Research & PolicyFree
Literature Review Synthesis for Researchers
The synthesis organized around what your papers disagree about, with every claim cited to the paper and the page it came from.
River reads the papers you include and writes the synthesis around the claims they contest, rather than around the order you read them in. Every finding lands in an evidence matrix first: one row per paper, with the population, the design, the measure, the direction of effect and the estimate beside it. The prose is then written out of that matrix, so a paragraph about a contested claim names the studies on each side and the methodological difference separating them.
Unlike the writing guides that rank for this, which stop at organize by theme and identify where studies disagree, this fills in the column that decides whether a disagreement is real. Two papers reporting opposite effects on productivity are frequently measuring different things, one asking employees and one reading logged output. Advice about structure cannot catch that. The matrix can, and where a paper does not report enough to fill the column, the draft names the gap instead of assuming the value.
Built for doctoral candidates writing the review chapter, postdocs turning a reading pile into a related work section, and applied researchers defending an evidence claim to a committee. Reach for it once screening is finished and the PDFs are in a folder. Merge the library first with the Zotero duplicate merge plan. When the matrix itself is the deliverable and the prose comes later, the extraction table goes field by field instead, and the results section writer handles your own model output. Reading already rated against your questions lives in the annotated bibliography pack.
"Most studies found" is a synthesis method
The sentence most studies found an effect reads like a summary of the literature. It is a synthesis method, and the Cochrane handbook names it as one: vote counting, applied informally through wording such as 3 of 10 studies showed improvement, or the majority of studies, or few studies. Its verdict on the version keyed to statistical significance is blunt, that the method has serious limitations and is unacceptable, because a study too small to reach significance gets counted as evidence against the effect.
Take an illustrative set of eleven papers on remote work and productivity. Seven report an increase and four a decrease, so seven of eleven point one way. On a two-sided sign test that is p = 0.55, which is no evidence: nine of eleven would give 0.065, and you need ten before the count clears the conventional threshold. Now sort them on what was measured. All six that asked employees to rate their own output point up, six of six, p = 0.031. Four of the five reading logged output point down.
The literature was never in conflict. Six papers measured perception and five measured throughput, and the pooled count buried both findings under a word. So a review that has to make an evidence claim needs three matrix columns before it needs an outline: what each study measured, which direction its effect ran, and whether an effect estimate can be recovered at all. PRISMA 2020 asks reviewers to describe the synthesis methods and the rationale for them in item 13d, and a count of studies is one of those methods.
How it works
Send the papers
The included set as PDFs or a reference export, plus the one claim your review has to settle.
The matrix comes first
One row per paper: population, design, what was measured, the direction of effect, and the estimate.
Disagreements get sorted
Contested claims separate from real consensus once the matrix is sorted on measure and on design.
Argue it in chat
Push back on a grouping, add the three papers you found late, or ask for a shorter related work section.
What you get
- The synthesis organized by contested claim, with the studies on each side of every one
- An evidence matrix as a Sheet: population, design, measure, direction and estimate per paper
- Every claim in the prose cited to the paper and the page it was read from
- Direction counts reported with their sign test, so a count of studies carries a number
- The apparent disagreements that collapse once the papers are grouped by measure
- The papers whose effect direction could not be recovered, listed rather than dropped
Common questions
My supervisor said my review reads like a list of summaries. What actually fixes that?
Paragraphs organized around a claim instead of around a paper. A summary paragraph opens with an author and reports what they did. A synthesis paragraph opens with a contested claim, names the studies on each side, and then says what separates them. The matrix is what makes the second one writable, because the split has to be visible in a column before it can be argued in a sentence.
Can it write the review if I only give it the topic?
No, and that is the one thing it refuses. A review assembled from a topic rather than from papers you actually included is fabrication, and it is the failure mode that ends careers. Send the PDFs, a reference export, or at minimum the title, design and reported finding of each study. What it will do from a topic alone is help you scope the question.
How many papers does it need?
It works from about eight, and the direction counts get meaningful later than people expect. On eleven papers you need ten pointing the same way before a two-sided sign test clears 0.05, so a small set usually ends in a described pattern rather than a tested one. The draft says which of the two it is, per claim, instead of letting a count imply a result.
Is this a meta-analysis?
No. A meta-analysis pools effect estimates with their precision, and that needs numbers most papers in a mixed set do not report in a common measure. This works one rung down: direction of effect, counted and tested, with the effect estimates shown alongside wherever they exist. Where enough estimates share a measure, it says so and tells you a pooled analysis is available.
Will it invent citations?
No claim reaches the draft without a paper and a location behind it. Every sentence in the synthesis carries the paper it came from and the page or section the passage sits on, so a supervisor can check any line in one click. Where the matrix has an empty cell, the prose says the study did not report that rather than filling it in.
What if the papers genuinely agree?
Then the review says so and spends its length elsewhere, on the boundary conditions and on what nobody has tested. Manufactured controversy is the mirror image of a summary list, and it is just as easy to spot. Real consensus with a named gap is a stronger chapter than a fake disagreement, and the matrix shows which one you have.
Does it handle interview studies and qualitative papers?
Partly. A direction of effect does not exist in a qualitative paper, so those rows carry the reported finding, the analytic approach and the participant count instead, and they are grouped rather than counted. Mixed sets work well, since the qualitative papers usually explain why the quantitative ones split. A set that is entirely qualitative wants thematic synthesis, which is a different method from this one.
Literature Review Synthesis for Researchers
Fill in the form and your workspace opens with the work already underway.