What stays local and what does not
This is the part that matters most, so here it is plainly. The PDF binary is opened and parsed in your browser. The file is not uploaded, and its images, embedded attachments, and metadata are never read or transmitted.
The text extracted from the pages is a different matter. To write a summary, that text is sent to a server AI service, along with the file name and the page count. So the words in your document do leave your device. The run is also recorded in your account history, inputs and output together, so you can find it again later, and a clear history action removes the runs you have not explicitly saved.
What that means in practice: this is fine for a report, a paper, a manual, or a public filing. It is not the right tool for a document you are contractually or legally required to keep off third-party services. For those, the local browser tools that never send anything, like PDF to Text, will extract the text without any of it going anywhere.
- The PDF file: read locally, never uploaded
- The extracted text: sent to the AI service and stored in your run history
- File name and page count are sent alongside the text
The three summary styles
TL;DR gives you the gist in a few lines. Use it to decide whether a document is worth reading at all, which is the most common reason people summarize something.
Key takeaways is the default and produces a bulleted list of the points that matter. This is the one to use when you need to act on a document rather than just triage it, because a list is easier to scan and easier to paste into notes than a paragraph.
Executive summary is longer and more structured, and it is built to be shared. Use it when someone else has to understand the document without reading it. Whichever you pick, the summary comes back as plain text you can copy or download as a .txt file.
Where it will not help
A scanned PDF produces nothing. If the pages are images of text rather than real text, the extraction finds nothing to send and the tool stops before making any network request, telling you the document is almost certainly a scan. There is no character recognition here, so run the file through an OCR tool first and summarize the text that comes out.
Very long documents get truncated. The summary is written from roughly the first 20,000 characters, which is about 8 to 10 pages of dense text. A warning appears when your document is longer than that. For a book or a long report, split it into sections and summarize each one rather than trusting a summary built from the opening chapter.
Password-protected PDFs cannot be opened at all, since there is no password field. And the usual caution about summaries applies: it is a compression of the document written by a language model, so verify anything you are going to rely on against the original text.
- Scanned PDFs need OCR first, since there is no text to extract
- Roughly the first 20,000 characters are summarized
- Encrypted PDFs cannot be opened