How to Write a Narrative Review That Editors Want to Publish

Page nine was the problem. My co-author had built a table there with fourteen rows, one per study, listing sample size, setting, design and "key finding." He'd spent two weekends on it and he was proud of it.

"Reviewers expect to see the evidence," he said.

"They expect to see what the evidence means," I said. "This table tells me what each paper found. It doesn't tell me why three of them disagree with the other eleven."

We argued for most of an afternoon. He wasn't wrong that the table had value. But I'd handled enough narrative reviews by then to know that this exact pattern is what gets them rejected. Fourteen careful summaries, stacked in a row, with the actual thinking left for the reader to do. We eventually cut the table to five columns, reorganized it around the two competing explanations in the literature, and rewrote the text around that tension. The paper was accepted after one round.

That argument is the whole post in miniature. Most narrative reviews fail because they summarize when they should synthesize, not because the authors missed papers.

What a narrative review is actually for

A narrative review is an argued interpretation of a body of literature. It doesn't try to answer a narrow question with pooled data. That's what systematic reviews and meta-analyses do, and they come with their own machinery: registered protocols, exhaustive searches and reporting standards like the PRISMA statement.

Narrative reviews get dismissed as the "weaker" cousin, and I think that's unfair. They're the right tool when a field is conceptually messy, when the evidence spans incompatible methods, when you need to trace how an idea developed, or when the useful contribution is a framework rather than an effect size. A systematic review can tell you whether an intervention works on average. A good narrative review can tell you why the trials keep disagreeing and what the next study should measure.

The trade-off is real, though. You give up the defensibility of a fully reproducible search, and in exchange you're expected to bring judgment. If you don't bring judgment, you've given something up for nothing. Editors notice.

Choosing a topic: pick a question, not a territory

The most common topic mistake I see is choosing a territory instead of a question. "Sleep and cognition in older adults" is a territory. "Why do sleep-extension interventions improve memory in some older-adult trials but not others?" is a question. You can write a review about the second one. The first one turns into a textbook chapter.

A first-year master's student I supervised, Tom, came to me wanting to review "digital health interventions for diabetes." He'd already found over two hundred papers and was drowning. I asked him what had surprised him while reading. After a pause he said that the apps with the best engagement numbers didn't seem to produce the best glucose outcomes, and nobody seemed to talk about it. That became his review. It came to about forty papers, it made a clear argument, and he finished it in a semester.

Before you commit, run three checks:

• The one-sentence test. Can you write the argument of your review in a single sentence before you've written anything else? If not, you don't have a topic yet, just a reading list.

• The existing-review check. Search for reviews published in the last five years on your question. If a strong one exists, your review needs a different angle, a newer evidence base, or a disagreement with it. "There's no review from my country" is rarely enough.

• The scope check. If your core literature is under ten papers, consider a short commentary instead. If it's over a hundred, narrow the question or consider whether you actually need a systematic or scoping review.

Also check the target journal's author instructions. Many journals publish narrative reviews only by invitation, and some will desk-reject unsolicited ones regardless of quality. A short pre-submission inquiry to the editor saves months.

Searching: narrative doesn't mean unsystematic

This is where I disagree with a lot of the advice in circulation. You'll hear that narrative reviews don't need a described search strategy. Technically, no reporting guideline mandates one. In practice, reviewers increasingly ask, and a review with no stated search reads as cherry-picked even when it isn't.

The quality criteria most editors I know have in mind, explicitly or not, are close to those in SANRA, the Scale for the Assessment of Narrative Review Articles. It asks whether the review explains its importance, states its aims, describes the literature search, references appropriately, reasons from evidence and presents data appropriately. Read it before you start, not after.

You don't need a PRISMA flow diagram. You do need a short, honest paragraph. Something like this:

We searched PubMed, Embase and CINAHL from January 2010 to March 2026 using combinations of terms for pressure injury, prevention, repositioning and nursing workload. We supplemented this with backward and forward citation searching of key papers and with guidance documents from national nursing bodies. We prioritized randomized trials and large observational studies, and included qualitative studies where they explained implementation findings. This was not a systematic search, and studies may have been missed.

That last sentence matters. Stating your limitation plainly is more persuasive than pretending you haven't got one.

Practically, I'd do the search in three passes: a structured database search to get the core literature, citation chasing in both directions from the five or six most important papers, and a deliberate hunt for disconfirming evidence. That third pass is the one people skip. If your review argues that X explains the inconsistency, go looking specifically for studies where X was present and the effect still didn't show up.

The EQUATOR Network library is worth browsing too. There's no single dominant reporting guideline for narrative reviews, but you'll find the standards for the primary studies you're reviewing, which helps you judge how well they were reported.

Organizing the literature around ideas, not papers

The order in which you read papers shouldn't become the order in which you write about them. Yet that's exactly the structure of most rejected reviews: "Smith et al. found... Jones et al. found... In contrast, Lee et al. found..."

The tool that fixes this is a concept matrix. Rows are papers; columns are the ideas, variables or explanations you're tracking. It's simple, and it forces you to see patterns across papers instead of within them.

Once the matrix exists, the argument often falls out of it. In this example the "Notes on adherence" column is where the story is.

I learned that from a research nurse named Dana, who wrote one of the best narrative reviews I've handled. She was reviewing repositioning schedules for pressure injury prevention, and the trial literature was a mess of null and positive results. Most authors would have concluded "evidence is mixed." Dana had worked night shifts for eleven years. She added an adherence column almost on instinct, and it showed that the null trials were mostly the ones where the intensive schedule simply wasn't delivered. Her review argued the field had been testing staffing capacity, not repositioning frequency. Reviewers loved it because it came from knowledge none of them had.

That's what a narrative review lets you contribute. Your clinical, methodological or field experience is a legitimate lens. Use it.

Writing synthesis instead of summary

Here's a before-and-after from a draft I edited (details changed).

Before:

Martinez et al. (2019) found that mindfulness training reduced burnout in residents. Okafor et al. (2021) found no significant effect. Chen et al. (2022) reported a modest reduction at three months that was not sustained at six months. Further research is needed.

After:

Trials of mindfulness training for resident burnout disagree mainly on timing of follow-up. The two trials reporting benefit measured outcomes within three months, while the trial with six-month follow-up found the effect had faded. None of the three changed scheduling or workload, which suggests the training may offer short-term coping without addressing the conditions driving burnout. Future trials should measure outcomes at six months or later and test mindfulness alongside workload changes.

The second version does four things the first doesn't. It names the pattern before citing anyone. It explains the disagreement. It offers an interpretation and flags it as an interpretation. And it turns "further research is needed" into a specific recommendation someone could actually act on.

Some sentence frames I find myself suggesting again and again:

• "The studies that found X share one feature that the others lack: ..."

• "These findings are compatible with two explanations. The evidence favors the first because ..."

• "Most of this literature measures A, but the clinical question is about B."

• "The apparent consensus rests largely on a single cohort, so ..."

Every paragraph should open with a claim, not a citation. If you can't write the claim, you haven't finished thinking about that section yet.

A structure that works

Journals vary, but this skeleton survives most author guidelines:

1. Introduction. Why this question matters now, what existing reviews have and haven't done, and your aim in one or two sentences.

2. Methods (brief). Databases, dates, terms, supplementary searching, how you selected and weighed studies, and the limitations of that approach.

3. Thematic sections. Organized around the ideas in your concept matrix, each opening with a claim.

4. Synthesis or framework. Where you pull the themes together. A figure or model often belongs here.

5. Implications and gaps. Specific enough that a reader could design the next study.

6. Limitations. Of the evidence and of your review.

Keep tables of individual studies as supplementary material unless the journal wants them in the main text. Synthesis tables, organized by theme, earn their place much more easily.

The mistakes I see most often

• The annotated bibliography. One paragraph per paper. The most common reason narrative reviews get rejected.

• Uncritical weighting. Treating a pilot study with twenty participants as equal to a multicenter trial. Say how much weight each piece of evidence deserves and why.

• Citing reviews of reviews. If a claim matters to your argument, cite and read the primary study.

• Hidden advocacy. An argument is good. Ignoring the evidence against it isn't. Reviewers in your field know the inconvenient papers.

• "Further research is needed." Always true, never useful. Say what research, measuring what, in whom.

• Stale searches. If your search ended eighteen months before submission, rerun it. Reviewers check.

Authorship, conflicts and AI tools

Narrative reviews attract authorship problems because the work is less visible than in an empirical study. The ICMJE recommendations apply here as much as anywhere: everyone listed should have contributed substantially to the conception or interpretation, drafted or critically revised the text, approved the final version and agreed to be accountable for it. Senior names added for weight don't qualify.

Conflicts of interest matter more in reviews than people assume, because interpretation is the whole product. If you've consulted for a company whose product you're discussing, disclose it. The COPE guidance on conflicts and on text recycling is worth a read if you're adapting a thesis chapter or a previous review.

If you used AI tools to search, screen or draft, say so in your methods or acknowledgments according to the journal's policy. AI tools can't be authors, and you're accountable for every claim and every citation. Verify each reference against the source. Fabricated or mismatched citations are now one of the faster routes to a desk rejection.

Before you submit: a short checklist

• Can I state the review's argument in one sentence, and does it appear in the introduction?

• Is my search described honestly, including its limits?

• Does every section open with a claim rather than a citation?

• Have I explained, not just reported, the main disagreements in the literature?

• Have I included the strongest evidence against my interpretation?

• Are my gaps and recommendations specific enough to act on?

• Did I check the journal accepts unsolicited narrative reviews?

• Are authorship, conflicts and AI use disclosed?

For a longer, practical treatment, the widely cited article by Green and colleagues on writing narrative literature reviews for peer-reviewed journals still holds up.

Where to start this week

If you're sitting on a review idea, don't open a blank document. Open a spreadsheet, list the ten papers you already know matter, and add columns for the ideas you think connect them. Within an hour you'll know whether you have an argument or just a pile.

Much of the drag in review writing is mechanical: re-finding papers, losing track of which note came from which source, copying summaries between tools. I use PACR, an AI-native research workspace, to keep searching, reading, notes and drafting in one place, so the thread from a search result to a sentence in the draft doesn't get lost. If you want to try building your first concept matrix there, you can set up a free workspace and start with a literature search on your question.

Whatever tool you use, the thinking is the part nobody can do for you. Dana's adherence column didn't come from software. It came from eleven years of night shifts and the nerve to say the field had been asking the wrong question. Find your version of that column.