How to report AI use in a systematic review or meta-analysis
If an AI tool helped with your review, readers and editors need to know what it did, how you know it worked and what you checked. Three sources set the expectations: the PRISMA 2020 reporting checklist, the 2025 joint position statement of Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence, and the ICMJE recommendations.
This guide turns them into a short list of what to declare and where, with wording you can adapt. It is general guidance, not legal advice, and the policy of your target journal comes first.
What counts as something to declare
The position statement says to declare AI use that makes or suggests judgements. Its examples are deciding whether a study is eligible, appraising studies including risk of bias, extracting data, synthesising data from two or more studies, assessing the certainty of the evidence, and drafting text that summarises the strength of the evidence, its implications or a plain-language summary. AI used only to improve spelling, grammar or structure generally does not need listing, but check the journal's policy.
What PRISMA 2020 asks for
- Item 7, search strategy: the full strategy for every database, which includes any AI-drafted string as it was run.
- Item 8, selection process: how many reviewers screened each record, whether they worked independently and, if applicable, details of automation tools.
- Item 9, data collection: the same for collecting data, with details of automation tools.
- Item 11, risk of bias: the tools used, how many reviewers assessed each study and, if applicable, details of automation tools.
- Item 13d, synthesis methods: the model, how heterogeneity was assessed and the software package used.
Let SynthGap run the first pass
Enter an area. SynthGap proposes candidate topics, checks each against existing systematic reviews, and lists the primary studies behind it. Start on the Free plan: no card needed, and every run shows every topic it finds.
Sign up freeWhat to say about the tool
- Identity: the name of the tool, its version and the dates you used it.
- Purpose: what you used it for and which steps of the review it affected, with a reference to its user guide and any changes you made.
- Justification: why it is suitable, with evidence from published evaluations and how you validated it on your own review.
- Verification: what you checked, and, where practical, the prompts, outputs, data and code, shared in a repository or a supplement.
- Interests: any financial or other interest you have in the tool, and who funds it.
- Limitations: the known limits and possible biases of the tool, and the effect they might have on your results.
Where to put it
Put it where the journal says: the Methods, the Acknowledgements or a dedicated disclosure section. If there is a lot of detail, use a table or a supplement. The ICMJE recommendations say that AI used for writing assistance belongs in the acknowledgements, that AI used for data collection or analysis belongs in the methods, and that chatbots cannot be listed as authors because they cannot take responsibility for the work.
In the protocol too
The position statement asks that the decision to use AI is considered and reported as part of protocol development. A protocol sentence can follow this pattern, adapted to your review: 'We will use [tool, version] for [purpose] at the [step] stage, [as its user guide describes / with these settings]. We consider it suitable because [published evaluations and our own pilot on N records, with the result]. A person will [check / decide] [what]. Known limitations are [limitations].'
A worked example for screening
'Two reviewers screened titles and abstracts independently, with [tool, version] ranking the records. Screening stopped after [N] consecutive irrelevant records, and [M] randomly chosen unscreened records were checked to estimate how many relevant studies remained. In a pilot on [K] records screened by hand, the tool kept [x] of [y] relevant records.' Replace the brackets with your own numbers from the protocol; the figures are yours to set, not ours.
Sources
The figures and rules on this page come from these sources.
- Flemyng E, Noel-Storr A, Macura B, et al. Position statement on artificial intelligence (AI) use in evidence synthesis across Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence 2025. Campbell Systematic Reviews 2025;21:e70074.
- Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71.
- International Committee of Medical Journal Editors. Use of AI by authors. Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals.
Questions
Do I have to report using ChatGPT to polish my English?
Generally not: the position statement says that AI used to improve spelling, grammar or structure does not usually need listing. Some journals ask for any use to be declared, so check the instructions for authors.
Can an AI tool be listed as an author?
No. The ICMJE recommendations say that chatbots and other AI tools cannot be authors, because authorship carries responsibility for the accuracy, integrity and originality of the work.
If SynthGap suggested my topic, do I need to say so?
It is good practice. A topic suggestion made by AI informs a judgement about the review, so mention the tool in the methods or the acknowledgements and describe how you verified the suggestion, for example by searching for existing reviews and counting the primary studies.
Where can I find the RAISE recommendations?
RAISE stands for Responsible use of AI in evidence SynthEsis. The 2025 position statement of Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence, published in Campbell Systematic Reviews, summarises what the recommendations ask of authors. A newer version was in development in late 2025, so look for the latest.
Keep reading
General methodological guidance, not medical advice. SynthGap's suggestions are AI-generated from the published literature, so verify them before registering a protocol.