PDF Summarizer (Key Points)

Summarize a PDF into key points — in its own words

The PDF Summarizer turns a long PDF into a short list of key points. Drop a report, a policy, meeting minutes, a research paper or a chapter of study notes, press one button, and you get the sentences that matter most — the findings, decisions, numbers and requirements — in the order they appear, each with its page number.

This is an extractive summary. The AI does not write anything: it scores every sentence of your document for importance, and the page keeps the highest-scoring ones. Each key point is a sentence copied word for word from the PDF, so the summary cannot introduce facts, figures or claims that are not in the original.

Reading the PDF, splitting it into sentences and OCR for scanned pages all happen in your browser. Only the sentence text of the pages you choose is sent, and only when you press the button.

How to summarize a PDF

1. Drop a PDF onto the upload box. Its text is extracted on your device and split into sentences; headings, page numbers and repeated headers and footers are set aside.
2. Optionally type a page range under Pages to summarise — for example a single chapter — and check the sentences with “See the sentences that would be sent”.
3. Press Pick key points with AI.
4. Choose Short, Medium, Long or drag the slider. The list updates instantly without sending anything again.
5. Switch to Highlight in full text to see the chosen sentences in context; tap any sentence to add or remove it. Then copy the key points or download them as .txt or Markdown.

PDF Summarizer (Key Points) features

How the key points are chosen

After extraction, the text is rebuilt into paragraphs from the layout (line spacing, indents and font size), then split into sentences. Short lines without a full stop, and larger text, become headings; they give structure in the highlight view but are not scored. The numbered sentences go to TypeSafe’s Jev model, which rates each one on a four-step scale — not a key point, minor detail, useful supporting point, or key point a reader must not miss — and returns a probability-weighted score with a confidence.

The browser then ranks the scored sentences and keeps the top share you asked for — 10% for short, 20% for medium and 35% for long, with a minimum of three — and shows them in document order. Sentences scored as filler (below 25%) are never added automatically, so a short memo can produce fewer points than the percentage suggests.

Limits worth knowing

• Very long PDFs: one summary scores up to 400 sentences, roughly 15–25 pages of dense prose. Long reports are opened with the first 60 pages selected; summarise them chapter by chapter with a page range.
• Scanned pages are read with OCR first. Faint, skewed or handwritten pages produce broken sentences, and broken sentences score badly. Clean them up with PDF OCR if needed.
• Other languages work, though scoring is most reliable in English, and the sentence splitter knows English abbreviations best.
• Tables, slides and forms have few real sentences, so an extractive summary says little about them. For tables, use the PDF Table Extractor.
• An extractive summary can lose connecting context — “this” or “they” may point to a sentence that was not picked. Use the highlight view to read around a point.

Privacy

The file never leaves your device. Pressing the button sends one request containing the sentences of the selected pages and a title (the document’s first heading); the server does not keep them. If you need to search a document instead of summarising it, use Ask PDF, and for the raw text, PDF to Text.