Research & PolicyFree tool
Literature Review Synthesis for Researchers
A synthesis of the papers you've screened, organized by claim, with an evidence matrix and every sentence cited to the page it came from.
Free to start · Opens in your River workspace, ready to edit and share
How to synthesize sources in a literature review
To synthesize sources in a literature review, you write about what a body of studies establishes together, not what each paper found on its own. A good synthesis is organized around claims about the evidence. Each paragraph states a claim, brings in the studies that support or complicate it, explains any disagreement and says how strong the evidence is. The review closes with what no study has tested yet.
One habit matters more than any other: before you call a literature divided, check what each study actually measured. Papers that seem to contradict each other may have measured different things under the same name.
Fix the question and the papers first
You can't synthesize a topic. "Remote work" gives you nothing to organize around, but "Does remote work change individual productivity?" does, because every paper either bears on it or doesn't. The University of Toronto's writing advice says a review must be organized around and related directly to the thesis or research question1 you are developing.
Then settle which papers are in. Synthesis works on a fixed set, so finish searching and screening before you start, and note why you dropped each paper you excluded. If you screen against formal criteria, the PRISMA systematic review template has a log for those decisions and the flow diagram that reports them.
Build an evidence matrix
Before you write a sentence, put the papers in a table with one row per paper and one column for each thing you will compare. Johns Hopkins' Sheridan Libraries call this a synthesis matrix, a way to record the main points of each source and document how sources relate to each other2. A row tells you about one paper. A column tells you about the literature.
Columns for an evidence matrix
| Column | What to record | Example |
|---|---|---|
| Design | Trial, field experiment, survey, interview study or review | Randomized field experiment |
| Sample | Who took part, how many and where | 180 call-center staff at one firm |
| Measure | What was measured, and how | Calls handled per shift, from system logs |
| Direction | Which way the effect ran, significant or not | Higher output when remote |
| Estimate | The effect size and its confidence interval, if reported | +4% (95% CI 1% to 7%) |
| Finding | The paper's own words for its main result | "Remote staff handled more calls per shift" |
| Quality | The weakness that most limits how far you can trust it | Only volunteers took part |
| Location | The page, section or table it came from | p. 9, Table 3 |
The example entries are illustrative.
Keep the paper's own wording next to any value you code, so anyone checking your work can see what you read rather than trust what you wrote. If you've annotated your sources against your research questions in an annotated bibliography, start the matrix from those notes.
Never fill a cell with a guess
If a paper doesn't report its sample size or effect estimate, write "not reported" rather than estimating it. How many papers reported an estimate at all is worth a sentence in the review.
Sort the matrix before you compare
Mixed findings need a second look. Sort the matrix by what was measured, then by design, then by population. If the direction of effect changes with one of those columns, the conflict may be two groups of studies answering different questions.
When a split explains a disagreement, the split is your finding, and it leads the paragraph. Report how you grouped the papers, and why. The SWiM reporting guideline for systematic reviews, published in The BMJ in 2020, asks authors to describe the groups used in the synthesis and the reasons for them3. Its examples are groupings by population, intervention, outcome and study design.
Count directions, not significant results
Writing "most studies found an effect" is a synthesis method, even if it doesn't feel like one. The Cochrane Handbook calls it vote counting, and notes that phrases like "most studies found" and "the majority of studies"4 apply it informally, and perhaps unintentionally.
Counting only the statistically significant results is, in Cochrane's word, unacceptable4: a study too small to reach significance gets counted as showing no benefit. Instead, classify each study by the direction of its effect alone, then test the count with a sign test.
In the handbook's worked example, 10 of 12 studies favor one side. Typing =2*BINOM.DIST(2, 12, 0.5, TRUE) into Excel gives p = 0.039, where 2 is the smaller of the two counts.
All 20 papers: 12 found higher productivity with remote work, 8 found lower. Sign test: p = 0.50, no evidence either way.
Self-rated output (11 papers): 11 higher, 0 lower. p < 0.001.
Logged output (9 papers): 1 higher, 8 lower. p = 0.04.
Claim for the review: in these papers, remote workers rated themselves more productive, while their logged output was lower.
- Direction only. Each paper counts by the way its effect ran, significant or not.
- Sort first. The pooled count looked like a stalemate. Sorted by measure, it is two consistent findings.
- Report sizes too. A sign test ignores how large each effect was, so show the estimates beside the counts.
Small groups can't settle much. With five studies, even five out of five gives p = 0.06, so report the count without claiming a result.
Write the synthesis around claims
The University of Toronto's guide names the giveaway of a review that only summarizes: it's usually a bad sign to see every paragraph beginning with the name of a researcher1. Open each paragraph with a claim about the literature instead, and bring the studies to it.
State the claim
Say what the studies establish together, in one sentence, before you name any of them.
Bring in the evidence
Cite the studies on each side, and name each one's design in the sentence that uses it.
Explain the split
Say what separates the sides: measure, design, population or date. If nothing does, say that.
Say what's unsettled
Close with what this evidence can't answer, so each gap grows out of a paragraph.
Match your verbs to the design. A survey tells you what respondents reported, and a cohort study finds an association. Save causal verbs like "raises" for designs that can support them, such as randomized trials.
For the overall structure, UNC's Writing Center lists three ways to organize the body of a review: chronologically, thematically or methodologically5. Claim-led paragraphs work inside any of them. Choose method when findings track design, and chronology only when the field changed its position.
Name the gaps, then check every citation
Write the gap statement from the matrix, not from memory. Sort it and look for what no paper covers. You might find no trials in small firms, no study longer than six months, or no paper that measured both kinds of output.
Then check every citation against the page it came from. A 2015 meta-analysis in PeerJ pooled 28 studies that checked whether references in medical journals supported the statements citing them. It found that roughly one in every eight to nine references was seriously incorrect6, meaning the cited source did not support the statement at all.
The authors suggest placing each citation next to the words it supports, not in a group at the end of a sentence or paragraph. A reference and citation check compares each citation with the passage it points to.
Before you hand in the review
- Every paragraph opens with a claim, not an author
- Each disagreement was checked against measure, design and population
- Every count of studies counts directions and has a test, or is called a description
- Causal verbs appear only where the design supports them
- Every citation sits next to the words it supports and points to a page you checked
- Each gap names something no paper in the matrix covers
To have the matrix built and a first draft written for you, attach your papers to the tool at the top of this page. It cites each sentence to the page it came from.
How it works
Start with the papers you've already screened, attached as PDFs or pasted as a list, and a line on the question your review has to settle. River reads each paper and records its design, sample, measure, direction of effect and page in an evidence matrix. Then it writes the synthesis in your workspace, organized by contested claim unless you choose theme, method or chronology.
It works only from the papers you give it, and marks any value a paper doesn't report as "not reported" rather than guessing. If you don't state a question, it works one out from the papers and puts it at the top for you to change. When it's done, it tells you the largest split it found.
Attach your papers
Attach the PDFs or paste each paper's details, plus the question your review has to settle.
River builds the matrix
It reads each paper and records the design, sample, measure, direction of effect and page.
Get your synthesis
A draft organized by claim, each sentence cited, with the evidence matrix as a sheet beside it.
Push back and refine
Question a grouping, add papers you found late, or cut it down to a related work section.
What you get
- A synthesis organized by claim, naming the studies on each side of every contested point
- An evidence matrix as a sheet, with design, sample, measure, direction, estimate and page per paper
- A citation on every sentence, pointing to the paper and the page, section or table
- Counts of studies by direction of effect, with a sign test once eight or more contribute
- The disagreements that dissolve when papers are sorted by measure, design or population, and those that hold
- A list of what each paper didn't report, and a gap statement built from what no paper covers
Common questions
What is the difference between summary and synthesis in a literature review?
A summary reports what one source found, while a synthesis says what several sources establish together. A summary paragraph usually opens with an author and describes their study. A synthesis paragraph opens with a claim, cites the studies on each side of it, and explains what separates them, such as a different measure, design or population.
Can a literature review include sources other than journal articles?
Yes. Government reports, theses, conference papers and industry surveys can all belong in a review, as long as they say how they reached their findings. The tool reads them like papers, and the matrix marks whatever a source leaves out. For tracking competitors instead, the competitive intelligence report template dates every fact and links each conclusion to the evidence behind it.
How do I write a literature review synthesis with River?
In the box at the top of this page, attach your included papers as PDFs, up to 10 at a time, or paste them as a list. Add the question your review must answer. River builds an evidence matrix as a sheet, then writes a synthesis organized by claim. Both open in your free River workspace, where you can edit, share or ask for changes.
Can it write a literature review from just a topic?
No. A review built from a topic rather than from papers you actually included would have to invent its evidence, so the tool won't write one. Give it only a topic and it drafts a scoping note instead, with the question framed, inclusion criteria and search terms, so you can find and screen the papers first.
Will it make up citations?
No. It cites only the papers you supply, and ties each sentence to the page, section or table it came from, so you can check any line. When a paper doesn't report a value, the matrix says so. Before you submit, a reference and citation check can test whether each source says what your sentence claims.
Is this the same as a meta-analysis?
No. A meta-analysis statistically pools comparable effect sizes, and many sets of papers don't report enough of them. This tool counts the direction of each effect instead and tests the count with a sign test once eight or more papers contribute. Any effect estimates the papers report stay in the matrix. If enough of them are comparable, a meta-analysis of that subset may be worth running.
Is the literature review synthesis tool free?
Yes, it is free to start. Your synthesis and its evidence matrix open in a River workspace, where you can edit them, share them or ask River to change them. Paid plans add more AI use and access to the most advanced models.
Sources
University of Toronto. “The Literature Review: A Few Tips On Conducting It.” advice.writing.utoronto.ca (opens in a new tab)
Johns Hopkins University Sheridan Libraries. “Synthesize - Write a Literature Review.” guides.library.jhu.edu (opens in a new tab)
The BMJ. “Synthesis without meta-analysis (SWiM) in systematic reviews: reporting guideline.” pmc.ncbi.nlm.nih.gov (opens in a new tab)
Cochrane. “Chapter 12: Synthesizing and presenting findings using other methods.” cochrane.org (opens in a new tab)
UNC Writing Center. “Literature Reviews.” writingcenter.unc.edu (opens in a new tab)
PeerJ. “Quotation accuracy in medical journal articles—a systematic review and meta-analysis.” pmc.ncbi.nlm.nih.gov (opens in a new tab)
Built by
Chandler Supple
Co-founder & CEO, River
Chandler has spent the last six years working in AI, building tools for millions of users and Fortune 100 companies. In 2025, he co-founded River to build a new kind of operating system for AI agents.