An AI literature review generator drafts a synthesis of published research on a question: what the field has found, where studies disagree, and what remains open. In Scholarly, the review runs as a Deep Research investigation — the agent reads across the open literature and any papers you attach, then returns a cited draft organized by theme, at the depth you choose.
The hard part of a literature review was never the writing — it's the reading: locating the relevant studies, holding their findings in mind at once, and noticing that two well-cited papers quietly contradict each other. That's the part an investigation agent is good at. It reads more sources than a person can hold in working memory and keeps track of who found what, under which conditions.
The output is deliberately a draft. It gives you the map of the field — the themes, the load-bearing studies, the disagreements — with citations attached, so your time goes into judgment: which findings you trust, which methods convince you, and what your own review should argue.
A review rarely starts from nothing. You usually have a folder of PDFs — the papers your advisor sent, the ones the last proposal cited, the ones you found at midnight. Attach them and the agent treats them as the core of the review, reading each one rather than guessing from titles, and extending the search outward from there.
Because the review lives in your Scholarly workspace, the papers stay useful afterward: chat with any of them individually, pull them into the next investigation, or turn the finished review into slides for a lab meeting or a podcast for the commute.
The agent searches the open web, which reaches a great deal of published research — open-access journals, preprint servers, government and institutional publications — but it is not a subscription database, and it cannot read papers behind paywalls it can't access. For those, upload the PDF and it reads your copy.
And like any AI system, it can misread a study or overstate a finding, which is why every claim carries its citation. Verify the sources you plan to rely on, and treat the draft as the fastest possible route to a review that is still, in the end, yours.
