11 ChatGPT Prompts for Reviewing a PDF (Copy-Paste, by Task)

The most effective ChatGPT prompt for reviewing a PDF follows one pattern: role + task + output format. "Review this document" gets you a generic summary; "Act as a contracts analyst. List every clause that creates a payment obligation, in a table with page references" gets you something you can act on.

Below are 11 copy-paste prompts organized by what you're actually trying to do — a quick first pass, a deep review of a specific document type, a critique, an extraction, or a fact-check. There's also a short section on why ChatGPT sometimes fails on PDFs and how to fix each failure, because no prompt works on a file the model can't read.

Key Takeaways

• Every strong PDF review prompt combines three things: a role, a specific task, and a required output format.
• Run a three-question workflow: confirm ChatGPT read the whole file, ask your real question, then verify the numbers.
• Scanned PDFs are the #1 failure point. OCR the file first, or nothing else on this page will help.
• Treat every figure, date, and quote in the output as unverified until you check it against the source page.

Before You Prompt: 3 Checks That Change the Output

Most disappointing results come from the file or the setup, not the prompt. Run these three checks first — they take under a minute.

1. Confirm your PDF has a text layer. ChatGPT reads the text embedded in a PDF. If your file is a scan (a photo of each page), there's no text to read, and the model will guess or stall. The test: try selecting a sentence with your cursor. If you can't highlight words, it's an image. Fix it by running the file through any OCR tool first, then upload the searchable version.

2. Split very long documents. Upload limits and context windows mean a 200-page report may be truncated or read unevenly — users in OpenAI's own community forum report the model skipping details or inventing content that contradicts the file. For long documents, split by section and review in batches, or work chapter by chapter.

3. Decide your role and format before you type. "Summarize this" forces ChatGPT to guess what you need. "Summarize this for a finance director, focusing on cost changes, in five bullets" doesn't. Every prompt below has a placeholder for your context — fill it in.

The Prompts, Organized by Review Goal

First-pass summary

Use these when you've just received a file and need to know what it is and whether it deserves a closer read.

Before summarizing: tell me the title of each major section of this document, and quote the final paragraph word for word.

This is a reading check disguised as a prompt. If ChatGPT can't name the last section correctly, it hasn't processed the whole file — don't trust anything it says next.

Summarize this document for a [your role — e.g., marketing manager]. Cover: the main argument, the three strongest pieces of evidence, and anything that directly affects [your project]. End with three questions I should ask the author.

Deep review by document type

The document type determines what "review" means. A contract review hunts obligations and risks; a paper review interrogates methodology.

Contracts and legal documents:

Act as a contracts analyst. Review this agreement and list: (1) every clause that creates a payment, renewal, or termination obligation, (2) any liability caps or indemnities, and (3) anything unusual compared to a standard agreement of this type. Use a table with clause references. Flag anything ambiguous rather than interpreting it.

Research papers and academic PDFs:

Act as a peer reviewer for a journal in [field]. Evaluate this paper's methodology, sample, and whether the conclusions follow from the data. List the three weakest claims with the section each appears in. Separately, list what a reviewer would ask the authors to add before acceptance.

Business reports and proposals:

Review this report as a skeptical executive. Identify: the decision the report is asking for, the evidence offered for it, the two weakest pieces of evidence, and any risks the report doesn't mention. Keep it under 300 words.

Example scenario (illustrative): an operations lead receives a 40-page vendor contract at 4 p.m. with a decision due the next morning. She runs the contracts prompt above, gets a table of payment and auto-renewal clauses in under a minute, and spends her evening checking those six clauses against the original text instead of skimming all 40 pages. The AI didn't do the review — it told her where to look.

Critique and peer-review simulation

Act as a [hostile reviewer / demanding editor / cautious investor]. Give this document a score out of 10, then list exactly what loses points, quoting the relevant lines. Don't suggest fixes yet — I want the problems first.

Now argue the opposite: make the strongest possible case FOR this document's main claim, then the strongest case AGAINST it. Use only evidence that appears in the document itself.

Extraction: tables, dates, and action items

Extract every date, deadline, and named owner from this PDF into a table with three columns: item, owner, date. Leave a cell blank if the document doesn't specify — do not infer.

List every number in this document that refers to money, percentages, or quantities, with the sentence it appears in and the page or section reference.

Verification (anti-hallucination)

For each claim in your previous answer, quote the exact sentence from the document that supports it. If any claim is not directly supported by the text, mark it "NOT IN DOCUMENT."

What important topics does this document NOT cover? List questions a reader would expect this type of document to answer but that are missing or vague.

The follow-up pattern

One more habit beats any single prompt: treat the review as three questions, not one. First, confirm coverage ("What are the section titles? Quote the last paragraph."). Second, ask your real question. Third, verify ("Quote the source sentence for each claim"). Uploading a file and asking one big question is how you get confident, wrong answers. (OpenAI's own guide to working with files recommends the same specific, task-shaped prompting style.)

Why ChatGPT Fails on PDFs (and the Fix for Each)

If you take one thing from this table: the failure is usually diagnosable in under a minute, and the fix is rarely "write a longer prompt."

Skip the Upload: Reviewing PDFs Inside WPS Office

The ChatGPT workflow above works, but it has two costs: you upload your file to a third-party server (a real concern for confidential contracts and unpublished research), and the answers arrive without page references, so verification means scrolling.

WPS Office has the same prompt-driven review built into its PDF reader. Open the PDF, launch WPS AI from the toolbar, and you can use the exact prompts from this article in the chat panel — summaries come back with source citations pointing to the page each point came from, and follow-up questions stay in the context of the open file. Built-in modes cover the common cases directly: AI PDF Summary for key points, Chat with PDF for Q&A, AI Explain for selected passages, and translation across 100+ languages.

Two honest conditions: AI features require a recent version of WPS Office, and free accounts have daily usage limits (check the current allowance for your region). For a deeper workflow walkthrough, see our guides to using an AI summarizer for long documents and the best PDF editor with AI summarization.

download WPS Office free — review your next PDF where it already lives.

Frequently Asked Questions

Can ChatGPT read a scanned PDF?

Not reliably. ChatGPT reads the text layer of a PDF, and a scan is an image with no text layer. Run the file through an OCR tool first to create a searchable PDF, then upload that version. Test any PDF by trying to select a sentence — if you can't highlight words, it's an image.

Is there a file size or page limit for PDF uploads?

Yes, and it varies by plan and changes over time — check OpenAI's current documentation for your tier. Regardless of the stated limit, very long documents are read less reliably. For 100+ page files, split the document by section and review in batches for noticeably better results.

How do I stop ChatGPT from making things up when reviewing my PDF?

Require evidence with every answer: "Quote the exact sentence supporting each claim, and mark anything unsupported as 'NOT IN DOCUMENT.'" Then spot-check the quotes against the original, especially numbers, dates, and names. Never treat AI-extracted figures as verified.

Can ChatGPT review a PDF written in another language?

Yes. It can summarize and answer questions about documents in most major languages, and you can ask for the output in English ("Summarize this German contract in English"). For full-document translation with layout preserved, a dedicated tool like WPS PDF's translation mode is more practical.

Is it safe to upload confidential PDFs to ChatGPT?

That depends on your organization's data policy and your ChatGPT plan — enterprise tiers handle data differently from consumer accounts. If the document is sensitive, check your policy first, or use a tool that processes the file inside your own document suite, such as WPS Office's built-in AI.

The Bottom Line

A good ChatGPT prompt for reviewing a PDF is a role, a task, and a format — and a good review is three questions: confirm it read everything, ask what you actually need, then verify the output against the source. Start with the prompt that matches your document type, and keep the verification prompt for anything with numbers in it.

This week, take the next PDF that lands in your inbox and run the reading-check prompt before anything else. If you'd rather skip the upload step entirely, download WPS Office free and ask the same questions right inside the document — summaries come with page citations, so the verification step takes seconds, not minutes.