A free allowance per visitor, no account needed.

Result

what the machine actually readscut at h2, then 2,000-char windowsmax 26 chunks / pagegrouped by meaningno account

1. Fetched
2. Extracted
3. Chunked
4. Embedded
5. Grouped

passages read
slots used of the 26-passage cap
unique text in passages, chars
in passages incl. overlap and intro, chars
characters past the cap

Paste a URL above. The page is cut into passages at its h2 headings; the ones inside the cap are read, and anything the cap cuts off is not.

What does this mean?

The reading shows which passages of your page a retriever can receive, and how they group by meaning — the question semantic chunking asks, put to passages already cut. It is a preview of the machine's view, not a guarantee of what any assistant does with it.

Important passage outside the cap?

Move it earlier on the page, or split the page so that passage has a budget of its own. This reading takes at most 26 passages from a page.

Models and retrieval caps vary.

This does not establish that any assistant has indexed the page. Each system does its own chunking — fixed windows, headings or semantic chunking — and embeds and caps in its own way; this is the pipeline the product runs. Sources and methodology

How to read your result

Four readings decide whether a page is legible to a retrieval system. Here is what each one means when it is bad.

Chunks that never reach the index

A retrieval system reads a fixed number of passages from a page and ignores the rest, so anything below that cut is invisible no matter how good it is. Vectorscope applies the same cap the product applies and names the passages that fell outside it.

If your answer, your differentiator or your FAQ sits below the cut, the fix is not more words, and semantic chunking will not rescue it either: a well-cut passage past the cap is still a passage nobody reads. Move that passage up, or split the page so it has a passage budget of its own.

The extraction ratio

The extraction ratio is how much of the page turned into readable text — a low one means most of what you shipped was navigation, boilerplate or markup that no model will ever see. A long page with a low ratio is a short page as far as retrieval is concerned.

Extraction happens before any chunking, semantic chunking included, so text lost here is lost to every strategy. Common causes, in the order we see them: text rendered only by JavaScript, an article wrapped in so much chrome that the body is a minority of the document, and images carrying words that exist nowhere in the HTML.

Groups, and what a tight group means

Groups are the semantic part of this reading, the nearest it comes to semantic chunking: passages the embedding places close together in meaning, whatever heading they sit under. One enormous group usually means the page says one thing many times, which reads as thin to a retriever even when it reads as thorough to an editor.

Several small groups with one passage each is the opposite problem: a page that changes subject constantly gives a retriever nothing complete to quote.

The passage a question actually retrieves

Asking the page a question returns the passages that answer it, ranked, using only that page — which is the closest thing to watching a model decide what your page is for. If the returned passage is not the one you would have chosen, the page is not organised the way you think it is.

This is also the fastest way to test a heading rewrite: change the heading, run the page again, ask the same question.

What is semantic chunking?

Semantic chunking is splitting a page into passages where its meaning changes, instead of every fixed number of characters or at every heading. Vectorscope does not cut that way: it cuts at each h2 and then in fixed windows, and its semantic step comes after the cut, when the passages are grouped by meaning.

Fixed-length chunking

Fixed-length chunking cuts every N characters and is the default in most pipelines because it is trivial to implement. It also cuts through the middle of a sentence, separates a heading from the answer under it, and produces passages that begin with half a word.

Vectorscope uses it only inside a long section: a window of 2,000 characters with 200 characters of overlap, ending at a sentence break where one is close enough.

Heading-aware chunking

Heading-aware chunking uses the page's own headings as the cut lines, so a passage arrives with the heading that names it. This is Vectorscope's first cut, and only an h2 makes one: an h3 stays inside the passage of the h2 above it, so a section of six h3 questions is one passage, or several windows of one.

It also means your heading text is doing retrieval work. A section called "Overview" contributes nothing to the passage; a section called "Can I cancel at any time?" contributes the question itself.

Semantic chunking, and how it differs from grouping

Semantic chunking compares neighbouring sentences and cuts where the meaning shifts, which gives passages that stand alone but needs an embedding pass over the page before a single chunk exists. Vectorscope does not do that step. Its semantic part is the grouping: every passage is embedded once it is cut, and passages that sit close in meaning are drawn as one group.

So the map asks the question semantic chunking asks — where does one idea end? — of the passages your headings already made. A group spanning three h2 sections says the headings split one idea; a long section whose windows land in two groups says a single heading holds two.

Is my content ready for AI search?

This page answers that question for one URL at a time, honestly, which means naming what it cannot see as clearly as what it can.

What this tool can tell you

How much of your page becomes readable text at all, as a ratio and not an adjective.
Exactly which passages a retriever would receive, in order, with the ones beyond the cap marked.
Which passages share a subject once they are grouped by meaning — the question semantic chunking asks, put to your own page.
Whether your passages stand alone, or lean on the passage before them to make sense.
Which passage a real question retrieves from your page, using the page's own embeddings.
Whether the page asks to be left out of an index at all — a noindex directive is reported rather than ignored.

What it cannot tell you

Whether any specific assistant has your page in its index. Nobody outside those companies can see that.
How a model ranks your page against a competitor's. That needs a corpus of both, which is the product rather than the demo.
Whether you are cited in an answer today. That is a measurement over time on prompts you choose, not a property of one page.

Using Vectorscope to test a RAG pipeline

If you build retrieval rather than content, this page is a reference implementation you can point at any URL in about twenty seconds.

1Run a page you already index, and compare the passage boundaries with your own chunker's — fixed windows, headings or semantic chunking.
2Note the extraction ratio: if yours is lower on the same URL, your extractor is discarding body text, not boilerplate.
3Ask the page a question you know the answer to, and check whether the top passage is the one that contains it.
4Run a page that is far too long. The passages beyond the cap are the ones your own top-k will also never reach.

The chunker (h2 cuts, then 2,000-character windows with 200 of overlap), the embedding model and the cap here are the ones the product runs in production. A demo using a simplified pipeline would flatter the pipeline it exists to demonstrate.

Who is Vectorscope for?

Vectorscope is for anyone whose page has to be read in passages: the people who write it, and the people who build the retrieval that reads it. Both get the same reading — the cut, the cap, and the semantic chunking question of which passages belong together.

Content and SEO teams

For checking that the answer a page exists to give survives into one passage a retriever can quote, above the cap, under a heading that names it.

Engineers building RAG

For a reference reading of any URL to hold your own chunker against, whether it cuts on fixed windows, headings or semantic chunking.

Editors restructuring a page

For deciding where a heading belongs: the groups show which sections share a subject and which section is really two.

Once the passages are right, the words inside them are next: PatternScope reads a draft for the stock phrases and repeated sentence shapes a reader reacts to.

Words used on this page

The terms this page relies on, each defined once, because a page that uses them without defining them is a page an assistant cannot explain to anybody.

Semantic chunking: Cutting text into passages where its meaning changes, rather than at a fixed length or at a heading. On this page the meaning-based step is the grouping of passages that are already cut, not the cut itself.
Chunk: A passage of a page, cut so it can be read on its own. Retrieval systems store and return chunks, never whole pages, which is why a page's structure matters more than its length.
Embedding: A list of numbers that stands for the meaning of a passage, so that two passages about the same thing sit near each other. Comparing embeddings is how a retriever finds a passage without matching words.
Cosine similarity: A measure of how close two embeddings point in the same direction, from 1 for identical to 0 for unrelated. It is the number behind every claim that two passages are about the same thing.
Retrieval-augmented generation: Answering a question by first retrieving passages from a source and then writing from them, rather than from the model's memory alone. Every AI answer that cites a page did this, and the page it cited was a chunk.
Extraction: Turning a fetched HTML document into the readable text a model receives, discarding navigation, scripts and boilerplate. What extraction drops, no model ever sees.

Keep reading

Longer answers to the questions this page can only summarise.

What semantic chunking is, and what it trades awayThe three strategies, side by side, on the same page.How to optimise content for AI searchWhat actually changes a passage's chances of being retrieved.MethodologyEvery formula behind the readings above, stated in full.The Vectorscope skill for Claude CodeThe same reading, run locally inside Claude Code, with sources.

Questions people ask about this

How does AI Mode choose which sources to cite?

Nobody outside Google can state the mechanism, and this page will not pretend to. What is observable is that cited pages tend to contain a passage that answers the query completely on its own, and that pages whose answer is spread across several sections are cited less often than their content deserves. Vectorscope shows you whether your page has such a passage.

How do I get cited in AI Overviews?

Give the answer its own passage, under a heading that names the question, high enough on the page to survive the retrieval cap. That is not a trick — it is the shape retrieval rewards, and you can check whether your page has it by running it here.

Is this an embedding visualization tool?

Yes, for one page at a time. It embeds every passage of a URL and groups them by meaning, so you can see which parts of the page are near each other and which stand alone. It does not plot a corpus of thousands of documents.

What is the difference between heading-aware and semantic chunking?

Heading-aware chunking cuts at the page's own headings; semantic chunking cuts where the meaning changes, whether or not there is a heading there. The first is cheap and follows the author; the second needs every sentence embedded before a single cut is made.

Does Vectorscope use semantic chunking?

Not to cut the page. Vectorscope cuts at every h2 — an h3 does not start a passage — and splits a long section into 2,000-character windows with 200 characters of overlap. The semantic step comes after: each passage is embedded and passages close in meaning are grouped, which shows where semantic chunking would disagree with your headings.

Do you store the page I test?

The reading is kept briefly so it can be shared with a link, and the fetched page itself is not retained after the reading is produced. No account is needed, so there is nothing to attach it to.

Is it really free?

Yes, with a rate limit rather than a paywall. The limit exists because every reading spends an embedding call, and it resets on a fixed window that the page tells you about before you hit it.

On this topic

Is my page ready for AI search?

Term: Chunking RAG llms.txt Structured Data robots.txt
Read: AI reads your page in pieces. Three free skills show which ones Google Search in October 2026: the page changed more than the ranking
Check: Vectorscope skill for Claude Code

Every day Crawls with the Vectorscope reading beside every page, re-checked after each fix. How the product works

Other free AI search tools

What this cannot show you

One page on its own. Whether that page answers the questions people actually ask an assistant, how it stands against a rival page on the same subject, and which topics your site has no page for at all — those need a corpus of the whole site, measured over time.

Ask for an invite