Close reading is not the production of a shorter abstract. It asks how evidence was obtained and under which conditions a conclusion holds. AI reading tools can explain terms and locate passages, but may miss negation, qualifications, figures, or supplementary methods.
When this workflow is useful
- Deciding whether a paper should inform a project or review
- Checking relationships among methods, samples, analysis, and results
- Comparing designs, evidence, and limitations across papers
A practical sequence
- 01
Identify the central claim
State the question, principal claim, and declared scope. Keep the authors’ interpretation separate from the results they report.
- 02
Reconstruct the design
Identify provenance, controls, measurements, preprocessing, analysis, and exclusions. Consult registrations, protocols, code, data, and supplements when they affect interpretation.
- 03
Trace claims to evidence
Map consequential claims to figures, statistics, or analyses. Examine effect size, uncertainty, missing observations, and plausible alternatives.
- 04
Write traceable notes
Use AI for bounded questions and confirm answers in the source. Record the question, design, results, limits, and exact page or section locations.
Quality-control questions
- Do notes distinguish author claims, observations, and your interpretation?
- Can important statements be traced to the source?
- Were design limits on causation and generalisation considered?
- Was AI output checked in full context?