GuidesPath

Literature reviewer

Group papers by what they claim and measure, so the disagreements surface before you write the first section.

Intermediate
Lessons5
Time~155 min
Tools5 tools
Before you start
  • Two papers in your field you have read end to end
  • A reference manager with the PDFs attached, not just the citations
0%0/5 steps
Path artifact

A review map: 4–6 clusters, two anchor papers each, three disagreements with the design choice behind them, and gap-led headings.

Finish every lesson and you should have: A review map: 4–6 clusters, two anchor papers each, three disagreements with the design choice behind them, and gap-led headings.

Set the review boundary

15 minElicit

Write four lines that decide what belongs: population, intervention or concept, context, and years. Name one neighbouring topic you exclude. Put the boundary into Elicit and check which suggestions quietly break one of the four lines.

Pitfall
Widening the boundary for one good paper. Cite it as an exception instead, or the review keeps growing until the deadline decides for you.
Verification
You can reject a strong paper and point at the boundary line it fails.
→ See playbook step 1 in detail

Build the source graph

30 minConnected Papers

Seed Connected Papers with two DOIs you have read end to end. Triage the neighbours: each keeps a seed relation and a reason to open it, or it goes. Verify title, year, and venue in Semantic Scholar before anything reaches your library.

Pitfall
Reading the graph as a quality ranking. Proximity means shared citations, which the weakest papers in the field also have.
Verification
Every kept candidate names the seed it came from and a reason beyond graph position.
→ See playbook step 2 in detail

Cluster by claim

45 minZotero

Open every kept paper and write three lines: the claim, the outcome it actually measured, and the question it cannot answer. Group in Zotero by the claim line. Same keyword plus different outcome means two clusters, not one.

Pitfall
Filling the measure line from the abstract. Abstracts report the finding the authors liked, rarely the outcome they preregistered.
Verification
Each cluster has two anchor papers, one methods note, and one limitation note from the full text.
→ See playbook step 3 in detail

Find the disagreement

35 minConsensus

Ask Consensus the yes-or-no version of your question and collect paper pairs that point in opposite directions. Open both sides and write down which design choice explains the split: sample, outcome measure, follow-up length, or funder.

Pitfall
Settling for “findings are mixed”. It hides that one study measured grades at eight weeks and the other measured confidence at one.
Verification
The map holds three tensions with paper pairs, each explained by a design difference.
→ See playbook step 4 in detail

Outline from the gaps

30 minClaude

Draft headings for what is established, what is contested, and what your next search must resolve. Use Claude to compress the cluster notes into candidate headings, then delete every heading you cannot put named papers under.

Pitfall
Keeping a heading that sounds right but has no papers under it. It survives the outline and collapses the day you write the section.
Verification
Each heading lists its papers and the gap that justifies the section.
→ See playbook step 5 in detail
Finish line

You should now have: A review map: 4–6 clusters, two anchor papers each, three disagreements with the design choice behind them, and gap-led headings.

  • You can reject a strong paper and point at the boundary line it fails.
  • Every kept candidate names the seed it came from and a reason beyond graph position.
  • Each cluster has two anchor papers, one methods note, and one limitation note from the full text.
  • The map holds three tensions with paper pairs, each explained by a design difference.
  • Each heading lists its papers and the gap that justifies the section.