If you're writing a systematic review or a meta-analysis, there's one figure the journal will ask for without exception: the PRISMA flow diagram. It isn't optional or decorative. It's the visual proof that your study selection was systematic and not "I kept looking until I had enough studies to make my point" — and along with the Method section, it's the first thing a reviewer checks to decide whether the rest of the manuscript can be trusted.
This article covers the four phases of the diagram with a full worked example, the mistakes I see most often in review, and how to build it without depending on PowerPoint or doing the arithmetic by hand: at the end there's a free generator that computes each box's numbers for you, so they can never contradict each other.
What the PRISMA diagram is and where it came from
PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses: a guideline for the minimum information a systematic review or meta-analysis needs to include to be reproducible and evaluable. It isn't just the diagram — it's a full 27-item checklist (title, structured abstract, eligibility criteria, information sources, synthesis method, risk of bias, funding...) — but the flow diagram is the piece almost everyone associates with PRISMA because it's the only purely visual one.
The guideline has a history: it started as QUOROM in 1999, became PRISMA in 2009, and was substantially updated in 2020 (Page et al., 2021), with changes to the flow diagram itself. If your review is from 2021 onward, the version you need to declare is 2020, not 2009 — a small detail, but one a Q1 reviewer does check, because it signals whether you've read the current guideline or copied an old template from a colleague.
The four phases of the diagram, box by box
The diagram has four phases, always in the same order, and every box should carry a number you can derive from the one before it.
1. Identification
How many records you found in total. If you only searched databases (Web of Science, Scopus, PubMed, PsycINFO...), that's a single figure. If you also tracked reference lists, contacted authors, or searched grey literature, PRISMA 2020 asks you to separate "records identified from databases" from "records identified from other sources," because these are distinct search processes a journal wants to be able to evaluate on their own.
2. Screening
From the records identified, duplicates are removed first (the same study often turns up across several databases). What's left are the records screened: reviewed by title and abstract, without reading full text, excluding the ones that clearly don't meet inclusion criteria. This is the highest-volume, lowest-detail phase — no need to break down reasons here, just the total excluded.
3. Eligibility
Records that survive screening are assessed at full text. And here's where PRISMA asks for something almost nobody gets right the first time: the reason for every exclusion, one by one, each with its own count. "Did not meet criteria" is not a reason — it's a label that says nothing. A reviewer wants to see something like "Excluded (n = 9): no control group" and "Excluded (n = 6): non-clinical population," not one undifferentiated total.
4. Included
The studies that survive full-text assessment are the ones included in the synthesis. If your review is purely systematic (a narrative synthesis, no meta-analysis), this is the last box. If you're also running a meta-analysis, PRISMA 2020 asks for one more breakdown: how many of the included studies feed into the quantitative meta-analysis, because these don't always match (some studies get described narratively but don't provide the data needed to compute an effect size).
A full worked example, with real numbers
Here's what it looks like in practice, with a fictional but realistic search:
218 records identified from databases
+ 14 records identified from other sources
− 37 duplicates removed
= 195 records screened by title and abstract
− 148 excluded
= 47 articles assessed for eligibility at full text
− 28 excluded: 12 no control group, 9 non-clinical population, 7 insufficient data to compute the effect
= 19 studies included in the synthesis
of which 14 included in the meta-analysis
Notice that each line is a subtraction from the one before it. That's the check a reviewer runs before anything else: if the numbers don't add up (say, if the three exclusion reasons don't sum to exactly 28), the whole diagram loses credibility, and with it, confidence in the rest of your methodology.
The mistakes that come up most
The arithmetic doesn't add up. By far the most common. It happens when the diagram gets built by hand, weeks after the actual screening, reconstructing from memory numbers that should have been logged as they happened. The fix isn't "be more careful": it's not relying on memory and computing the derived boxes automatically from the ones you actually recorded at the time.
Exclusion reasons left undifferentiated. I mentioned it above, but it's worth repeating because it's the second most common failure: a single "excluded for not meeting criteria (n = 28)" with no further detail. PRISMA requires the breakdown by specific reason at the eligibility phase, not before.
Mixing the 2009 and 2020 versions. Copying a template from an article published before 2020 (with fewer boxes and no databases / other-sources split) while stating in the text that you followed PRISMA 2020. A reviewer who knows the guideline spots it at a glance.
Not stating whether the review was pre-registered. If your review is genuinely systematic (not just a meta-analysis of studies you already knew about), registering it on PROSPERO before you start searching is what separates a systematic review from a narrative review with pretensions. It's stated in the abstract and the method, and it's one of the first things checked.
How to build it
There are several ways to build the diagram. The official PRISMA site offers Word and PowerPoint templates you can fill in by hand, box by box, editing text and repositioning arrows. It works, but it's slow and it's exactly where arithmetic errors creep in: nothing warns you when a number doesn't match the one before it. Draw.io and Excel are two other common options, with the same underlying problem.
The alternative I use myself, and have left open and free, is the PRISMA flow diagram generator: you give it the numbers that can't be derived from others (identified, duplicates, excluded at screening, full-text exclusion reasons) and it computes the rest of the boxes automatically, so the subtraction always checks out. It also generates the text for your Methods section from those same numbers, and you download the diagram as a high-resolution PNG with no watermark, in English or Spanish depending on what your manuscript needs.
What else PRISMA asks for besides the diagram
The diagram is the most visible piece, but PRISMA 2020 has a full 27-item checklist (available at equator-network.org, alongside the rest of the reporting guidelines by design type). What a reviewer checks first, beyond the diagram itself: the title identifying the work as a systematic review and/or meta-analysis, a structured abstract with objectives and limitations, eligibility criteria specified and justified before the search, the synthesis method (which model, which software, which heterogeneity), risk-of-bias assessment for every included study, and a statement of funding and conflicts of interest.
Frequently asked questions
Is the PRISMA diagram required for any meta-analysis?
Yes, if your meta-analysis is built on a systematic literature review (the vast majority of cases). If you're combining a handful of studies you already knew about without a systematic search process, that technically isn't a systematic review and you wouldn't need the full diagram — but say so explicitly in the manuscript, because otherwise a reviewer assumes it and flags its absence.
What if two exclusion reasons overlap for the same study?
Assign it to the first reason that applies, following a priority order you decide and document yourself. What matters is that each excluded study counts under a single reason, so the sum of reasons matches the total excluded at full text exactly.
Can I have more than one "other source" of identification?
Yes — reference lists, author contact, grey literature, trial registries. It's common to sum them all into a single "other sources" figure for the diagram, and detail each one separately in the method text if the number of studies each route contributed matters to your readers.
Does the diagram change if my review has no meta-analysis?
The structure is the same; the only thing omitted is the last line (how many of the included studies feed into the quantitative meta-analysis), because there is no meta-analysis to speak of. Everything else — identification, screening, eligibility with exclusion reasons, included — is reported the same way.
Can I use the same diagram for a scoping review?
Not exactly: there's a dedicated extension, PRISMA-ScR, with a diagram and checklist adapted for scoping reviews, which don't require risk-of-bias assessment or quantitative synthesis in the same way. If your work is a scoping review, use that extension and state it in the title.
Does the rest of the meta-analysis hold up to review?
The diagram is the first step. The free kit covers the rest: study selection, forest plot, heterogeneity and the APA 7 write-up, with the chart tools linked inside.
See the meta-analysis kit →Once the diagram is settled, the natural next step is the forest plot with the pooled effect, and checking heterogeneity across studies (I², Q, τ²) before deciding between a fixed- or random-effects model. If you also want to check for publication bias, the funnel plot is the usual complement. And once the full manuscript is written, the paper reviewer is free and tells you what a Q1 reviewer would object to before the journal does.