Chat with a PDF Online Answers that show their source

Is it safe to upload a PDF to an AI chat?

Only the passages picked for your question, plus the last six exchanges, are sent to an outside AI provider, and that provider's own rules apply to what it receives. The PDF, your messages and the answers stay in your account until you delete the chat. Deleting a chat cannot recall text that was already sent.

Can the chat read a scanned PDF?

Sometimes. When no page has any text, the first 8 pages are turned into images and sent to an AI service to be read. Scanned pages inside a PDF that also has normal text are not read. If the chat says it found no text, the scan could not be read yet.

Does the chat make things up?

It can. The AI is told to answer only from the passages it was sent and to cite them, and citations to anything else are dropped. A "not mentioned" answer only means those passages lacked it. Names, figures and dates still need checking on the cited page.

From pages to numbered excerpts

After upload, the text of each page is extracted and cut into excerpts of up to 220 words. Consecutive excerpts on the same page overlap by 40 words, so text near a boundary appears in both. Every excerpt gets an ID made of its page and position, such as P3-C1, and these IDs are what answers cite.

If no page yields any text, as with a fully scanned document, OCR is attempted instead: up to the first 8 pages are rendered with Ghostscript at 160 dpi and sent to OpenAI for transcription. OCR runs only when the whole document has no text, so scanned pages inside an otherwise digital PDF are not read.

Excerpts created from the three-page sample (627 words)

Which excerpts are sent with a question

Excerpts are chosen by word matching, not by meaning. Words of three or more letters in your question are compared with each excerpt, after a short list of common words such as "the", "what" and "does" is ignored. Each occurrence of a word adds 4 points and its presence adds 1. Up to eight excerpts are sent: any you selected yourself come first, and the best-scoring ones fill the remaining places. When fewer than two excerpts match at all, the top of the ranking is used anyway, which for a question with no matching words means the start of the document.

Excerpt scores for three prompts on the sample

What a page reference does and does not confirm

The model is told to answer only from the supplied excerpts and to cite them by ID. Each citation is checked against the excerpts that were actually sent; references to anything else are dropped, and the page number displayed comes from the excerpt record rather than from the model. A citation chip therefore reliably points to a real passage on that page.

The quoted wording next to it is a different matter. The model writes that quote, up to 240 characters, and it is not compared with the excerpt. Click the chip, read the source text and check that it says what the answer claims, especially for names, figures and dates.

Answers that need extra checking

• Summaries of long documents. With no matching words, eight excerpts of at most 220 words from the opening pages are all the model sees.
• "Not mentioned" answers. The model can only report that the excerpts it received lack the information; a passage using different wording may still exist.
• Totals, counts and comparisons across many pages, which rarely fit in eight excerpts.
• Tables and multi-column layouts. Extraction flattens them into a single run of words, so rows and columns can be mixed up.
• Scanned pages beyond the first eight, or scanned pages in a document that also has digital text.
• Accuracy checks. The instructions limit these to internal consistency within the excerpts; nothing is checked against outside sources.
• Legal, medical and financial conclusions. Read the cited pages yourself and get qualified advice before relying on them.

Plan, provider and storage

AI Chat is included in the AI plan: 1,000 messages per calendar month and PDF uploads up to 100 MB. The Free plan and the current Pro plan do not include it; older paid subscriptions keep the chat access they were sold with. A message counts against the allowance only when an answer is returned.

Each question is sent with the selected excerpts and up to the last six completed exchanges of the conversation. The configuration uses OpenAI by default, with gpt-4o-mini as the default model setting and DeepSeek as the fallback when the first provider fails. Those providers' own retention terms apply to what they receive.

The original PDF, the extracted excerpts, your messages, the answers and any notes stay with your account until you delete the chat or the account. Download Markdown exports the extracted text page by page under its excerpt IDs, plus notes and comments, which shows exactly what text the chat could work from. Deleting a chat cannot recall content already sent to a provider.

Messages you may see

• "This PDF is still being prepared for chat." Extraction or OCR has not finished.
• "This PDF did not contain extractable text, and OCR could not read it yet." Neither text extraction nor OCR produced words.
• "The AI provider could not answer right now. Please try again." Both providers failed; the attempt does not use a message.
• "You have used your AI message allowance for this month." The allowance resets with the next calendar month.
• "AI Chat requires the AI plan." The account has no AI Chat access.