Read a research paper efficiently in three passes. First decide whether the paper is relevant and record its research question. Next map the methods, data, results, and stated limitations. Finally verify the claims that matter to your purpose by reading the exact methods, figures, tables, and discussion passages behind them. Efficiency means allocating attention deliberately; it does not mean trusting the abstract or an AI summary as the whole paper.
Creator disclosure: Heni Hazbay creates Summarise Visually and has a direct interest in the app. The critical-reading method comes before the bounded product workflow, and the original paper remains the authority.
Publication and evidence note: This guide was prepared from the cited sources and reviewed project evidence. Its questions, ledger, and fictional example are editorial tools, not a critical-appraisal credential or a captured product test.
Start with a reading purpose, not a speed target
The same paper can be read for different reasons: deciding whether it belongs in a literature review, understanding a method, checking a cited claim, preparing for a seminar, or evaluating whether evidence applies to a real decision. Write the purpose in one sentence before reading. It determines which sections need close attention and which can wait.
Also record the exact paper: title, authors, venue, year, DOI or stable link, version, and whether you have the full text or only an abstract. A preprint, accepted manuscript, and version of record may differ. Do not combine notes from different versions without marking the change.
Nature Education’s overview of scientific papers explains the roles commonly served by sections such as the abstract, introduction, methods, results, and discussion. That structure is a useful navigation map, but disciplines and article types vary. Follow the paper in front of you rather than forcing every source into one format.
S. Keshav’s “How to Read a Paper,” originally published in ACM SIGCOMM Computer Communication Review and hosted in Stanford’s course archive, presents a three-pass approach in which each pass builds greater understanding. The protocol below uses the same broad progression while adding purpose-specific stop decisions, source-status labels, and a claim ledger; it is an editorial adaptation, not a claim that every discipline should use one fixed sequence.
Pass 1: orient and decide whether to continue
The first pass should answer what the paper asks, why it may matter, what kind of evidence it uses, and whether it fits your purpose. Inspect the title, abstract, section headings, figures and tables at a high level, conclusion or discussion overview, and references that connect it to your question.
Create a short orientation record:
- Research question or central claim, in your own words.
- Paper type, such as experiment, observational analysis, review, theory, methods paper, or case study, if the paper states or clearly supports that classification.
- Population, data, material, or domain studied.
- Main comparison or analytical approach.
- The result the authors emphasize.
- One stated limitation or boundary you can locate.
- Relevance to your reading purpose.
Do not settle a consequential conclusion from the abstract. An abstract compresses the paper and may not show all exclusions, measures, uncertainty, null findings, or limitations. Its job in this pass is orientation.
Stop or continue questions after pass 1
Stop and record why when the paper is outside your scope, only an abstract is available for a task that requires methods, the population or outcome is plainly irrelevant, or you have found a corrected or retracted status that makes the chosen version unsuitable. Stopping is an evidence-based allocation decision, not a judgment that the paper has no value.
Continue to pass 2 when the question is relevant, the evidence type can inform your task, or an important claim needs checking. If relevance is uncertain, identify the one section or table most likely to resolve it and read that before committing to a full pass.
Pass 2: map how the paper reaches its result
Read the introduction for the problem and proposed contribution, then the methods and results in an order that lets you connect each claim to how it was produced. You may alternate between a result and its method rather than reading every page linearly.
For empirical work, record the design, sample or dataset, inclusion and exclusion rules, measures, intervention or exposure, comparison, time frame, analysis, and missing-data treatment when they matter to your purpose. For qualitative, theoretical, review, or methods papers, adapt the fields to the evidence and reasoning actually used. Never invent a sample or experimental comparison where none exists.
Inspect figure and table labels, units, denominators, axes, uncertainty displays, and captions. Ask whether the text describes the same quantity the visual shows. A striking chart cannot be understood safely without its definitions and comparison. Equations, appendices, and supplementary methods may also carry essential conditions.
Separate at least four kinds of statement in your notes:
- Reported observation: what the paper’s data or analysis reports.
- Author interpretation: what the authors think the observation means.
- Stated limitation: a boundary the paper identifies.
- Reader inference: your own proposed connection or concern.
Labelling these prevents an inference from being repeated later as a finding.
Stop or continue questions after pass 2
Continue to a close third pass if the claim is central to your work, the method is unfamiliar, the result depends on choices you have not understood, or the paper conflicts with another source. Pause and seek domain help when you cannot interpret a statistical procedure, measurement, specialist method, or reporting convention that materially affects the conclusion.
For a background citation with limited importance, a checked map may be sufficient. Record what you did not read. “Methods and main result checked; supplement not reviewed” is more useful than a polished note that implies complete coverage.
Pass 3: verify the claims you will use
The third pass is selective and exact. Take each claim you expect to repeat, teach, cite, or use in a decision and trace it to the supporting passage, analysis, table, or figure. Read the relevant method and stated limitations. Check whether your wording preserves population, comparison, measure, time, direction, uncertainty, and conditions.
Then test alternative interpretations. Does the design support association, prediction, or a causal claim? Is a subgroup result being generalized to everyone? Is absence of evidence being described as evidence of no effect? Does the conclusion extend beyond the measured outcome? These are questions to investigate, not assumptions that every paper is flawed.
Read cited sources when a consequential premise depends on them. A citation in a paper is not itself proof that the cited source supports your paraphrase. If you cannot access or evaluate a needed source, mark that limit rather than filling it with confidence.
Copyable claim ledger
This ledger is the practical asset for the third pass. Use one row per claim; do not combine several conclusions under one source location.
| Claim I may use | Exact source location | Evidence or analysis | Population and conditions | Uncertainty or limit | My status |
|---|---|---|---|---|---|
| Write a narrow claim | Page, section, figure, or table | What directly supports it | Who, what, comparison, and time | Authors’ limit plus unresolved issue | Supported / revise / do not use |
Add a separate quotation field if exact wording matters. Otherwise write a faithful paraphrase and retain attribution. A completed ledger row should be understandable to you later without making the paper sound more certain, broad, or causal than it is.
Example ledger row using a fictional paper
Suppose a fictional study compares two reminder formats in one volunteer sample over one week. A poor ledger claim would be “Format A improves memory.” A bounded entry might be: “In the fictional one-week volunteer comparison, participants assigned to Format A completed more of the paper’s defined recall items than the Format B group.” The ledger would still need the sample, allocation method, measure, result location, uncertainty, and limitations. The example demonstrates wording only; it reports no real evidence.
Convert the reading map into study notes
Once claims are checked, transfer them into fields rather than a single prose blob. The research paper summary template separates question, method, sample, findings, uncertainty, and limits. The guide on turning a research paper into study notes adds source-located retrieval prompts.
If an AI-generated condensation is involved, apply the AI summary accuracy checklist to important claims. Compare against the paper, not against a second generated answer. Agreement between two outputs does not verify the source.
A bounded Summarise Visually workflow
Current project review supports a bounded PDF extension handoff that supplies selectable text and supports selected summary modes. This is not evidence of OCR, whole-page visual understanding, complete supplement handling, scientific-validity assessment, or a home-screen document picker. Installed extension availability is not claimed here.
After completing pass 1 yourself, choose a bounded text selection you are permitted to use. A Detailed result can be treated as a candidate condensation; Key Points can be treated as candidate map entries; Q&A can be treated as candidate prompts. None of those modes establishes that a method is sound or a conclusion justified.
Place the generated statements in the left column of a checking sheet, then locate each in the paper. Record omitted conditions, visuals, equations, or references. Only checked wording should enter the claim ledger. This keeps the app in an organizational role while the reader owns interpretation and verification.
Limits and verification
- A three-pass strategy is a prioritization method, not a quality score or a universal reading sequence.
- Abstracts and generated summaries can omit design choices, null findings, uncertainty, and limitations.
- Figures, tables, equations, appendices, and supplements require direct inspection; selectable text may not preserve their meaning.
- Critical appraisal can require statistical, methodological, clinical, legal, or other domain expertise beyond this guide.
- Peer review does not guarantee that a paper is correct, and this workflow is not peer review.
- Retractions, corrections, conflicts of interest, reporting standards, and later evidence may change how a paper should be interpreted.
For consequential use, read the complete relevant sections, confirm the version and status, inspect the original visuals and data descriptions, and obtain qualified review where appropriate. Keep a record of sections not read and questions not resolved.
Sources and evidence
The section-navigation description draws on Nature Education’s “Scientific Papers”, and the broad multi-pass progression is informed by S. Keshav’s “How to Read a Paper”, originally published in ACM SIGCOMM Computer Communication Review and hosted in Stanford’s course archive. Both sources were accessed July 20, 2026. The stop-or-continue questions, claim ledger, and fictional example are editorial adaptations and are not presented as a validated reading intervention.
Product claims are limited to reviewed project evidence for supplied selectable text through a bounded PDF extension handoff and selected summary modes. No OCR, full-paper comprehension, methodological assessment, accuracy rate, research recommendation, or current-device paper walkthrough is claimed. See Methodology for the site’s evidence standard.