Text Summarizer

You paste a text or upload a document, choose a format and a length, and the summary appears next to your original. In the key-points and study-notes formats, the sentences the summary was built from are highlighted in your text, so you can see at a glance what was kept — and, more importantly, what was left out.

The summarizer extracts; it does not comment. It is instructed never to add a fact, a figure, a date or an example that is not in the text, never to introduce the summary with 'This text is about', never to give an opinion. It keeps the main information and the order of ideas, and prefers omitting a nuance to rendering it halfway. A summary that says less than the original but nothing the original did not say is the goal.

Two hundred characters and three sentences are the minimum — below that there is nothing to summarize. Up to twelve thousand characters, about two thousand words, are handled in one pass, which covers most articles, reports and chapters.

• paste a text or upload a PDF, Word or text document
• five formats, four lengths, result in seconds
• the sentences kept are highlighted in your original
• nothing is added that the text did not say

Paragraph is a written summary in one piece: complete sentences, no bullets, readable as a text. It is the format for a quick read of an article, a report or an email thread. Key points is a list of bullets, one idea each, with the source sentences highlighted in your text. It is the format for reviewing a document you will need to discuss or act on.

Study notes is the revision format: short headings, telegraphic bullets under each, and a closing 'Remember' line that fits in one sentence. It is built for students revising a chapter or a set of lecture notes. Executive brief is the professional format: three sections titled Context, Findings and Recommendations, each in short bullets, with the rule that the recommendations must come from the text — if it makes none, the brief says so.

Simplified is a written summary in plain language for a reader of about fourteen: short sentences, one idea each, everyday vocabulary, technical terms explained. It is the format for turning a specialist text into something anyone can follow.

• Paragraph: a readable summary in one piece
• Key points: bullets, with the source sentences highlighted
• Study notes: headings, telegraphic bullets, a 'Remember' line
• Executive brief: Context, Findings, Recommendations
• Simplified: plain language, technical terms explained

The length slider has four steps — very short, short, medium, detailed — and it is graduated in a way you can check: for written formats the target is a share of the source's word count, for list formats a share of its sentence count. 'Short' on a thousand-word article gives a summary of roughly a hundred and fifty words; 'detailed' about four hundred.

The tool enforces the length rather than merely requesting it. Language models are poor at hitting a length target — measured in published tests, a large model misses by more than the tolerance about seventy percent of the time. So the summarizer counts what was actually produced, and if it is more than thirty percent off, it asks once more, telling the model exactly how far off it was and whether to cut ideas or add ones from the text. A single retry fed with the measured gap brings the miss rate close to zero.

In practice this means the length you choose is the length you get, within a reasonable margin, and the word counts shown above the summary let you verify it.

The dominant flaw of an automatic summary is not inventing — that is rare, and the tool is instructed against it — but omitting. A summary leaves things out by definition, and the thing left out can be the one that mattered to you. The only way to spot that is to see what was kept.

In the key-points and study-notes formats, each bullet is tied to the source sentences it came from, and those sentences are highlighted in your text on the left. Read the unhighlighted parts: that is what the summary decided you could do without. If an important sentence is among them, choose a longer setting or a different format.

The highlighting is exact by construction. The text is split into numbered sentences before it reaches the model, and the model returns numbers, not positions. A number that does not exist is discarded, never approximated. What is highlighted is what was used.

Upload a PDF, a Word (.docx) or a text file of up to eight megabytes and the text is extracted and summarized as if you had pasted it. A research paper, a report received by email, a downloaded article or a chapter from a course pack can be summarized in two clicks, in whichever format suits the reading you need to do.

Scanned PDFs are the exception: a photographed page saved as PDF holds an image, and nothing can be extracted from it. Open the document and paste the text if you can select it. Documents over twelve thousand characters are cut at that point, and the result says so; a long report is better summarized section by section anyway, since a single summary of forty pages loses too much.

Paste a chapter or your lecture notes, choose Study notes, and you get headings, bullets under each and a one-line 'Remember'. Then read your original with the highlights on: the unhighlighted parts are what the notes skipped, and a quick check tells you whether anything you need for the exam is among them.

Study notes work best on texts of a few pages at a time — one chapter, one lecture. The format has its own floor (two headings of two bullets at least) and a cap of about twenty words per bullet, so the notes stay telegraphic instead of turning into a second text nearly as long as the first, which is what early versions of the tool produced.

A summary is a starting point for revision, not a substitute for it. Reading the notes and then re-reading the highlighted sentences in context is a far better use of the tool than reading the notes alone.

The Executive brief format turns a report, a set of meeting notes or a long email thread into three sections — Context, Findings, Recommendations — that can be pasted into a message or a slide. The rule that recommendations must come from the text matters here: the tool will not invent next steps, and if the document proposes none it says so rather than filling the section.

For reading rather than forwarding, Paragraph at 'short' gives a one-glance version of anything that lands in your inbox. Key points at 'medium' gives a checklist to walk through in a meeting, with the highlights showing where each item came from if someone asks.

Paste the abstract and introduction of a paper, or the full text of an article, and Key points at 'medium' gives you the claims and the evidence in order. The highlights show which sentences carried them. Reading the highlighted sentences in the original is often enough to decide whether the full paper deserves your time.

Figures and dates are copied exactly, and negations are preserved — the instructions single those out because a summary that turns 'no effect was found' into 'an effect was found' is worse than no summary. Still, read the summary as a summary: it is a map of the text, not the text.

The summary is written in the language of the text. Paste a Spanish article and the summary is in Spanish; a German report, in German; a French chapter, in French — including the fixed labels of the study-notes and brief formats, which are translated into that language. The interface stays in the language of this page.

Sentence splitting, which the highlighting depends on, uses the language-aware rules built into the browser engine plus a list of abbreviations for the languages served, so 'Dr. Smith' or 'z.B.' does not break a sentence in two. Quality is highest in English and the major European languages, where the underlying model has the most material.

A summary leaves things out; that is its purpose. Before using one for anything that matters, check three things. First, the figures, dates and names: they should be identical to the original. Second, the negations: 'no', 'not', 'never', 'without' — a summary that loses one reverses a claim. Third, in the formats with highlights, the unhighlighted parts of your text: is there a sentence there you needed?

The word counts above the summary tell you how much was removed. A reduction of eighty-five percent is a very different object from a reduction of forty, and knowing which you are reading helps you trust it appropriately.

• figures, dates and names must be identical
• negations must survive: a lost 'not' reverses a claim
• read the unhighlighted parts for anything you needed
• the word counts show how much was removed

Ten summaries a day without an account, each on up to twelve thousand characters. That covers a student's day of revision, a professional's inbox and a reader's stack of articles. No card, no email, no sign-up.

A free account multiplies both limits by four. The counter runs over a rolling twenty-four hours and is applied by account and by network. There is no paid tier for the summarizer: the limits keep the service fast for everyone.

The text is sent to the server, summarized, and the result is returned. It is not stored, not added to any dataset, and not used to train anything. Uploaded documents are read in memory and discarded; the file is never written to disk. Only a usage counter is kept for the daily limits.

The summary is produced by a language model hosted under a business agreement that excludes training on submitted text. If you summarize confidential material — an internal report, a draft, a private thread — this is the policy that applies.

The summarizer shortens a text by keeping what matters and shows you what it kept. The paraphrasing tool rewrites a text in different words at the same length, or shorter or longer, but keeps everything. The document analyzer on this site answers questions about a document in conversation, which is the right tool when what you need is not a summary but a specific answer buried in forty pages.

So: a long text to read quickly — summarizer. A text to say differently — paraphraser. A question about a document — analyzer. And a text to translate — the AI translator, which handles documents too.

The summarizer works on prose. Tables, spreadsheets and code have no sentences to extract and will produce poor summaries; the document analyzer handles those better. Texts under two hundred characters or three sentences are refused — there is nothing to summarize. Texts over twelve thousand characters are cut, and the result says so.

A summary is only as good as the text: a rambling document produces a rambling summary, and a text that buries its point produces a summary that may miss it. When a summary seems off, the highlights will usually show why — the tool kept the sentences that looked important, and the text did not put its importance where a reader would expect.

• prose only: tables, spreadsheets and code have no sentences
• two hundred characters and three sentences minimum
• twelve thousand characters maximum per summary

Take a six-hundred-word news article about a city council voting to convert a downtown car park into housing: the vote count, the number of homes, the objections from local businesses, the timeline, a quote from the mayor, a quote from an opponent, and a paragraph of background on the housing shortage.

Paragraph, short gives about ninety words: the decision, the numbers, the main objection, the timeline. Key points, medium gives six bullets — the vote, the homes, the businesses' objection, the mayor's position, the opponent's, the timeline — with six sentences highlighted in the article. Reading the unhighlighted part, you see that the background paragraph on the housing shortage was dropped: important context, and a reason to choose 'detailed' if you are writing about the story yourself.

Executive brief gives Context (the shortage, the car park), Findings (the vote, the numbers, the objections) and Recommendations — and here the tool writes that the article proposes none, because a news article does not. Simplified gives a plain-language paragraph a fourteen-year-old can follow, with 'rezoning' explained in five words. Same article, four readings, each the right one for a different purpose.

Meeting notes and long email threads are the texts people most often need summarized at work and least often get right, because they are not written to be read — they accumulate. Paste the notes and choose Key points at 'medium': you get one bullet per decision or action, and the highlights show which line of the notes each came from, which matters when someone asks 'who said that'.

For an email thread, paste the messages in order, oldest first, and choose Paragraph at 'short' for a briefing or Executive brief to forward. The instruction to keep the order of ideas means the summary follows the thread's chronology, and the rule that recommendations must come from the text means the brief will not invent a next step nobody agreed to.

Remove signatures, disclaimers and quoted replies before pasting: they are not sentences worth summarizing and they use up the character limit.

A chapter of a textbook or a set of lecture notes is longer than twelve thousand characters more often than not, so the first step is to split it: by section, by lecture, by heading. Summarize each part with Study notes at 'medium', and you get a set of notes that follows the structure of the course. Paste the 'Remember' lines of each part together, and you have a one-page overview of the whole.

Then use the highlights. For each part, read the unhighlighted sentences once: they are what the notes skipped, and a fast check tells you whether an example, a definition or an exception you need for the exam is among them. The summary is the map; the highlighted sentences are the territory you should know by heart; the unhighlighted ones are what you decide you can skip — with your eyes open.

The text is split into sentences before it reaches the model, and each sentence is numbered. The model is asked for a summary in the requested format and length, and for the numbers of the sentences it drew on. It is told to keep the main information and the order of ideas, to prefer omitting an idea to rendering it halfway, and never to add anything. It is not told what is important — that judgement is the model's, and it is generally good: a sentence with a figure, a decision or a claim ranks above one with a transition or a restatement.

The length is then checked against what was produced, and corrected once if it is off. The sentence numbers are checked against the list, and any that do not exist are discarded. Only then is the result shown. Every step that can be verified mechanically is verified mechanically; the model is trusted with the one thing only a reader can do, which is to say what matters.

A summary is a shorter version of a text that keeps its main content and order — the Paragraph format here. An abstract is a summary written by the author of a paper, before the paper, following a fixed structure; the summarizer can produce something close to it from the paper's text, but a real abstract is written, not extracted. Key points are the ideas of a text as a list, without the connecting prose. A brief or executive summary is a summary organised for a decision — what is the situation, what was found, what should be done.

Study notes are a summary organised for memory: headings, telegraphic bullets, one line to remember. Simplified is not shorter for its own sake but easier: it may be nearly as long as the original, with the difficulty removed. Knowing which of these you need is the first decision on this page, and the format buttons are named to make it a quick one.

A slide holds five bullets of ten words; a report holds five thousand words. Getting from one to the other is a summary problem, and Key points at 'very short' is built for it: three to five bullets, one idea each, each traceable to a highlighted sentence in the source so that you can answer a question from the audience with the original in front of you. For a speaker's note under the slide, Paragraph at 'short' gives the sentence or two to say.

For a social post or a newsletter blurb about a longer piece, Paragraph at 'very short' gives the hook — the one or two sentences that say what the piece is about — without inventing a claim the piece did not make, which is the mistake that gets summaries of research into trouble. Read it once for hedges: 'may', 'suggests', 'in this sample' should survive into the post.

A literature review is a stack of papers reduced to what each one found, and the summarizer takes the mechanical part of that off your hands. For each paper, paste the abstract, introduction and conclusion — usually under twelve thousand characters together — and choose Key points at 'medium'. You get the question, the method, the main findings and the stated limitations, each tied to a highlighted sentence you can quote.

Two rules keep this honest. First, the summary keeps figures exactly and preserves negations, but you are the one citing: read the highlighted sentences in context before you attribute a finding to a paper. Second, a summary of the abstract summarises what the authors claim, not what the paper shows; for the papers that matter to your argument, read the results section yourself. The tool decides which papers deserve that reading, and that alone saves days.

The most common use of a summarizer is not to replace reading but to choose it. A folder of twenty articles, a search that returned thirty papers, a week of newsletters: summarize each at 'very short', and in ten minutes you know which five deserve an hour. The ones you set aside, you set aside knowing what they said — which is different from ignoring them.

For this use, Paragraph is the format: a couple of sentences you can scan in a row. Keep the word counts in view — a piece that compressed by ninety-five percent was probably thin; one that resisted compression probably has more in it than its length suggests. And when a summary makes you curious, that is the signal: open the original, read the highlighted sentences, and then the rest.

It is worth saying how the length slider came to work, because the story explains a behaviour you may notice. The first version asked the model for a summary of 'about three sentences'. It got summaries that were three sentences long and barely shorter than the source, because the model wrote sentences three times longer than the original's. Length in sentences controlled nothing. Written formats are now targeted in words, list formats in bullets, and each is measured after the fact.

The second problem was self-contradiction. Study notes were described to the model as 'two to four headings with two to four bullets each' — at least eight bullets — while the length target asked for three. The model followed the format and ignored the number, and no retry could fix it, because the instruction was impossible. The format descriptions now take the target as a parameter and never ask for more bullets than the target allows. These two fixes are why a 'very short' summary is now genuinely very short, and why the word counts above the result can be trusted.

• Free AI chat
• Paraphrasing tool
• Free grammar checker
• AI translator
• AI document analyzer
• AI homework helper