Hundreds of papers. One question.A cited first draft, built from real ones.
Ask your research question. IdeaM searches the open scholarly record, puts the real candidate papers in your outline for you to choose from, and drafts a literature review from the ones you keep — organised by question, with an evidence table and a verdict for each, and every citation pointing to a paper you can open and read.
Every researcher knows this pile.
You have a question that matters to your thesis, your grant or your clinic. Somewhere out there are the forty papers that answer it — scattered across databases, behind different search boxes, half of them behind paywalls.
So the weeks go to searching, downloading, skimming abstracts and copying references by hand, and the writing starts from a blank page. And the AI tools that promise to help are known for inventing references that do not exist.
Review Writer does the gathering and the first draft the other way round: it can only cite what it actually found, so you start from real papers and spend your time on the judgement only you can bring.
From a question to a cited draft, step by step.
Open Build, choose Review Writer, and answer five questions. Then you choose the papers — nothing is written until you have.
It asks you first
- Your research question
- The field or literature
- Which years to cover
- The kind of review — narrative, systematic-style or scoping
- Whose name goes on the title page
Then, in order
- 1Search
OpenAlex, Semantic Scholar, Crossref, Europe PMC (which includes PubMed) and arXiv, searched together. Duplicates are merged, papers without an abstract or off the question are screened out, and the rest are ranked by relevance, citations and recency.
- 2You choose
The candidate papers land in your outline, one item per paper with its authors, journal, DOI, citation count and abstract. Delete the ones you don’t want.
- 3Read
Legal free full text is read where it exists — PubMed Central (found from the DOI when an index did not record it), arXiv and every open copy Unpaywall lists — and the abstract otherwise. Each reference says which, and the Methods give the share read in full.
- 4Weigh
An evidence table for every paper — design, population, size, intervention, comparator, duration and findings, taken from its own text. Outcomes that papers name differently are matched into questions, and each question is weighed by study design and size, with a confidence rating — High, Moderate, Low or Very low — and the reasons for it.
- 5Write
One argument, organised by question rather than paper by paper: a bottom line first, then for each question what the strongest evidence shows, why studies disagree, and a verdict — drafted from the papers you kept, and nothing else.
- 6Check
Every citation must match a retrieved paper; anything that does not is removed, and every cited paper is looked up again at its registry. A Citation Check in your outline lists what was removed and any sentence worth checking against its paper.
- 7Deliver
A PDF opens at once, and BibTeX and RIS files for your reference manager are saved beside it.
How a review gets written, and a real one to read
The whole workflow on one page — the sources, your choice, the reading, the citation check — with a real excerpt.
Word for word from the sample
Unedited. The numbers in brackets are the review’s own citations; each points to a paper IdeaM retrieved and confirmed exists.
Compared with no intervention or usual care, intermittent fasting lowers fasting glucose at 6 months (high confidence) [1, 2]. Compared with no intervention or usual care, intermittent fasting has inconsistent effects on HbA1c over 3–6 months (moderate confidence) [1, 3, 4].
Specifically, when comparing IF against an active alternative, one randomised controlled trial demonstrated a benefit when comparing 5:2 MR against metformin and empagliflozin [5], whereas another randomised controlled trial comparing an intermittent energy restriction diet to a continuous energy restriction diet found no difference in HbA1c [9].
Carter, S., Clifton, P. M., & Keogh, J. B. (2018). Effect of Intermittent Compared With Continuous Energy Restricted Diet on Glycemic Control in Patients With Type 2 Diabetes. JAMA Network Open, 1(3), e180756. https://doi.org/10.1001/jamanetworkopen.2018.0756
Read the real thing
Written by IdeaM from one research question — unedited first draft; every citation is a real paper.
The papers and the review, in one outline you control.
Nothing disappears into a black box. The candidate papers sit in your outline with their records, and the review lands beside them, a section per item. Reorder the themes, rewrite a paragraph, delete a paper — and Export makes the PDF again from the outline as it stands, checking every citation again.
- Delete a paper before writing and it is never cited.
- Move a question, merge two sections, or rewrite a verdict in your own words.
- Type a citation number that is not in the reference list, and the next PDF removes it.
- Run Build again with a sharper question, and compare the two lists side by side.
- Literature Review — Endocrinology
- Candidate Papers
- Vitale & Kim, 2020 — The Effects of Intermittent Fasting…one paper
- Carter et al., 2018 — Effect of Intermittent Compared With…one paper
- Does intermittent fasting improve glycemic control…? A Systematic-Style Review
- Abstract
- Bottom Linethe answer first
- Introduction
- Methods
- Evidence Tablea row per paper
- HbA1ca question, with its verdict
- Discussion
- References
- Citation Check
What you take away.
When the review is written, you get the forms a researcher actually uses.
Review article PDF
A title page with the abstract, a bottom line with a summary-of-findings table, the evidence table, a section per question ending in a verdict and its confidence, in-text citations as [n], and a numbered reference list in APA style marked “read in full” or “abstract only”. Systematic-style and scoping reviews add the found → screened → included counts.
BibTeX and RIS
Every cited paper, ready for Zotero, EndNote, Mendeley, JabRef or Overleaf. Also in Export ▸ References (BibTeX) and References (RIS).
The paper trail
A Methods section written from what was actually searched and kept, and a Citation Check that shows every citation removed and why.
Three kinds of review.
A readable synthesis of what a field knows and where it is heading.
A documented search and screening, reported with its found → screened → included counts. Screening is rule-based and done by one person, so it is “systematic-style”, not a full systematic review.
A map of how much research exists on a topic, and where the gaps are.
Written by the free AI on your own computer, or with your own Google AI key. A real twelve-paper run on a MacBook Air took about a quarter of an hour.
Everything around the reading, too.
A review is the tip of a lot of reading and thinking. IdeaM holds all of it.
Your Second Brain
Save papers, notes, talks and web pages, then ask your own material questions and get answers from it.
Live Books
Keep a topic outline current: refresh it from the web when the field moves, and see what changed.
Capture on your phone
A thought after a seminar, a reference someone mentions — Quick Capture saves it in seconds, even offline, straight from your Home Screen.
Outlining built for argument
Drag sections around, zoom into one theme, and see the shape of the whole review at once.
You have the final say
Every AI change to your outline is shown to you first. Snapshots and unlimited undo keep every version safe.
A strong, fully cited first draft — that you shape.
The search, the first reading and the referencing are done, and every reference is a real paper. Read the papers that matter most, check the claims against them, and rewrite in your own argument. The thinking, the judgement and the final word stay yours.
Bring the question you have been circling.
Ask it once. Get back the real papers, and a cited first draft to read, check and make your own.