How to read research papers faster: the three-pass method

Updated 13 October 2026 · ChatPDF guides · 5 sources

AI can miss or misstate things. Summaries and answers from AI tools, ChatPDF included, can leave out important details or get them wrong. Check anything that matters against the document itself, especially legal, medical, financial and exam material.

Most people read research papers too thoroughly too early. They start at the first line and push through dense methods before knowing whether the paper is relevant. A better approach, described in a widely shared note by computer scientist S. Keshav, is to read in passes, each deeper than the last, and to stop as soon as you have what you need ¹.

Pass 1: the five to ten minute skim

The first pass gives you a bird's-eye view ¹:

1. Read the title, abstract and introduction carefully.
2. Read the section and subsection headings, and nothing else in the body.
3. Read the conclusions.
4. Glance over the references and tick off the ones you already know.

Keshav suggests that by the end you should be able to answer "the five Cs": the category of paper (a measurement, a new method, a review), its context (which other work it builds on), whether its assumptions seem correct, its main contributions, and its clarity ¹. For papers outside your field, or ones you only need to know about, this is often where you stop.

On iPhone, the Preview app makes the skim quicker: tap the page count to see thumbnails and spot figures and section breaks, and use Go to Page to jump straight to the conclusions ⁵.

Pass 2: up to an hour, with care

In the second pass you read the paper properly but still skip the hardest details such as proofs ¹. Focus on:

• Figures and tables. Are axes labeled? Are there error bars or confidence intervals? Many weak papers show it here first.
• The method in one sentence. Who or what was studied, how many, compared with what.
• References to follow up. Mark the ones you have not read.

At the end, you should be able to summarize the main thrust of the paper, with supporting evidence, to someone else ¹. Write that summary down in three to five sentences: it becomes your note on the paper (see turning a PDF into notes).

Pass 3: only for the papers that matter

The third pass is for papers you are reviewing, building on or relying on heavily. Keshav describes it as virtually re-implementing the paper: re-creating the work in your head, making the same assumptions and challenging each one. He estimates it takes beginners about four or five hours and experienced readers about an hour ¹. Most papers you meet do not need this.

Using AI carefully

AI tools can speed up the first two passes, but they cannot do the judging for you.

Good uses:

• A quick orientation before your own first pass: "What question does this paper ask and what is its main finding?"
• Explaining jargon or a method you do not know: "Explain what a difference-in-differences design is, as used in the methods section."
• Finding where something is discussed: "Which section describes the participants and how they were recruited?"
• Generating questions to test yourself after reading.

Risky uses:

• Trusting the summary's numbers. A study of AI summaries of medical evidence found factually inconsistent statements and overly convincing claims, and found models struggled to identify the most important information, with more errors on longer texts ². Earlier work on summarization found unsupported content in the summaries of every system examined ³.
• Asking whether the paper is "good". Judging correctness needs the third-pass thinking that only you can do.
• Citing references the AI names. AI tools can make up sources or summarize them incorrectly, so consult and cite the original yourself ⁴.

Check any number or claim you plan to use against the figure or table it comes from; working with tables and numbers explains why tables are a weak spot for AI tools.

Where ChatPDF helps

With ChatPDF you can add a paper from Files and ask the orientation and jargon questions above, then follow up. The "Summary" quick action gives three key points, which is a reasonable starting point for pass 1. Answers do not include page numbers, so use them to find where to read rather than as a source. Papers whose PDF is a scan without a text layer will not work; most papers downloaded from journals or preprint servers have real text.

Common questions

Sources and further reading

1. Keshav S (2007). How to read a paper. ACM SIGCOMM Computer Communication Review 37(3):83–84. doi:10.1145/1273445.1273458 web.stanford.edu
2. Tang L, Sun Z, Idnay B, et al. (2023). Evaluating large language models on medical evidence summarization. npj Digital Medicine 6:158. doi:10.1038/s41746-023-00896-7 nature.com
3. Maynez J, Narayan S, Bohnet B, McDonald R (2020). On faithfulness and factuality in abstractive summarization. Proceedings of ACL 2020:1906–1919. doi:10.18653/v1/2020.acl-main.173 aclanthology.org
4. MLA Style Center (2025). How do I cite generative AI in MLA style? (Updated and revised). style.mla.org
5. Apple Support. Open and view PDFs and images in Preview on iPhone (iPhone User Guide). support.apple.com