You have a set of quantitative studies on the same question, and you need to say what they collectively show. A pooled effect estimate is not on the table — the designs differ, the outcomes are measured differently, or the sample is too heterogeneous to justify one number. That does not mean you are limited to a list of study-by-study summaries. It means you need a structured, transparent narrative synthesis that a coauthor can follow and a reviewer can audit.
The core task is to convert scattered results into a defensible argument: what is consistent, what conflicts, what explains the difference, and what remains unknown. This guide walks through that sequence — from scoping the question to a finished, handoff-ready synthesis — with checkpoints, a worked example, and completion criteria you can apply before you send the manuscript out.
What does "synthesis without meta-analysis" actually require?
A narrative synthesis is not a looser version of a meta-analysis. It is a different method with its own rigor requirements. Where a meta-analysis leans on statistical pooling, a structured narrative synthesis leans on explicit comparison: you define the question, document how you found and selected studies, extract comparable information, and then reason across studies in a way that another researcher could reproduce.
Three commitments separate a rigorous synthesis from a literature summary:
- A pre-specified question and eligibility criteria. You decide what counts as relevant before you start reading, so selection is not driven by which papers you happened to find.
- A documented search and selection trail. The path from search to included studies is recorded, not reconstructed from memory.
- A transparent synthesis method. You state how you compared findings — by outcome, by design, by population — rather than asserting a conclusion.
NIH guidance frames careful assessment of prior knowledge, strengths, weaknesses, and knowledge gaps as core to rigorous research practice, alongside transparent methods and reproducibility. That framing applies directly here: your synthesis is itself a piece of research, and it should be as auditable as the studies it summarizes.
Start by diagnosing your starting state
Before drafting anything, identify which situation you are in. The right workflow depends on it.
You are revising a manuscript. A reviewer asked you to reconcile conflicting findings or to justify a claim that spans several studies. You likely already have a reference list; what you need is a defensible comparison structure and a clear statement of what the evidence does and does not support.
You are closing a project. You have a body of results and prior literature and need a synthesis section that will survive a lab handoff. Here the priority is the evidence trail — someone else must be able to pick up your files and understand how each conclusion was reached.
You are starting a new review. You have a question but no corpus. The priority is a planned, documented search and a screening process that links multiple reports of the same study.
You are preserving an evidence trail. The synthesis exists, but the method behind it is implicit. Your job is to make the decisions visible.
Each state shares the same backbone — question, search, selection, extraction, comparison, reporting — but the entry point and the amount of new searching differ.
How do you turn a broad topic into a synthesisable question?
A synthesisable question names the population, the exposure or intervention, the outcome, and the comparison. It also names what would make a study ineligible. Write both down before searching.
Cochrane guidance recommends planning searches from eligibility criteria rather than from a topic label. That single discipline prevents the most common failure in narrative synthesis: a corpus assembled by keyword drift, where later studies are included because they are adjacent to the topic rather than because they meet a stated criterion.
At this checkpoint, produce a one-page protocol note containing:
- The review question in one sentence.
- Inclusion criteria (population, design, outcome, date range, language).
- Exclusion criteria, with the reason each is excluded.
- The outcomes you will compare across studies, and how you will group them.
- The synthesis approach — for example, grouping by outcome and then by design, or by population.
This note is your first artifact. It is also the document you return to when a coauthor asks why a particular study was left out.
How should you search and select studies so the trail holds up?
Cochrane's search guidance is specific and worth following closely even when you are not conducting a full systematic review: use multiple sources, combine free-text terms with subject headings, document the process, deduplicate records, screen in stages, link multiple reports of one study, and rerun searches before publication.
In practice, that means:
- Build the search from your criteria, translating each concept into free-text terms and subject headings.
- Search more than one source so that no single database's coverage determines your corpus.
- Record the exact search — terms, sources, dates — as you go, not afterward.
- Deduplicate before screening, so the same paper is not counted twice.
- Screen in stages: title and abstract first, then full text.
- Link reports of the same study, so a trial reported in three papers counts as one study, not three.
- Rerun the search before you submit, so the synthesis reflects the current literature.
This is where a scholarly search interface earns its place in the workflow. Using WisPaper search to run and revisit your searches keeps the discovery step in the same environment where you will organize what you find. If you prefer to start from a public entry point, the WisPaper Scholar Search page explains the paper-finding starting point.
Checkpoint: you should be able to hand a coauthor your search strings, the sources searched, the dates, and the number of records at each screening stage. If you cannot, the trail is not yet auditable.
Where do you keep the corpus and the decisions about it?
Screening generates two things that must stay together: the papers and the reasons they were included or excluded. Separating them is how evidence trails decay.
Save or upload the papers you are considering into a single organized library, and keep your inclusion and exclusion log alongside it. My Library is the place to save and upload papers and organize them, which makes it the natural home for the screened corpus. When the corpus is assembled, you can ask questions grounded in those papers using Library QA, which works on papers uploaded to My Library or added from search results.
Use Library QA for orientation, not for conclusions. Good questions at this stage include: which studies report the outcome I care about, which ones use a design I planned to exclude, and where do the reported populations differ. The answers help you build the comparison table; the table, not the answer, is what goes in the manuscript.
Checkpoint: every included study is in the library, and every excluded study has a recorded reason. The count of included studies matches the count in your screening log.
What goes into the extraction table?
Your extraction table is the instrument of synthesis. Design it before you extract, and extract the same fields from every study so that comparison is possible.
| Field | What to record | Why it matters for synthesis |
|---|---|---|
| Study identifier | First author, year, and linked reports | Prevents double-counting one study |
| Design | Study type and key structural features | Groups studies for comparison |
| Population | Who was studied and how they were selected | Explains divergence in findings |
| Exposure or intervention | What was measured or delivered | Establishes comparability |
| Outcome definition | Exactly how the outcome was measured | Flags outcomes that are not truly the same |
| Direction of effect | Which way the result points | Enables consistency and conflict mapping |
| Reported uncertainty | Whatever the study reports | Keeps claims narrower than the evidence |
| Limitations noted by authors | The study's own caveats | Feeds the gap analysis |
Two disciplines make this table useful. First, record the outcome definition verbatim enough that you can tell whether two studies are measuring the same thing. Second, record direction of effect separately from magnitude — without pooling, direction is often the most defensible cross-study comparison you can make.
How do you compare findings across studies without pooling?
Comparison proceeds in layers, and each layer answers a different question.
Layer 1 — Consistency. Sort studies by outcome and ask whether the direction of effect agrees. Where it does, you can state a consistent pattern across the studies you included, naming the designs and populations involved.
Layer 2 — Conflict. Where direction disagrees, do not average the disagreement away. Look for the structural difference: a different population, a different outcome definition, a different design. Your extraction table should already contain the candidate explanation.
Layer 3 — Explanation. Group studies by the characteristic that seems to track the divergence. If studies with one design point one way and studies with another design point the other, that grouping is your finding — and it is more informative than a pooled number would have been.
Layer 4 — Gaps. Use the limitations your authors reported, plus outcomes no included study addressed, to state what remains unknown. NIH guidance treats identification of knowledge gaps as part of rigorous practice, and in a narrative synthesis the gap statement is a result, not a formality.
When the corpus is large or the comparison structure is complex, a longer-running agent workflow can help you work through the literature systematically. Scholar Agent is designed for longer research tasks in the areas of literature review, inspiration, and reproduction, which fits the stage where you are mapping studies against each other rather than running a single search.
Checkpoint: for every outcome you plan to discuss, you can point to the studies on each side of the consistency question and name the characteristic that explains any divergence.
Worked example: reconciling a conflicting outcome across six studies
Suppose a manuscript claims that a particular exposure is associated with a measured outcome, and a reviewer notes that the cited studies disagree. You have six studies, and pooling is inappropriate because the outcome is measured three different ways.
Step 1 — Protocol note. You state the question, the six-study corpus, and the decision to compare by outcome definition rather than by pooled estimate.
Step 2 — Search and selection record. You document how the six were identified, note that two additional reports describe the same study as one of the six, and link them so the corpus is six studies, not eight papers.
Step 3 — Extraction. You complete the table above for all six. The outcome-definition column reveals three measurement approaches.
Step 4 — Comparison. Sorting by measurement approach, you find that studies using two of the three approaches point in the same direction, while studies using the third do not. The third approach captures a narrower construct.
Step 5 — Revised claim. The manuscript claim narrows: the association is consistent across studies measuring the outcome in the two comparable ways, and the divergence is attributable to a different construct rather than to contradictory evidence.
Step 6 — Gap statement. You note which populations and which measurement approaches remain unstudied.
Step 7 — Trail. The protocol note, search record, extraction table, and comparison notes are filed together so a coauthor can retrace every step.
This example shows the payoff of structured narrative synthesis: the reviewer's objection is answered not by asserting confidence but by showing the structure of the evidence.
How do you keep the method trail intact for coauthors and peer review?
A method trail is only useful if it survives the project. Three practices keep it alive.
Keep artifacts in one project space. The protocol note, search record, screening log, extraction table, and comparison notes belong together, not scattered across email threads. Research Projects is the WisPaper area for creating and managing a project, which gives the synthesis a single home that a coauthor or successor can open.
Write the method section from the artifacts, not from memory. Describe the search, the selection, and the synthesis approach as you actually performed them. PRISMA 2020 provides a reporting checklist, expanded explanations, an abstract checklist, and flow-diagram templates for systematic reviews, including transparent reporting of search, selection, and synthesis methods. Even when your work is a narrative synthesis rather than a full systematic review, the checklist's emphasis on transparent reporting of search and synthesis methods is a useful standard to write toward.
Verify the reference list before submission. A synthesis carries a long reference list, and a fabricated or mismatched entry undermines the whole trail. TrueCite checks BibTeX against real academic databases to flag hallucinated references — a concrete final pass before the manuscript goes out.
Checkpoint: a coauthor who has never seen the project can open it, read the protocol note, follow the search record to the included studies, and reproduce your comparison from the extraction table.
When is the synthesis ready?
Use these completion criteria as a gate before submission or handoff. Each is binary: either you can demonstrate it or you cannot.
- The review question and eligibility criteria are written down and dated.
- The search is documented with terms, sources, and dates, and was rerun before submission.
- Records were deduplicated, and multiple reports of one study are linked.
- Screening was staged, and every excluded study has a recorded reason.
- The extraction table is complete for every included study, with outcome definitions recorded.
- Every outcome discussed has an explicit consistency statement naming the supporting studies.
- Every conflict has a proposed structural explanation drawn from the extraction table.
- The gap statement identifies specific unstudied populations, outcomes, or designs.
- The method section describes the search, selection, and synthesis approach as performed.
- The reference list has been checked for fabricated or mismatched entries.
- All artifacts live in one project space a coauthor can open.
If every box is checked, the synthesis is auditable. If any box is unchecked, the gap is in the trail, not in the prose — fix the artifact first.
How does the workflow change when you are starting fresh?
When you begin a new review rather than revising an existing one, the sequence is the same but the front end carries more weight. Spend your effort on the protocol note and the search record, because everything downstream inherits their quality.
Two adjustments help. First, run a scoping pass before committing to eligibility criteria: a broad search tells you whether the literature is large enough to synthesize and whether the outcome is measured consistently enough for cross-study comparison. Second, decide your comparison structure early — by outcome, by design, or by population — because that decision determines which fields your extraction table needs.
If the scoping pass reveals that the field is thin or the question is unsettled, you may need to reframe before investing in extraction. Idea Discovery supports a workflow that takes research topics, background, references, and experimental constraints and generates research ideas — useful when the scoping pass suggests the question itself needs sharpening.




