Questions and prompts that work well for PDFs (templates)
Updated 13 October 2026 · ChatPDF guides · 5 sources
AI can miss or misstate things. Summaries and answers from AI tools, ChatPDF included, can leave out important details or get them wrong. Check anything that matters against the document itself, especially legal, medical, financial and exam material.
- Good prompts are specific, name the part of the document, say what format you want and let the tool say "not mentioned".
- Chat-with-PDF tools answer from a few retrieved passages, so using the document’s own terms helps them find the right text. 1,2
- Ask for a quote you can search for, then check it in the PDF: models can sound sure and still be wrong. 3
- For studying, prompts that make you answer questions beat prompts that give you more to reread. 4,5
The same PDF can give you a vague, generic answer or a precise, useful one depending on how you ask. The templates below work with most AI PDF tools and with Apple's Writing Tools on selected text. Replace the parts in brackets.
Five rules behind every good prompt
- Be specific about the task. "Summarize" is vague; "Summarize the methods section in three sentences for a non-expert" is not.
- Point to the place. Name the chapter or section by its heading. Chat-with-PDF tools search your document for relevant passages before the model answers, and the heading's words help that search 1.
- Use the document's terms. A passage pulled out of context can be hard to find and easy to misread, so matching the author's vocabulary helps 2.
- Say the format. Bullets, a table, five questions, one paragraph.
- Allow "not mentioned". Add "If the document does not say, tell me." It reduces, but does not remove, invented answers 3.
Templates by goal
| Goal | Prompt |
|---|---|
| Quick overview | "In three sentences: what is this document, who is it for, and what does it conclude?" |
| Key points | "List the [5] most important points, each in one line with the supporting number or example." |
| One section | "Summarize the section titled '[heading]' in [4] bullet points." |
| Definitions | "How does the document define [term]? Quote the definition if there is one." |
| Rules and exceptions | "What is the rule about [topic], and what exceptions or conditions apply?" |
| Numbers | "List every figure the document gives for [metric], with the period and unit for each." |
| Deadlines and actions | "List every deadline, date or required action, with who must do it." |
| Compare | "Compare [option A] and [option B] as described in the document, in a two-column table." |
| Explain simply | "Explain [paragraph or concept] as if to a first-year student, then give one example." |
| Find support | "Does the document say [claim]? Quote the exact sentence, or say it is not mentioned." |
| Critique | "What limitations or open questions does the author mention?" |
| Study | "Write [5] questions that test the main ideas of [section], without the answers." |
Follow-up chains
The second and third questions are where the value is. Some reliable chains:
- Zoom in: "What are the main findings?" then "Tell me more about finding 2: what evidence supports it?" then "What does the document say about its limitations?"
- Pin down a rule: "What is the refund policy?" then "Are there conditions or exceptions?" then "Quote the sentence about the time limit."
- Study loop: "Quiz me with 5 questions on chapter 2." Answer from memory, then "Here are my answers: [answers]. Which are wrong or incomplete according to the document?"
The study loop is worth the effort: practice testing and spreading it out over time are the two techniques rated most useful in a large review of learning research 5, and testing beat rereading for long-term retention in classic experiments 4. More in how to make practice quizzes from a PDF.
Prompts that tend to go wrong
- "Summarize everything in detail" on a long PDF: you get a generic overview. Go chapter by chapter (see summarizing long PDFs).
- "Is this correct?" or "Is this a good contract?": the model will give an opinion that sounds authoritative. Ask what the document says, then judge yourself, or ask a professional.
- "What page is this on?": many tools do not track pages, and a made-up page number is worse than none. Ask for a quote and search for it instead.
- Five questions in one message: each gets a thinner answer, and it is harder to check. Ask one at a time.
- Questions about the world rather than the document: "What is the current interest rate?" is not in your PDF. The model may answer anyway from general knowledge, which can be outdated.
Checking the answer
Take a key word or number from the answer, search the PDF for it and read the surrounding paragraph. If the tool gave a quote and you cannot find it, treat the whole answer as unconfirmed. The full routine is in can you trust AI PDF summaries?
Where ChatPDF helps
ChatPDF has these patterns built in as shortcuts: the Summary quick action ("Summarize this document in 3 key points"), Insights ("What are the main insights from this document?"), and suggested prompts for a summary, a five-question quiz, flashcards and key takeaways. Everything above works as a typed question too, and you can ask in your own language even if the PDF is in another one. Page numbers are not part of its answers, which is one more reason to ask for quotes you can search.
Common questions
What is the best first prompt for a new PDF?
"In three sentences: what is this document, who is it for, and what does it conclude?" It gives you orientation, and you can then ask about the parts you need.
Does telling the AI to answer only from the document stop it from making things up?
It helps, but models can still slip. Models can still produce statements the source does not support, so check important answers in the PDF. 3
Should I ask for a summary or a quiz when studying?
A quiz. Answering questions from memory improves long-term retention more than rereading summaries. 4,5
Sources and further reading
- Lewis P, Perez E, Piktus A, et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems 33 (arXiv:2005.11401). arxiv.org
- Anthropic (2024). Introducing Contextual Retrieval. anthropic.com
- Ji Z, Lee N, Frieske R, et al. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys 55(12):248. doi:10.1145/3571730 (arXiv:2202.03629). arxiv.org
- Roediger HL, Karpicke JD (2006). Test-enhanced learning: taking memory tests improves long-term retention. Psychological Science 17(3):249–255. doi:10.1111/j.1467-9280.2006.01693.x psychology.ecu.edu
- Dunlosky J, Rawson KA, Marsh EJ, Nathan MJ, Willingham DT (2013). Improving students’ learning with effective learning techniques. Psychological Science in the Public Interest 14(1):4–58. doi:10.1177/1529100612453266 (summary page by the Association for Psychological Science). psychologicalscience.org