PaperfyPAPERFYAI paper library and reviewHomeGuide
Sign up freeLog in
Paperfy Guides
how to do a literature reviewLiterature review6 minFor researchers and students starting a literature review

How to do a literature review: five steps froma pile of papers to an account you can defend

You start your literature review and everything goes smoothly—right up until the papers are collected. Then the folder passes 50 PDFs and you lose all sense of what to do next. When reviews stall, the cause is rarely reading ability. It's process design. A literature review isn't a recap of the papers you've read—it's the process of organizing prior research so that it explains your own question. This article breaks that process into five stages.

Published: 2026-07-22

In this article

The overall shape: a five-stage processStage 1: Collect — record your searchStage 2: Select — inclusion and exclusion criteriaStage 3: Read — skim with AI summaries, then read the originalsStage 4: Compare — build a summary tableStage 5: Write — turn the table into a narrativeA workspace that supports all five stagesHow rigorous does your review need to be?Start with Paperfy

The overall shape: a five-stage process

A literature review runs through five stages: collect, select, read, compare, and write.

Most people who get stuck are trying to do all five at once. Finish one stage at a time. Going back a stage is fine; skipping one is not. That rule alone changes how the work moves.

Stage 1: Collect — record your search

Search with term combinations derived from your question, not whatever words come to mind. Note the database, the exact search query, and the search date.

It feels tedious, but a review with no record of how you searched can't be reproduced later—not even by you. Even an informal review needs at least this much documentation to explain to advisors and reviewers.

Stage 2: Select — inclusion and exclusion criteria

You don't need to read everything you collected. First define your inclusion and exclusion criteria: population, time period, study design, language, and so on.

When you exclude a paper, jot the reason next to it in your list. A one-liner like "excluded: pediatric population" saves you from wondering, months later, why that paper isn't there.

Stage 3: Read — skim with AI summaries, then read the originals

For the papers that remain, first use AI summaries to grasp the basic frame—aims, population, methods, results, limitations—then read the originals in depth, starting with the papers central to your review.

Wherever the summary and the original give you different impressions, treat the original as correct. And always verify quoted numbers and conclusions against the original PDF and its figures. A review that skips this step is built on sand.

Stage 4: Compare — build a summary table

Arrange the papers you've read in a single table along shared axes: population, methods, outcomes, main results, limitations. This summary table is the backbone of your review.

For how to build the table—and how to keep papers from getting mixed together—see the article on summarizing multiple papers.

Stage 5: Write — turn the table into a narrative

Finally, turn the table into prose. The basic order: what is known → where results diverge or fall short → what hasn't been examined → how that connects to your question.

Don't march through the papers one by one—group them by finding and write about the findings. That's the difference between an annotated list and a review.

A workspace that supports all five stages

In Paperfy, collected PDFs are laid out one page per paper, with the AI summary, figures, tables, and original text in one place. The back-and-forth between skimming (stage 3) and close reading with notes (stage 4) happens right there.

Summaries are hand-editable, so you can record exclusion reasons and "where this goes in the review" alongside each paper—even if the review takes months, every paper keeps a record of your decisions.

How rigorous does your review need to be?

One last thing: the rigor required for the background section of a student project and for a systematic review in a submitted manuscript are very different.

If systematic comprehensiveness and bias assessment are required, the review must be designed to standards such as PRISMA. Check early with your advisor which kind of review yours needs to be.

Key takeaways

  • Record the search query, search date, and database.
  • Write down inclusion/exclusion criteria first, then log each exclusion reason.
  • Verify cited figures and conclusions against the original PDF.
  • Build the summary table before writing the narrative.

Don't let months of review work get lost

With Paperfy, each paper keeps its PDF, AI summary, figures, tables, and decision notes together—so you can return to any of it at any point in a long review.

Use Paperfy nowView demo

Literature review

Explore This Topic

Read the full cluster for this workflow, from first steps to practical use.

Browse all guides

Read Next

These guides continue the same workflow.

AI tools for literature reviewAI for literature reviews: delegate the tasks, keep the judgmentsummarize prior research with AISummarizing prior research with AI: from a stack of summaries to a usable synthesismanaging research literatureManaging prior research: citations and understanding are two different jobs

Sign up free and create your paper library.

Store, organize, summarize, and review your own PDFs with AI.

Sign up free