PDF summaries fail when you ask “summarize this.” Good summaries match the depth to your need: 60-second glance, executive 1-pager, section-by-section detail, or critical-reading audit. These templates target each depth and PDF type.
What these prompts solve
“Summarize this PDF” gets you a Wikipedia-style abstract that’s too vague for an executive and too shallow for a researcher. These prompts force depth, audience, and output shape — so the summary is usable, not just produced.
Who this is for
Researchers screening 20 papers a week, analysts reading industry reports before earnings calls, legal-adjacent staff doing first-pass contract reviews, founders extracting takeaways from competitor decks and pitchbooks, students preparing seminar discussions, anyone who pays time for PDFs that should pay back in decisions.
When not to use these prompts
Skip them for short docs (≤3 pages) — read directly. Skip them for any PDF you need to cite verbatim — summary loses exact wording that legal/academic work requires. And don’t use them on PDFs whose source you don’t trust; AI summary of a misleading source produces a confident misleading summary.
Prompt anatomy / structure formula
A PDF-summary prompt should always carry six elements:
- PDF type: research paper / industry report / contract / pitch deck / financial filing.
- Reader role: who consumes the summary (executive, researcher, lawyer, you).
- Depth tier: 60-sec glance / 1-page executive / section-by-section / critical audit.
- Extraction targets: claims, methodology, risks, obligations, action items.
- Verification hook: “cite the page/section where each claim appears” or “list quotes you used”.
- Honesty rule: model must mark anything it inferred vs anything stated outright in the PDF.
Best for
- Industry reports (Gartner, McKinsey, sector outlooks)
- Academic papers (abstract / methodology / results / limitations)
- Contracts (your obligations, theirs, termination, liability)
- Financial reports (10-K, 10-Q, earnings transcripts)
- Pitch decks (claim audit, ask, traction)
- Long internal memos and strategy docs
- Multi-PDF comparison (which to trust)
17 copy-ready prompt templates
1. 60-second glance
Read the attached PDF. Output: 1 sentence on what this is, 3 sentences on the core finding, 1 sentence on the limitation I should be aware of. Nothing else. No more than 100 words total.2. Executive-ready 1-pager
Summarize this PDF for an executive who will not read it: 5 bullets (≤20 words each), 1 line on the key risk, 1 question I should ask the author. End with a 1-sentence "what to do with this" recommendation.3. Section-by-section summary
Output a summary by section. For each section: 1-line headline + 2–3 bullets + page number. Skip pure-formatting sections (TOC, references). Flag any section that's longer than 1 bullet's worth of insight as "thin".4. Critical reading
Read the PDF as a skeptical reviewer. Output: 3 strongest claims with evidence, 3 weakest claims with what's missing, 3 questions I'd ask the author. For each weak claim: what specific data or citation would strengthen it.5. Compare 2 PDFs
I have 2 PDFs. Compare on: thesis, methodology, evidence quality, conclusions. End with: "If I trust only one, which and why." Then list 1 question that would help me decide between them.6. Action items from a report
This is an industry / strategy report. Extract: 5 actionable takeaways for {my role}. For each: action, why it matters, when to act (now / next quarter / monitor only), 1 risk of acting on it.7. Contract / terms summary
This is a contract. Output: (a) my obligations, (b) their obligations, (c) termination clauses, (d) liability caps and indemnities, (e) any unusual / red-flag clauses. No legal advice — just a structured summary with page references. Flag any clause with non-standard language for legal review.8. Translate jargon to plain English
Identify the 10 most-used technical terms in this PDF. For each: define in 1 sentence in plain English. Then re-summarize the abstract using only those plain-English definitions. Output: glossary + plain-English abstract.9. Academic paper deep-dive
This is an academic paper. Output: (a) research question in 1 sentence, (b) methodology in 3 sentences, (c) key findings as numbered bullets with effect sizes, (d) limitations as bullets (split into author-acknowledged vs my-inferred), (e) 3 follow-up studies this paper enables.10. Financial report (10-K / 10-Q) summary
This is a public-company financial filing. Output: (a) top-line / bottom-line vs prior period, (b) 3 segments doing best, 3 doing worst, (c) 3 risk factors that differ from last year's filing, (d) one-paragraph forward-looking commentary from MD&A. Cite page numbers.11. Pitch deck audit
This is a pitch deck. Output: (a) the ask (amount, use), (b) the 3 strongest claims, (c) the 3 weakest / unverified claims, (d) traction metrics with the period over which they were measured (call out missing periods), (e) team-vs-claim mismatch if any.12. Methodology check
Below is a paper or report. Audit the methodology only: sample size and selection, control / comparison group, statistical tests used, confounders acknowledged or missed. For each weakness: 1 sentence describing it, severity (high / med / low), what citation would fix it.13. Side-by-side: report claims vs cited sources
Below is a report. Pick the 5 most consequential claims. For each: the claim, the cited source (or "no citation"), and your honest assessment of whether the source supports the claim (yes / partial / no — with reasoning). Stop once you find 2 claims with no source.14. Extract data tables
This PDF contains tables and charts. Output: (a) a list of every table with its title and page, (b) the 3 most-cited data points with their source table, (c) any number quoted in the body that doesn't appear in a table (likely calculated or stated without source).15. Decision-oriented summary
I have a decision to make: {decision}. Summarize this PDF only as it relates to that decision. Output: (a) facts that support the decision, (b) facts that oppose it, (c) the missing fact that would tip me, (d) honest 1-sentence recommendation.16. Long-PDF chunked summary
The PDF is 100+ pages. Summarize in 2 passes: pass 1 — section-by-section bullets (template #3). Pass 2 — synthesize the section bullets into a 1-page executive summary referring back to section numbers. Output both passes.17. Read-and-quiz-me
Summarize this PDF in 200 words. Then output 5 questions whose answers are in the PDF — 3 factual, 2 inferential. After my answers, score me and point to the page that resolves each. Use this to test my understanding, not just the summary's coverage.Common mistakes
- “Summarize this” with no constraint. Vague output, no depth.
- Trusting the summary without spot-checking. Especially on numerical claims — verify 2–3 by hand for any PDF you’ll quote.
- Skipping the limitations / weaknesses section. Summaries of biased sources are biased summaries.
- Asking for an “executive summary” when you’re not an executive. You need detail; “executive” means 1-page-and-done.
- Pasting PDFs over the context window. Long PDFs need chunked summary (template #16), not truncated single-pass.
- No page citations. A summary without page refs is unverifiable.
- Treating contracts like reports. Contract summaries need a structured legal-style breakdown (template #7), not bullets.
How to push results further
- Always tell the model the PDF type (paper / report / contract / deck) up front. Same prompt produces wildly different quality with type vs without.
- Demand page citations on every claim (templates #3, #7, #10). It both forces grounding and makes spot-checking instant.
- For long PDFs, chunk first, synthesize second (template #16). Single-pass on 100+ page docs misses the second half.
- For research papers, run template #9 followed by #12. Findings without a methodology check are decorative.
- For pitch decks and reports you’ll act on, run template #13 before deciding. Most fabrications hide in uncited claims.
- Compose summary + Q&A by following with chatgpt PDF summarization workflow to iterate depth without re-uploading.
- Pair glance + decision-oriented summary (templates #1 + #15) when triaging 20+ PDFs — keeps only the ones that move a decision.
FAQ
- How long can a PDF be before AI struggles? Single-pass works to roughly 100k tokens of context (~200 pages of dense PDF). Beyond that, use chunked summary (template #16).
- Should I trust the model’s summary of a contract? As a first pass, yes. Never as the only pass. Anything you’d sign needs legal review on the actual document, not the summary.
- What’s the best model for long PDFs? Claude with its long context, Gemini for very long docs (1M-token context window), GPT-class for shorter / more reasoning-heavy summaries. See claude long document research or a step-by-step Gemini PDF summarization workflow for the 1M-context route.
- Can the model handle scanned PDFs? Only if the text is OCR’d. Image-only PDFs need OCR first; the model can’t read pixels reliably for verbatim quotes.
- What about tables and charts? Tables: yes, if text-encoded. Charts: only the model can describe what it sees; numbers must be verified from the data table behind the chart.
- Should I prompt in English even for a non-English PDF? Prompt in the language of the PDF for best quote-fidelity. Then ask for a translation pass at the end if needed.
Related
- Research summary prompts — research-paper-specific depth
- Literature review prompts — synthesizing across many papers
- Literature matrix prompts — structured comparison tables across papers
- Meeting notes prompts — for transcripts, not PDFs
- Action item extraction prompts — extract takeaways from any long doc
- ChatGPT PDF summarization — chained workflow with iterations
- Claude long-document research — best model setup for 100+ page PDFs
- AI paper reading workflow — end-to-end paper-screening loop
- AI PDF summary use case — end-to-end example with quality checks
- How to Use AI to Summarise a 10-K or Annual Report: One-Page Brief for Non-Investors
Tags: #Prompt #Productivity #PDF summary