Long Document Summarizer with Decisions for ChatGPT
Summarize a long report, contract or paper into what it says, what it means for you, and what to do next, with page references so you can verify every claim.
Fill in the variables
Prompt
<role>
You are an analyst who reads long documents for busy decision-makers. You report what the document says, flag what matters for the reader's situation, and keep your own opinions clearly separated from the author's.
</role>
<context>
Document (paste the text or attach the file): {{document}}
Who I am and why I am reading this: {{reader_context}}
The decision or question this document should help with: {{question}}
Level of detail I want: {{detail}}
</context>
<constraints>
- Every claim in the summary must be traceable: add (p. X) or (section name) after it. If the document has no page numbers, quote a few words from the source sentence instead.
- Do not add information that is not in the document. Where the document is silent on something I asked, say "not addressed in the document".
- Separate three things with headings: what the document says, what it means for me, what to do. Opinions appear only in the second and third.
- Preserve exact figures, dates, defined terms and names. No rounding of numbers that matter.
- If the document contradicts itself or a figure is inconsistent, flag it.
- Length: the summary should take under 3 minutes to read at the "standard" detail level.
</constraints>
<goals>
- I can answer my question without reading the full document.
- I can find and verify any claim in under a minute.
- Nothing important is left out, and nothing is invented.
</goals>
<instructions>
1. One-paragraph overview: what the document is, who wrote it, its purpose, and its main conclusion.
2. Key points: 5–10 bullets of what it says, each with a reference.
3. Relevant to me: how the points above apply to my context and question, including anything that changes a decision.
4. Gaps and caveats: what the document does not cover, assumptions it makes, anything that looks inconsistent.
5. Suggested next steps: 3–5 concrete actions or questions to ask.
6. Glossary: up to 5 defined terms from the document, in one line each, only if they are needed to understand the summary.
</instructions>
<output_format>
Headed sections in the order above. References in parentheses. No introduction before the overview.
</output_format>
Example input
Document: (a 14-page SaaS vendor agreement; excerpts) p.2 "Term: 24 months from the Effective Date, renewing automatically for successive 12-month periods unless either party gives 90 days' written notice." p.4 "Fees increase annually by the greater of 5% or CPI." p.5 "Customer may add seats at any time at the then-current list price; seats may not be reduced during a Term." p.7 "Uptime commitment 99.5% monthly; sole remedy is service credits of 5% of monthly fees per 0.5% below commitment, capped at 25%." p.9 "Customer data is deleted 30 days after termination; export available in CSV on request during the Term." p.11 "Either party may terminate for material breach uncured after 30 days' notice." p.12 "Limitation of liability: 12 months' fees, excluding breach of confidentiality and data protection obligations." Reader context: founder of a 20-person company evaluating this tool for the whole team Question: Should we sign as is, and what should we negotiate? Detail: standard
Example output generated with Claude Sonnet 4.6 from the example input above
Overview
This is a 24-month SaaS subscription agreement with automatic 12-month renewals (p. 2), annual price increases of at least 5% (p. 4), seat counts that can only go up during a term (p. 5), a 99.5% uptime commitment remedied by capped service credits (p. 7), 30-day data deletion after termination (p. 9), and liability capped at 12 months' fees with carve-outs for confidentiality and data protection (p. 12). Its main effect is to lock in a two-year commitment with rising fees and limited exit options.
Key points
- Initial term is 24 months, then auto-renews yearly unless 90 days' notice is given (p. 2).
- Fees rise every year by the greater of 5% or CPI, with no cap stated (p. 4).
- Seats can be added at list price at any time but cannot be reduced during a term (p. 5).
- Uptime commitment is 99.5% per month; the only remedy is credits of 5% per 0.5% shortfall, capped at 25% of monthly fees (p. 7).
- Data is deleted 30 days after termination; export is CSV, on request, during the term only (p. 9).
- Termination for breach requires 30 days' notice and an uncured breach (p. 11); no termination for convenience is mentioned.
- Liability is capped at 12 months' fees, except confidentiality and data protection (p. 12).
Relevant to me
For a 20-person company, the seat clause is the biggest exposure: if headcount drops, you keep paying for unused seats until the term ends (p. 5). The 5%-or-CPI escalator means a minimum 10.25% increase over the two-year term before any seat growth (p. 4). The 90-day notice window on a 24-month term is easy to miss and triggers another full year (p. 2). The 25% credit cap means a bad outage month is worth at most a quarter of that month's fee (p. 7). The 30-day deletion window after termination is short for a migration (p. 9).
Gaps and caveats
- No termination for convenience is mentioned anywhere in the excerpts; confirm it is absent from the full text.
- The CPI index and reference period are not defined (p. 4).
- Export is "on request" with no format detail beyond CSV and no response time (p. 9).
- Whether service credits require a claim within a deadline is not addressed in the document.
- No security or data-location commitments appear in the excerpts; not addressed in the document.
Suggested next steps
1. Negotiate a 12-month initial term, or a termination-for-convenience right after 12 months with 30 days' notice.
2. Ask for seat reductions at renewal, or a 10–20% downward flexibility mid-term.
3. Cap the annual increase at 5% flat, and define the CPI index if it stays.
4. Extend data retention after termination to 90 days and require export within 5 business days of request.
5. Put a calendar reminder 120 days before the end of the initial term for the renewal notice.
Glossary
- Term: the 24-month initial subscription period, followed by 12-month renewal periods (p. 2).
- Service credits: discounts on future fees given for missed uptime; the only compensation available under this agreement (p. 7).
Why this works & how to customize
Why it works: the traceability rule (every claim carries a page or quote) turns the summary into something you can verify in a minute, and it stops the model from blending the document's content with its own advice. The three-way split, says / means for me / do next, is what decision-makers actually want and what generic "summarize this" prompts never provide.
How to customize: set detail to "brief" for a half-page version or "thorough" for section-by-section coverage. For academic papers, rename the sections to Findings, Methods and limitations, Relevance, and ask for the sample size and effect sizes explicitly. In NotebookLM, upload the PDF as a source and paste only the prompt; the citations come out as clickable references. For very long documents in ChatGPT, attach the file rather than pasting, and ask for the summary per chapter first.
How to customize: set detail to "brief" for a half-page version or "thorough" for section-by-section coverage. For academic papers, rename the sections to Findings, Methods and limitations, Relevance, and ask for the sample size and effect sizes explicitly. In NotebookLM, upload the PDF as a source and paste only the prompt; the citations come out as clickable references. For very long documents in ChatGPT, attach the file rather than pasting, and ask for the summary per chapter first.
Comments (0)
Sign in to comment.
No comments yet.