ResearchRabbit

Every search starts from seed articles. The search algorithm article says ResearchRabbit builds a candidate list from papers connected to those seeds: one cites the other, they share a reference or citation, they share authors, or their titles and abstracts look semantically similar. It then ranks the candidates by how connected each is to the whole seed set, which is why results shift as you select more papers.

Search settings decide where candidates come from: all connections, references only or citations only, which the guides call Similar, Earlier and Later Work, plus the seeds’ authors or authors of related papers. The AI article states that large language models do not power the core recommendations. That keeps the method explainable, but the tool does not read full texts or write about them; for an agent that reads papers and drafts a cited report, the Undermind profile covers that case.

The database article counts 310+ million articles, drawn mainly from Crossref, Semantic Scholar and OpenAlex and updated weekly. Only articles with open access metadata, meaning a publicly available title, author list, date, abstract and reference list, can be indexed; the paper itself may still sit behind a paywall. A paper can be missing because its metadata is closed, because none of the three providers has it, or because it is too new.

Citation and reference counts can run lower than on Google Scholar, which counts differently, and recent citations can lag. Versions of a paper with different DOIs appear separately. Papers with thin metadata can still be saved and their details filled in by hand.

For most searches it does not. The free tier article lists unlimited searches with up to 50 starting articles, citation maps, collections and subcollections, the Zotero importer, shared collections, BibTeX import and export, and notes. The cap bites when a whole reference list should act as one search, such as a gap check before submission, where RR+ takes up to 300 seeds. The RR+ article lists the filters that come with it:

• Keyword, for a word or phrase in the title or abstract.
• Publication date, to find foundational work or the current state of a field.
• Journal quality by quartile and h-index.
• Open access, so every result has a free PDF.
• Retractions, to exclude or review retracted articles.

RR+ also adds multiple projects, each with its own notes and articles, and Signals alerts.

A search can start from keywords, a DOI, an existing collection, or a BibTeX file exported from Zotero, Mendeley or EndNote. The Zotero importer links an account once and then imports chosen collections without erasing existing ones, from more than one Zotero account if needed. The way back is a file: the sharing guide lists BibTeX, RIS and CSV exports, and the Zotero guide says two-way sync is still being built.

Collections can be shared with a view-only public link or with invited editors and viewers, who also see your notes and any subcollections. If a citation map goes into a paper or slide as a figure, ResearchRabbit notes that many publishers expect the tool to be cited.

The Terms of Service name Litmap Limited, doing business as Litmaps, registered in Wellington, New Zealand. New Zealand law governs, disputes go to arbitration under the European Arbitration Chamber in Brussels, and users must be 18 or older. The data summary in the help center says ResearchRabbit does not sell or expose articles or notes and does not train AI models on them.

The Privacy Notice, shared across Litmap Limited’s services, says AI features run through third-party providers including Databricks and OpenAI for natural language processing and AI insights, and that servers are in the United States. It does not say which ResearchRabbit features, if any, send data to those providers.

1. Pick a topic you know and list ten relevant papers.
2. Use five of them as seeds in a free search, by DOI or through a BibTeX import.
3. Open Similar, Earlier and Later Work and record which of the other five appear in the first results.
4. Search the DOI of each paper that did not appear, to see whether it is in the database at all.
5. Export the collection as RIS and import it into your reference manager to confirm the records arrive complete.

Accept if most held-back papers surface and the export is clean. Reject if papers that are in the database still do not appear from five related seeds. No result is recorded here.