Reading papers well
Asking good questions of a paper is a skill, with or without AI.
The three-pass method
A widely used approach from the computer scientist S. Keshav. The first pass takes five to ten minutes: title, abstract, headings and conclusion, to decide whether to continue. The second pass takes up to an hour: read the body, skip the proofs and fine detail. The third pass is for the few papers central to your work, where you try to rebuild the argument yourself.
Questioning a paper with AI makes the first pass far more informative, and tells you which papers deserve the second.
Questions worth asking of any empirical paper
Keep a standard list and ask it every time. The answers become directly comparable across papers, which is most of the work of a literature review.
| Ask | Why it matters |
|---|---|
| What was the research question or hypothesis? | Tells you whether the paper is about what its title suggests |
| Who or what was studied, and how many? | Sets how far the findings generalize |
| What was compared with what? | No comparison, no causal claim |
| How large was the effect, and how uncertain? | "Significant" is not the same as "large" |
| What do the authors list as limitations? | Their own view of the weak points |
| Who funded the study? | Context for reading the conclusions |
Summaries are for triage, not for citing
If a paper will appear in your reference list, read at least the sections you're citing. A summary can be accurate and still miss the qualification that changes how you should use the finding.
Citing a paper you haven't read is how errors travel from one article to the next.
Take notes in your own words
Write two or three sentences per paper in your own phrasing: what they found, how far you trust it, and where it fits your project. These notes become the raw material for your literature review, and writing them yourself is what makes the content stick.