How to use AI for a literature review without fake citations
AI tools can speed up the slowest parts of a literature review: finding papers across fields, screening hundreds of abstracts and pulling out what each paper did. They can also invent sources and misstate findings. This guide shows how to use AI for the first part while keeping the review accurate enough to defend in front of a supervisor or examiner.
Where AI helps
- Turning your research question into search terms, including the vocabulary of neighbouring fields.
- Searching several scholarly databases at once rather than one by one.
- Screening many titles and abstracts to find the ones worth reading in full.
- Summarising what each paper studied, how and with what result, so you can compare them.
Where it fails
- General chatbots can produce references to papers that do not exist.
- Summaries can overstate a finding or miss the limitation that matters to you.
- Search coverage varies: paywalled journals, books and non-English work are often under-represented.
- AI cannot judge which papers matter most in your field; your supervisor and your own reading can.
A method that keeps the review accurate
- 01Write the question first
One or two sentences stating what you want to know. It decides what is relevant.
- 02Search real databases
Use tools that search and link to actual records (arXiv, OpenAlex, Semantic Scholar, Crossref, PubMed), so every paper on your list exists.
- 03Read the closest papers yourself
Use AI summaries to decide what to read, then read the key ten to twenty papers in full.
- 04Quote, then paraphrase
Keep the exact sentence that supports each claim with its page number, so you can always show the source.
- 05Check the reference list
Before you share the chapter, check every reference against a bibliographic database.
Turn the review into a contribution
A literature review is not a list of summaries. Group papers by approach, show where they agree and disagree, and end with what is still unknown. That gap is what your thesis or paper addresses, and it is much easier to defend when each side of it is backed by a quoted passage.
Questions
Can AI write my literature review?
It can help you search, screen and summarise, but the review has to reflect your own reading and judgement. Check your university’s policy on AI use, and never include a source you have not checked.
Which databases should I search?
A broad index such as OpenAlex or Semantic Scholar, the published record in Crossref, arXiv for recent preprints in quantitative fields, and your field’s own database, such as PubMed for biomedicine.
How many papers should a literature review cover?
It depends on the field and level. Aim to have found and read the closest recent work in full; a focused set of well-understood papers is better than a long list of skimmed ones.