Using AI to find a systematic review topic: what to trust and what to check
AI tools can search more of the literature, faster, than any person can, which makes them useful for the first pass of finding a review topic. They can also state things that are wrong with complete confidence, invent references and miss studies, so their output is a list of candidates, not an answer.
This guide sets out what AI can reasonably do at this stage, what it cannot, and a checklist that you can apply to the output of any tool, including SynthGap.
What AI can do well
- Search and group a large body of literature by meaning, not only by keywords, so that related studies with different wording are found together.
- Find the systematic reviews that cite a given study, which shows whether its topic has been synthesised.
- Draft candidate questions, with a suggested population, intervention, comparator and outcome, for you to judge.
- Summarise abstracts quickly so you can decide what to read.
What AI cannot do
- Guarantee that a topic is novel. A review may exist in a source or language the tool did not search, or may be registered but unpublished.
- Judge clinical importance. Whether an answer would change care is a human judgement.
- Be trusted with citations. Models can produce references that look right but do not exist or do not say what is claimed.
- Replace a protocol, duplicate screening or a risk of bias assessment.
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 freeA checklist for verifying a suggestion
- Open every study the tool cites, by its PubMed ID, and read the abstract. Confirm that it exists and says what the tool claims.
- Run your own search of PubMed, the Cochrane Library, PROSPERO and Epistemonikos for reviews on the question.
- Check the date of the newest review and whether new studies have been published since.
- Count the eligible studies and check that they are comparable enough to pool.
- Ask a clinician or supervisor whether the answer would matter.
- Write the question as a PICO and register a protocol before you screen.
Reporting the use of AI
Evidence-synthesis organisations, including Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence, have published joint guidance on the responsible use of AI in evidence synthesis, known as RAISE. Journals and the ICMJE recommendations ask authors to disclose the use of AI tools and what they were used for. If a tool helped you choose a topic, say so in the methods or the acknowledgements, and describe how you verified its suggestions. The full list of what to declare is in how to report AI use in a systematic review, and AI tools for systematic reviews covers the later stages of a review.
Questions
Can I trust an AI-suggested topic?
Trust it as a hypothesis. It is a starting point that you test by reading the primary studies and searching for existing reviews.
Does SynthGap check that a topic is novel?
SynthGap searches the published literature and checks each candidate against existing systematic reviews, then lists the studies behind each topic. Its suggestions are AI-generated, so verify them before you register a protocol.
Do I have to declare that I used AI?
Many journals and the ICMJE recommendations ask authors to disclose the use of AI tools. Check the instructions for your target journal, and describe what the tool did and how you checked it.
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.