Guide
How to use AI for a literature review responsibly
AI can speed up the reading and organising parts of a literature review. It cannot be trusted to remember which paper said what unless you structure the work so that it does not have to.
The attribution problem
Put ten papers into one chat and ask for the main findings, and you will get a fluent summary. What you will not reliably get is a correct account of which study each finding came from. Claims blend, and a blended claim is worse than no claim in academic work.
One paper, one branch
The fix is structural. Add your PDFs to the canvas and give each paper its own branch, connected to that paper only. Ask the same questions of each: research question, method, sample, findings, limitations. Because the branch can see just one source, there is nothing for it to confuse.
Themes come from connections
When you are ready to compare, create a branch for a theme and connect the papers relevant to it. Ask where they agree, where they conflict and what none of them address. The set of papers behind each theme is visible on the canvas, which makes the synthesis traceable. Highlight the passages you want to keep and they are saved to a list.
For the mechanics of connecting material to chats, see context types.
What still has to be you
Check every claim you intend to cite against the paper itself. Read the papers that matter in full. Follow your institution’s rules on AI use and disclose it where required. The canvas helps you organise the reading. It does not replace it.