Can ChatGPT read PDFs? Text, scans and upload steps

Updated October 3, 2026 · Ask Your Docs

Yes. Attach a PDF to a ChatGPT conversation and ask a specific question about it. Start with a fact you can check on the original page; an upload or confident answer alone does not show what the assistant read. OpenAI’s file upload guidance describes the current availability and limits.

Choose the next step based on your file:

• Selectable text: upload it and ask one known-answer question.
• Scanned pages: check the text layer and OCR output before relying on an answer.
• Charts or images: check whether your plan and upload route can retrieve visuals, then verify the values on the page.

For continuing work, a Project can retain reference files. Attaching a PDF in a Project conversation and adding it as a Project reference file are different routes for visuals. Product details below were checked against OpenAI documentation on October 3, 2026.

How to get ChatGPT to read a PDF

Open a ChatGPT conversation, use the attachment control, select your PDF and wait for the upload to finish. Start with a specific question rather than a request to understand everything. OpenAI documents file uploads on free and paid plans, with account settings and usage limits affecting availability. See the File Uploads FAQ for the current rules.

1. Choose a document you are permitted to upload. Give it a descriptive filename with a date or version.
2. Open the original in a PDF reader and identify one fact you can verify independently.
3. Attach the PDF and ask for that fact, its supporting passage and its location.
4. Compare the response with the original before expanding to a summary or comparison.

Use only the attached PDF. What is the campaign launch date? Give the filename, section heading and a short supporting quote. If the document does not state the date, say that it is missing.

This first question checks the connection between a response and its evidence. A convincing summary alone is harder to audit: it can sound right while omitting the one detail that matters to your decision.

Check whether the answer is supported

A source reference is a starting point for verification. Open the cited document, find the relevant passage and check that it supports the claim being made. Keep the filename and version visible when several PDFs discuss the same topic. A quote from last quarter can be accurate and still be the wrong source for this quarter.

For a practical acceptance check, choose three questions: one with an explicit answer, one requiring two passages, and one whose answer is absent. Write down the expected result before asking. This is a suggested evaluation method, not a claim that every assistant passes it. Record failures as well as successes.

For example, a fictional campaign brief might state a launch date but say nothing about renewal terms. A useful answer identifies the date and admits the missing terms. Do not reward an answer for completing every field if the document cannot support it.

Can ChatGPT read scanned PDFs, images and charts?

Text extraction, OCR for image-only scans and interpretation of embedded visuals are different routes. First try selecting and copying text from the original; then test a page with a chart or table, ask for its page number and exact evidence, and compare the answer with the page.

The PDF-to-Markdown tool can help inspect extracted text, but conversion does not perform OCR or preserve chart data and page layout. Keep the original for those checks.

A scanned PDF contains page images; a digitally exported PDF may contain selectable text. Try selecting a sentence in your PDF reader. If copying it produces nothing or garbled characters, inspect the text layer before relying on text extraction. OCR can add searchable text to a scan, but you still need to check names, decimal points and column order against the page.

OpenAI’s PDF visual retrieval guide documents visual retrieval for Enterprise PDFs attached in a conversation, including a Project conversation or an interaction with a published GPT. PDFs added as GPT knowledge or Project reference files use text-only retrieval. Other plans use text retrieval for PDFs and discard embedded images. If a chart matters, check the plan and route before trusting an answer about it.

When a chart matters, ask which values and labels support the conclusion, then inspect them yourself. If the upload route does not preserve that visual information, provide an accessible text table or use a supported image input. Keep that transcription beside the original so another person can check it.

How to ask about tables and long reports

For a table, request a small extraction before asking for analysis. Name the table heading, the period and the required columns. Check several rows, including a total and an exception. A correct-looking total cannot prove the individual rows were mapped correctly. If the source is an Excel workbook, working from that original can avoid ambiguities introduced when a spreadsheet is laid out as PDF pages.

Find the table titled “Regional campaign budget.” Return region, reporting period and budget exactly as stated. Keep currency and units. Flag unreadable cells; do not infer missing values. Then list the passages used.

For a long report, organize the task around sections: identify relevant headings, extract evidence, then synthesize the answer. Compare the section list with the document’s contents page. For exhaustive checks, maintain a review checklist outside the conversation; one broad response is weak evidence that every page received attention.

How large a PDF can ChatGPT read?

OpenAI’s documented upload limits include 512 MB per file and two million tokens for text and document files. Other account, storage and upload-rate limits apply. These numbers describe acceptance boundaries. They do not guarantee exhaustive analysis or correct retrieval of every passage in an accepted file.

If an upload fails, confirm the account and workspace you are using. Check the file’s size and whether it opens normally. Look at OpenAI’s service status before repeating the same request. Failed attempts can count toward rate limits. Keep the error text and time if you need support.

If only one document fails, create a smaller copy of the relevant section while preserving its title, date and original page references. If every attachment fails, investigate account limits or a service issue first. Compressing an image-heavy PDF may reduce bytes without making its text any more readable.

Can you keep PDFs for future chats or share them?

Yes: ChatGPT Projects group reference files, instructions and related conversations. They are a useful starting point when you want to continue working with the same source material in ChatGPT. Shared projects also support collaboration; check the sharing options and workspace controls available in your account.

Set a clear document owner and update process. Replace or remove obsolete reference material deliberately, and ask a known-answer question after a change. Name both documents when comparing versions. Keeping files available is different from proving that a particular response used the newest one.

A separate AI knowledge base becomes relevant when organizing and reusing sources across assistants is part of your workflow. Evaluate the setup, access controls and source output using your own files. Choose a native project if it already meets your needs; adding another system creates another place to maintain.

Where Ask Your Docs fits

Ask Your Docs is built around four steps: upload a file, organize it, connect your AI and ask with sources. Use the Connect your AI instructions to check the setup for your assistant. Verify a sourced answer after connecting before treating the workflow as ready.

Start with a small, representative set of permitted documents. Ask the same known-answer questions you used for direct uploads. Check that the source passages answer the question and that the files you expect are available. This makes the choice between a conversation, a project and a library concrete.

Frequently asked questions

Explore more document guides or return to the Ask Your Docs demo.