Responsible use of generative AI in systematic reviews
Systematic and systematic-style reviews take significant time and effort. So it’s natural to ask: could generative AI help? And if so, can it do so without compromising the quality, transparency and reproducibility that make reviews valuable in the first place?
What is RAISE?
This is where the Responsible use of AI in Evidence Synthesis (RAISE) recommendations come in. Developed by an international group of experts, RAISE offers practical guidance to help researchers think through when and how it may be appropriate to use AI tools in the review process.
Researchers remain responsible
RAISE emphasises a simple principle: researchers remain responsible and accountable for their work. AI tools may assist, but they do not replace scholarly judgement. This means:
- approaching AI outputs critically
- understanding the limitations of tools
- questioning whether tools perform as claimed
- being ready to verify what they produce.
If you do decide to use an AI tool, it’s important to be transparent. This includes being able to justify why the tool was used, how it contributed to the review and what steps were taken to check its outputs. AI systems cannot be listed as authors, and their responses should not be assumed to be accurate without scrutiny.
Active human oversight is essential throughout the entire review process.
Choosing tools carefully
When selecting tools, it is worth looking beyond convenience. RAISE encourages researchers to prioritise tools with clear validation, strong evidence of effectiveness and transparency about how they work. It is also important to consider how much influence a tool has over your findings. Using AI to support minor tasks, like brainstorming search terms or finding related papers using citation-based tools, carries very different risks from relying on it for screening, analysis or interpretation.
Red flags to look for
- limited opportunities for human oversight
- weak or absent validation evidence
- lack of transparency about how a tool works
- potential legal, copyright or data security concerns.
Questions to ask yourself
It can help to ask:
- What is the potential impact if the tool gets this wrong?
- Where and how will I verify its outputs?
- Can I clearly and transparently report what the tool did?
Remember: reviews are intellectual work
Systematic reviews are not just a sequence of mechanical steps. They are intellectual work. As a researcher, you shape the questions, interpret the evidence and situate findings within your field. AI may support parts of the process, but it cannot replace your disciplinary expertise or critical thinking.
Optional reflection: where do you feel comfortable?
You might find it useful to think about where you feel comfortable using AI in your own review work.
Scenario 1:
A researcher uses AI to generate a full draft of a systematic review report.
- Comfortable
- Maybe / depends
- Not comfortable
Scenario 2:
AI suggests screening exclusions, but a human reviewer makes the final decision.
- Comfortable
- Maybe / depends
- Not comfortable
These scenarios provide a starting point for thinking about how AI could fit into your review process.
Reach out for help
If you have any questions or need to discuss further, contact the Library.