PRISMA Flow Diagram Generator for Systematic Reviews

A PRISMA flow diagram is the figure in a systematic review that reports how many records were identified, screened and excluded, and how many studies were included, and why full-text reports were excluded. PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Item 16a of the PRISMA 2020 checklist asks authors to describe the results of the search and selection process, "ideally using a flow diagram". The diagram counts three different things:

PRISMA 2020 has four flow diagram templates. Pick by two questions: is this a new review or an update, and did you search only databases and registers, or other sources too?

Mention your sources in the description and the layout follows. The two new-review templates are typeset by code; the two update templates are drawn by the image model, so check every number on those. Scoping reviews usually adapt the same flow, reporting "sources of evidence". Cite the diagram's source as: Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71.

Reviewers check that every box follows from the one above it. Run these subtractions on your screening log before you generate. Vizcept sets your numbers as you wrote them and does not balance them, so a slip in your log shows up in the figure too.

1. Records identified − records removed before screening = records screened
2. Records screened − records excluded = reports sought for retrieval
3. Reports sought − reports not retrieved = reports assessed for eligibility
4. Reports assessed − reports excluded (all reasons added up) = reports of included studies
5. Studies included in the previous version + new studies included = total studies included in review (updated reviews only; the same holds for reports)

Three rules keep the boxes consistent:

• Count each excluded report once, under its primary reason, so the reasons add up to the reports-excluded total.
• Write down each database's count before you remove duplicates. Some citation managers remove duplicates as you import, and the PRISMA 2020 template asks for per-database counts where feasible.
• If an automation tool excluded records, give the human and the automation counts separately.

A worked example:

If one study has several reports, the number of included studies can be smaller than the number of reports — say both. Why the finished figure still needs checking.

When an image model draws a flow diagram, every box and digit is painted as pixels from your description, and nothing ties a painted digit to the number you wrote. Vizcept splits the job in two:

• Reading: a text model (Anthropic Claude Sonnet 4.6, with OpenAI GPT-5.2 as a fallback) reads your counts and returns which number belongs in which PRISMA 2020 box, as structured data.
• Drawing: code lays out the template from that data and sets every word and digit in a font. No part of a typeset figure is painted by an image model.

Three rules follow:

1. Every count above zero must appear in your text. If the model returns a number that is not in your description (say it subtracted 312 from 1,847 to get 1,535), the number is removed and the box shows a blank "(n = )". The check looks for the number anywhere in your text, not next to that box, so a worked-out count that equals another number you wrote gets through. A 0 stays only when your text has a 0 or says none or all.
2. Sources go where you screened them. Records screened together with your database records are listed in the identification box; only sources with their own retrieval and eligibility counts get the other-methods column.
3. The arithmetic is yours. The figure shows your numbers even when they do not add up; it does not balance them.

The image model draws the whole figure instead, so check every number on it, when:

• the layout is outside the two new-review templates: an updated review with a previous-studies column, the PRISMA 2009 four-phase layout, or a flow that is not a literature search, such as a cohort's sample selection;
• your description contains a character the fonts cannot set, such as Korean, Arabic or Thai script, or an emoji (Chinese is supported);
• a single box would list more than 16 sources or reasons;
• you attach an image, such as a sketch or an existing figure, before generating;
• reading, layout or saving fails partway, and the run is handed to the image model;
• you change a finished figure with Edit with AI.

The result does not yet show which way a figure was made. More on how Vizcept draws each kind of figure.

Yes, and you own the figure. Two things to do before you submit:

• Disclose the tool. A text model read your description to place the counts, so treat the figure as AI-assisted even though code drew it. Elsevier, Wiley and Taylor & Francis allow AI help with explanatory figures such as flow diagrams if you disclose it, and several want the disclosure in the caption. Springer Nature does not permit "creating content from text prompts" but allows "diagrams derived from author-developed content", so ask the editor. See the publisher policy table and AI disclosure statement generator.
• Cite PRISMA 2020. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71.

Writing a thesis? The conceptual framework generator covers the framework chapter too. Reporting a randomized trial rather than a review? The CONSORT flow diagram generator typesets that figure from your participant flow the same way.

No. A PRISMA flow diagram reports how records moved through a review. A quasi-experimental design diagram shows treatment and time: a comparison group, an interrupted time series, or both. Reporting flow is not the design. Draw the design on the quasi-experimental designs guide, and keep this tool for the review’s counts.

The six common ways to make the figure differ in what you type and who places the numbers.

The Word templates and the Shiny app place each number exactly where you type it, with no AI involved. Vizcept sets the digits from your text but a model decides which box each goes in, so read each box once. Prices checked 19 September 2026; SciDraw AI checked 25 October 2026.