AI Tools for Literature Reviews: What They Can (and Can't) Do

AI-assisted research tools (Elicit, Consensus, Research Rabbit, SciSpace, Semantic Scholar, and general tools like ChatGPT) have genuinely useful applications in a literature review — and genuine limits that matter for a systematic review specifically.

What they're actually good at

• Discovery and citation mapping — tools like Research Rabbit and Connected Papers visualize citation networks, helping you find related work you might have missed with keyword search alone.
• Summarizing individual papers — quickly getting the gist of a study's methods and findings before deciding whether to read it in full.
• Answering narrow factual questions across a paper set — tools like Elicit and Consensus can extract specific data points (sample sizes, outcomes) from multiple papers faster than manual reading.
• Drafting and editing assistance — general AI tools can help restructure or clarify your own writing.

Where they fall short for a systematic review specifically

• Not a substitute for a systematic, documented search — AI discovery tools don't replace a reproducible, multi-database Boolean search strategy that a reviewer can independently verify.
• Summaries can misrepresent findings — AI-generated summaries of papers have been shown to occasionally overstate, understate, or misattribute findings; they require verification against the source, not blind trust.
• Screening decisions still need human judgment — AI can help triage which papers to look at first, but final inclusion/exclusion decisions in a systematic review should be made and documented by the reviewers, per PRISMA expectations.
• Fabricated or hallucinated citations — general-purpose AI chatbots have a documented tendency to generate plausible-looking but non-existent references. Every AI-suggested citation must be independently verified to actually exist and say what it's claimed to say.

A reasonable approach

Use AI tools to speed up discovery, triage, and drafting — treat every factual claim, citation, and summary they produce as something to verify, not something to trust by default. Disclose AI tool use per your institution's or journal's policy where required.