How to compare research papers
Do not compare abstracts. Abstracts are advertisements. Compare the questions, the methods, and the evidence you could point at if someone disagreed.
Only compare papers that share a question
Two papers about “learning” are not comparable if one measures exam scores in a week and the other measures self-reported confidence. Write each paper’s question in a line. If the lines do not overlap, you may still cite both — but you are not comparing results. You are changing the subject.
A one-page grid
For two or three papers, a table beats a narrative:
| Paper A | Paper B | |
|---|---|---|
| Question | … | … |
| Who/what | sample or scope | sample or scope |
| What was done | design in a clause | design in a clause |
| Outcome | named measure | named measure |
| Headline number | with units | with units |
| Bound | the limitation that matters | the limitation that matters |
Fill it from the papers, not from memory. Add a page number to every cell you might reuse. Empty cells are findings: that paper did not report the thing you need.
Disagreement is usually about the measure
When two good papers “contradict,” look at the outcome first, then the sample, then the time window. True replication fights are rarer than two teams answering cousins of the same question. Your comparison should say that plainly.
Do not average them
A 2% effect in a large trial and a 40% effect in a lab of 24 people do not become 21%. If you need a synthesis, say which study is closer to your setting and why. That sentence is the comparison.
Tooling, honestly
PDF Lens opens one PDF at a time. It will not “diff” two papers. The practical pattern is: keep the grid in your notes, search each PDF for the outcome name, and extract a claim with its page. If you want a product that chats across a pile of mixed sources, that is a different shape than an AI PDF reader — see the NotebookLM note.