Word & Character Counter
Words, characters, sentences and reading time, live as you type.
Text & Writing
Text Summarizer
Extractive summaries that select real sentences, never invent them.
What it does. This is an <em>extractive</em> summariser: it scores every sentence by how much of the document’s distinctive vocabulary it carries, then returns the highest-scoring ones in their original order. Every word in the summary appears in your text. It cannot invent a fact, because it cannot invent a sentence.
Runs in your browserNothing uploadsNo signupWorks offline
How to use Text Summarizer
- Paste the text.
- Choose how much to keep — a percentage or a sentence count.
- Read the summary and the sentences it ranked highest.
Extractive and abstractive
There are two kinds of summariser and the difference matters more than the output length.
An abstractive summariser writes new sentences in its own words. That reads better, and it can state things your source never said — the failure mode people call hallucination. It needs a language model.
An extractive summariser selects sentences that already exist. It reads more choppily and it cannot fabricate, because every sentence is quoted from your input.
For summarising something you must not misrepresent — a legal document, a medical paper, meeting minutes — extractive is the safer instrument, and the choppiness is a fair price.
How the scoring works
Words are weighted by how distinctive they are to your document: a term that appears often here but is common in ordinary English tells you little, while one that appears often here and rarely elsewhere is probably the subject.
Each sentence scores as the sum of its word weights, normalised by length so a long sentence does not win by being long. Sentence position gets a small bonus, because the first sentence of a paragraph tends to carry its claim.
This is the classical TF-IDF approach and it is decades old. It has one large advantage over anything newer: you can inspect exactly why a sentence was chosen, which the tool shows you.
When it works and when it does not
| Text | Result |
|---|---|
| News article | Good — inverted pyramid suits extraction |
| Academic abstract | Good — already dense |
| Long report | Good, with a low keep percentage |
| Narrative or fiction | Poor — meaning is in sequence, not sentences |
| Dialogue or transcript | Poor — turns lose their context |
| Bullet lists | Poor — already compressed |
The pattern is that extraction works on text that argues and fails on text that unfolds. If the point of the piece depends on the order events are revealed, taking sentences out of it destroys the thing you wanted to keep.
Frequently asked questions
Does this use AI?
No. It uses TF-IDF sentence scoring, a statistical method that predates modern language models. That is a deliberate choice: it cannot hallucinate, because it only ever returns sentences from your own text.
Why does the summary read choppily?
Because it is extractive — real sentences lifted from the source and placed in order. An abstractive summariser rewrites them into smoother prose, but can also state things your source never said.
How much should I keep?
20–30% for a long report, 40% for an article. Below about 15% you usually lose the connective sentences and the result stops making sense on its own.
Is my text uploaded?
No. The analysis runs entirely in your browser, which matters when the document is a contract, a draft or anything unpublished.
Why did it pick a sentence that looks unimportant?
Usually because that sentence contains several terms that are distinctive to your document. The tool shows each sentence’s score so you can see the reason and adjust the length.