Using AI for a literature review — safely
AI can take real friction out of a literature review — but it can also hand you a beautifully formatted reference list of papers that don’t exist. The trick is knowing exactly which jobs to delegate and which to keep. The rule of thumb: let AI help you think and organise, but do the finding, reading, and citing yourself.
Where AI helps vs where it fails
| Good use | Dangerous use |
|---|---|
| Brainstorming search terms & synonyms | Asking it for a reference list |
| Explaining an unfamiliar concept | Trusting its summary of a paper you haven’t read |
| Summarising a paper you have obtained | Letting it write the synthesis |
| Suggesting a structure for a section | Pasting its claims in without checking |
A verify-everything workflow
- Ideate with AI — search terms, key debates, candidate concepts (see building a search strategy).
- Search real databases yourself — the actual papers come from indexed sources, not the model’s memory.
- Read what you cite. AI summaries are a preview, never a substitute — they miss nuance and sometimes invert findings.
- Verify every reference against an indexed database before it enters your synthesis matrix. If you can’t find it, it probably isn’t real (see AI hallucinations).
- Write the synthesis yourself. The argument across sources is the intellectual contribution — and the part you must own.
AI is not a systematic search
For a systematic review, an AI chat can’t replace a documented, reproducible search across named databases. It may help draft a search string or summarise included studies, but the search, screening, and appraisal still follow the systematic method — and your PRISMA diagram has to reflect what you actually did.
Get the free AI-in-Research Ethics & Disclosure Pack
Disclosure-statement templates, an acceptable-use checklist, and research-prompt patterns from The AI-Powered Scholar. We’ll email you the download link.
Frequently asked questions
Can AI write my literature review?
No — it invents references and the synthesis must be your own. Use it to find, summarise, and organise; read and cite the real papers yourself.
Where does AI genuinely help?
Search terms, explaining concepts, summarising papers you’ve obtained, structuring a section, and surfacing related work to chase down.
Why not trust AI reference lists?
Models generate citation-shaped text rather than retrieving real records, so references can be fabricated or mismatched. Verify each one.
Does AI replace a systematic search?
No — a systematic review needs a documented, reproducible database search AI can’t provide.