ChatPDF paper reader
Upload a research PDF, read the original pages immediately, and ask questions answered from retrieved evidence excerpts. Every page citation is verbatim-checked against the paper, and answers that fail verification are not billed.
Updated 2026-07-22
How source-grounded Q&A, citation verification, answer-length billing, generation records, and evidence notes work.
ChatPDF is a focused PDF reader with source-grounded question answering. It renders the original pages immediately while server-side processing, including OCR for image-dominant pages, continues in the background. You can search pages, switch to an extracted-text view, ask questions answered from retrieved evidence excerpts with verbatim-checked page citations, keep evidence notes, and track reading checkpoints that link back to source pages.
The server retrieves up to six evidence excerpts from the processed paper and generates the answer from those excerpts. Every page citation carries a quote that is verbatim-validated against the excerpts. An answer whose citations fail validation is demoted and not billed, and a reply stating the document cannot answer the question is not billed either. Multi-turn follow-ups keep the conversation context, with suggested next questions.
Billing is by answer length: an answer starts at 1 credit and never costs more than 2. Unlimited plans are unmetered. Cannot-answer replies and answers demoted for fabricated citations are not billed. Generation records keep a per-answer history with credit receipts, and the server stores only a hash of your question.
Yes. ChatPDF requires a WriteGo account so uploaded papers, processing results, generation records, and evidence notes share one authorization boundary. The original pages render immediately in the browser while server-side processing continues, and your generation history stores only a hash of each question rather than its text.
ChatPDF renders the original PDF pages immediately and adds an extracted-text view, page search, and zoom. Server-side processing — including server OCR for image-dominant pages — continues in the background, so reading never waits for extraction to finish.
Source-grounded Q&A retrieves up to six evidence excerpts from the processed paper and answers only from those excerpts. Each answer carries page citations, supports multi-turn follow-ups, and suggests questions you can ask next.
Quotes cited in an answer are validated verbatim against the retrieved excerpts. An answer whose citations cannot be verified is demoted and not billed, and a reply saying the document cannot answer the question is not billed either.
An answer costs from 1 credit and at most 2, based on reply length; unlimited plans are unmetered. Generation records keep a per-answer history with credit receipts, and the server stores only a hash of your question, not its text.
Save a passage or an answer without losing its exact quotes. Notes re-check every quote, and restoring an old answer re-verifies each citation against the current document before you rely on it.
Tables, equations, and charts rarely extract cleanly, so ChatPDF keeps a visual-evidence register for your manual transcriptions. Entries stay marked needs-review until you verify them against the original page.
Checkpoints for the research question, method, findings, and limitations link back to source pages instead of standing in as an unsupported summary.
ChatPDF is a single-paper reader. Cross-paper synthesis — the 5–30 paper evidence matrix, versioned source packages, and reference management — lives in the Literature Review Matrix, which keeps its own paper library, isolated from ChatPDF's on the backend.
Uploaded PDFs are processed server-side, with server OCR for image-dominant pages, and the original pages open immediately while processing continues. Password-protected or damaged files may remain unreadable, and pages that fail extraction stay visible in the original-page view.
Yes. Every answer is AI-generated from evidence excerpts the server retrieves from your paper — there is no local answer engine. Citations quote the excerpts verbatim and are validated before the answer is delivered.
ChatPDF replies that it cannot answer from this document, and that reply is not billed.
Generation records keep a per-answer history with credit receipts. For privacy, the server stores only a hash of each question. Restoring an old answer re-verifies every quote against the current document.
Not in ChatPDF. The 5–30 paper evidence matrix, source packages, and the DOI, Zotero, BibTeX, and RIS reference workflows live in the Literature Review Matrix, which keeps its own paper library separate from ChatPDF's.