Summarize a long document by defining the reader and decision, mapping the document, extracting evidence by section, and writing a synthesis tied to source locations. Check omitted limitations, numbers, dissent, and uncertainty. A shorter output should reduce detail without changing the document's meaning.
Who this is for: Professionals and students who need dependable summaries of reports, policies, research, transcripts, or technical documents.
- Define what the summary is for because relevance changes with the reader and decision.
- Build section-level notes before compressing an entire long document into one response.
- Audit omissions and altered certainty, not just factual errors that appear in the summary.
Set the summary contract
Name the audience, intended use, length, and required coverage. An executive decision brief needs findings, risks, choices, and unresolved questions. A study summary needs concepts, arguments, evidence, and definitions. Without a purpose, a model may favor prominent introductory material over details that actually matter to the reader.
State whether the summary may interpret or must remain descriptive. Require preservation of important uncertainty, conditions, exceptions, and disagreement. If recommendations are wanted, request them in a separate labeled section so readers can distinguish source content from generated analysis.
Map before compressing
Create an outline from headings, page ranges, and document structure. Note where methods, results, limitations, appendices, and definitions live. For scans or complex PDFs, check whether tables and text were extracted correctly. Visual structure may carry meaning that plain text conversion loses.
Summarize sections into evidence notes with source locations. Capture key numbers with units and denominators, named decisions with owners, and claims with their qualifications. This intermediate layer reduces the chance that one broad response overlooks a small but decisive section.
Synthesize without flattening
Combine section notes according to the summary contract. Group related points and explain relationships, but do not convert correlation into causation or a proposal into a decision. Preserve conflicting findings when the document does. Use direct quotations sparingly and verify them against the original page.
Make evidence traceable through page, section, timestamp, or document identifiers. For a multi-document summary, name the source behind each major point. If the source does not resolve a requested question, say that it is not established rather than filling the gap from general knowledge.
Run an omission audit
Compare the draft with the document map. Check every major section, table, conclusion, limitation, and appendix that affects interpretation. Search specifically for negations, exceptions, ranges, confidence language, and minority views. These elements are easy to omit and can reverse practical meaning.
Have a knowledgeable person verify high-impact summaries against original pages. Keep the source available beside the summary and record the review date. Current provider document features change, so consult official documentation for supported formats, file limits, and visual-processing behavior before relying on a specific workflow.
Summarize a research report for a school board
A 90-page report evaluates a tutoring pilot and contains mixed results across grade levels.
- Define the board's decision and require outcomes, implementation costs, limitations, and subgroup differences.
- Map the report into methods, overall findings, subgroup tables, implementation notes, and limitations.
- Extract section summaries with page references and preserve sample sizes plus uncertainty language.
- Draft a two-page synthesis that separates reported findings from possible policy implications.
- Audit every grade-level claim against its table and confirm that no subgroup result is presented as an overall effect.
Summary fidelity checklist
Review the draft against these questions before it replaces anyone's first reading of the source.
- Purpose: Does the summary answer the named reader's decision without adding a different agenda?
- Coverage: Are major sections, appendices, tables, limitations, and disagreements represented proportionately?
- Precision: Do numbers retain units, populations, time periods, and uncertainty qualifiers?
- Traceability: Can each material statement be located in the original document quickly?
- Boundary: Are source findings, reviewer interpretation, and generated recommendations visibly separated?
Common mistakes
- Asking for a generic summary and then using it for a decision the prompt never identified.
- Summarizing extracted PDF text without checking whether tables, columns, or footnotes were lost.
- Shortening cautious source language into a confident conclusion that the document does not make.
Try one
A summary says a program succeeded, while the source reports improvement only for one subgroup and no significant overall change. How should it be corrected?
The correction should remove the general success claim, report the overall result and subgroup result separately, retain the source's uncertainty language, and cite the relevant table or section. Evaluation should reward attention to population and scope. It should reject wording that treats a subgroup observation as proof for all participants or omits the overall finding.
Sources
- Gemini API document processing guideOfficial guidance for understanding PDF documents and their visual elements.
- Gemini API long context guideOfficial guidance for reasoning over large inputs and locating relevant material.