Connected Papers (connectedpapers.com) is a free, web-based literature-discovery tool that builds a single visual graph of papers related to one “origin” paper a researcher chooses. Rather than a ranked list of search results or a continuously expanding citation network, Connected Papers returns a one-shot, force-directed graph — drawn from Semantic Scholar’s open paper corpus — that clusters topically similar papers together and pushes dissimilar ones apart, giving a researcher a fast visual sense of a field’s shape before reading a single abstract.
This guide covers what Connected Papers actually is and how its similarity graph is built, its core features (Prior Works and Derivative Works views, the list view, multi-origin graphs, and saved-paper history), how it specifically differs from Research Rabbit — another graph-based discovery tool CASRAI covers separately — typical research-administration and researcher use cases, and its honest limitations.
What Connected Papers Is
Connected Papers launched as a side project built by a small team who, in their own telling, had spent years frustrated by the manual work of literature review and exploration. The tool takes a single paper a researcher already knows is relevant — the origin paper — and analyzes a pool of roughly 50,000 candidate papers around it to find those most similar in subject matter, then renders the result as an interactive graph: each node is a paper, node size reflects citation count, node color reflects publication recency, and edges connect papers the algorithm judges similar.
The underlying data comes from Semantic Scholar’s open paper corpus, the same academic graph that powers Semantic Scholar itself — Connected Papers is a separate product built on top of that data, not a Semantic Scholar feature. Because the graph is generated fresh from one (or a few) origin papers each time, using it feels less like searching a database and more like asking, “show me the neighborhood this paper lives in.”
How the Similarity Graph Actually Works
Connected Papers does not rank papers by direct citation count or publication date. Its similarity metric, as described by the tool’s own creators, is based on co-citation and bibliographic coupling (also called co-referencing): two papers that are frequently cited together by other papers, or that themselves cite a highly overlapping set of references, are presumed to be about a closely related subject — even if neither paper ever cites the other directly. This is a well-established approach in bibliometrics and information science, distinct from simply tracing who-cites-whom.
The practical effect is that a Connected Papers graph can surface papers that are conceptually close to the origin paper but sit outside its direct citation chain entirely — something a simple forward/backward citation trace would miss. The exact weighting formula behind the similarity score is not published in full technical detail on the tool’s public-facing site, so treat the graph as a strong discovery aid rather than an auditable, reproducible ranking.
Core Features
Prior Works and Derivative Works views
Alongside the main similarity graph, Connected Papers generates two curated lists from the same underlying paper set:
- Prior Works lists papers that are commonly cited across the graph’s papers — in effect, the shared ancestral literature the field builds on, typically surfacing seminal or foundational papers.
- Derivative Works lists papers that commonly cite the graph’s papers — the shared descendant literature, typically surfacing more recent state-of-the-art papers, reviews, or meta-analyses that build on the same foundation.
Both lists are a curated subset of the same similarity-graph computation, not an independent, exhaustive trace of every paper that cites or is cited by the origin — a meaningful distinction from how a citation-chaining tool like Research Rabbit handles the equivalent idea (see the comparison below).
Graph view, list view, and multi-origin graphs
Every graph can be viewed either as the interactive node-and-edge visualization or as a sortable, filterable list — useful when a researcher wants to scan titles and metadata quickly rather than navigate the visual layout. Since 2022, Connected Papers has also supported multi-origin graphs: a researcher can add a second, third, or additional origin paper to an existing graph, and the tool re-computes similarity against all selected origins at once. Each added origin refines the search rather than simply merging two separate graphs, which is useful when a research question sits at the intersection of two distinct bodies of literature rather than squarely within one.
Saved papers and account history
Creating a free account (via email or a Google sign-in) lets a researcher save individual papers and entire graphs for later reference and revisit a history of previously generated graphs, rather than losing that work at the end of a browser session. This is closer to a lightweight, ongoing reading list than to Research Rabbit’s persistent, collaboratively shareable Collections — it saves what was found, but does not, on its own, keep monitoring for new matching papers the way Research Rabbit’s Monitor feature does.
Reference-manager and Paperpile integration
Papers surfaced in a Connected Papers graph — including those in the Prior Works and Derivative Works lists — can be exported into common reference managers such as Zotero, Mendeley, and EndNote via each paper’s metadata, and Connected Papers has a dedicated browser-extension integration with Paperpile that adds a one-click “save to Paperpile” button directly on graph and paper-detail views. As with Research Rabbit’s equivalent integrations, this lets discovery inside Connected Papers feed directly into whichever citation-management tool a researcher already uses, instead of requiring bibliography data to be re-entered by hand.
Free tier and the paid Academic/Business plans
Connected Papers is free to start using with no account required. Account-free and free-account use is capped at five new graphs per month; a paid tier — split into an Academic plan (for individual academic, non-profit, or personal use) and a Business plan (for commercial use) — removes that monthly cap and allows unlimited graph generation, at a modest per-month price that is discounted for annual or quarterly billing over a straight monthly rate. Multi-origin graphs, the Prior/Derivative Works views, and reference-manager export are available on both the free and paid tiers — the paid tier’s main value is removing the graph-generation cap, not unlocking otherwise-hidden features.
Connected Papers vs. Research Rabbit: How They Actually Differ
Both Connected Papers and Research Rabbit are graph-based literature-discovery tools built on open bibliographic data, and it is easy to describe them as interchangeable at a glance. In practice they differ in specific, functionally important ways:
- One-shot graph vs. a growing, persistent network. A Connected Papers graph is generated once per origin paper (or origin set) and re-run manually if you want it refreshed; there is no background monitoring of a graph for new matching papers. Research Rabbit’s Collections are explicitly designed to grow over time — new papers can be added, the recommendation set refines as a collection grows, and its Monitor feature actively watches a collection and notifies the user when new matching papers are published.
- What “similar,” “earlier,” and “later” mean. Connected Papers computes one similarity graph per origin using co-citation and bibliographic coupling, then derives Prior Works and Derivative Works as curated subsets of that same graph. Research Rabbit instead offers three separately invokable modes — Similar Work (a content-similarity recommendation, distinct from the citation graph), Earlier Work, and Later Work (each an explicit backward or forward trace through the citation graph itself) — that a user switches between deliberately, rather than getting a single combined view by default.
- Free-tier limits. Connected Papers caps free use at five new graphs per month, after which a paid Academic or Business plan is required for unlimited graphs. Research Rabbit’s free tier is uncapped on searches and collections (with a 50-seed-article limit per search), with its paid ResearchRabbit+ tier raising that seed cap to 300 and adding features like integrity-monitoring alerts rather than removing a hard monthly usage ceiling.
- Multi-paper starting points. Connected Papers’ multi-origin graphs let a user add several origin papers to refine one graph. Research Rabbit achieves a similar effect by building a Collection from multiple seed papers and reference-manager imports, then continuing to add papers to it as the review progresses — a more open-ended, project-scoped model rather than a single refined graph.
- Reference-manager integration depth. Both tools export to Zotero, Mendeley, and EndNote. Research Rabbit additionally supports importing an existing reference-manager library directly as seed papers to start a Collection; Connected Papers’ most tightly built integration is its dedicated Paperpile browser extension, alongside general per-paper export to the other managers.
Neither tool is a strict upgrade over the other. A researcher who wants a fast, visual, one-time map of a field around a single known paper — and doesn’t need ongoing monitoring — tends to reach for Connected Papers. A researcher building an evolving, shareable, collaboratively maintained map of a topic over the course of a longer project tends to reach for Research Rabbit. Many literature-review workflows genuinely use both.
Typical Use Cases
- Orienting quickly to an unfamiliar field. Starting from one known-relevant paper, the graph view gives a fast visual sense of a topic’s major clusters, its most-cited foundational work (via Prior Works), and its current state-of-the-art (via Derivative Works) before committing to a formal search strategy.
- Building a thesis or dissertation bibliography. The tool’s own creators describe bibliography-building for a thesis as a core intended use case — generating a graph from a key paper, then exporting the relevant nodes into a reference manager as a starting reading list.
- Sanity-checking coverage before a systematic search. Running a known landmark paper through Connected Papers can surface adjacent, conceptually related work a keyword-based database search might miss — useful as a supplementary check, not a replacement for a documented, reproducible database search.
- Identifying candidate reviewers or related labs. Because the graph clusters by conceptual similarity rather than raw keyword match, it can surface authors and groups working on closely related problems who wouldn’t necessarily appear in a straightforward keyword search.
Limitations to Keep in Mind
- Not a reproducible, citable search. A Connected Papers graph is an algorithmic recommendation generated at a point in time; re-running it later is not guaranteed to return an identical result, and the exact similarity-weighting formula isn’t published in full. Like Research Rabbit, it does not on its own satisfy the documented, reproducible, multi-database search-strategy requirements of a PRISMA 2020-reported systematic review, scoping review, or meta-analysis.
- Coverage is bounded by Semantic Scholar’s corpus. Because the underlying data is Semantic Scholar’s open academic graph, coverage gaps in that corpus — for older literature, some non-English-language work, or disciplines with thinner open-metadata infrastructure — carry through to Connected Papers as well.
- Free-tier graph cap. Five new graphs per month on the free tier is a real constraint for anyone running many origin papers through the tool during an active literature review; heavier use requires the paid Academic or Business plan.
- A snapshot, not a monitored feed. Without manually re-running a graph, Connected Papers will not surface newly published papers that would now belong in it — unlike Research Rabbit’s Monitor feature, there is no built-in alerting for new matches over time.
- Disclosure practice. Because Connected Papers’ graph is algorithmically generated, any use of it as part of a documented search strategy for a systematic or evidence-synthesis project should be disclosed the same way other algorithmic discovery tools are — naming the tool, the date used, and how its suggestions were verified against primary sources — following the same general discipline covered under AI in literature search.
Frequently Asked Questions
Is Connected Papers free?
Yes, with a monthly cap. Anyone can use Connected Papers without an account, and a free account additionally allows saving papers and graph history. Free use is limited to five new graphs per month; a paid Academic plan (for individual academic, non-profit, or personal use) or Business plan removes that monthly cap for unlimited graph generation.
What data does Connected Papers use?
Connected Papers builds its graphs from Semantic Scholar’s open academic paper corpus. It computes similarity using co-citation and bibliographic coupling — identifying papers that are frequently cited together, or that cite highly overlapping sets of references — rather than ranking by raw citation count or keyword match alone.
What is the difference between Prior Works and Derivative Works?
Prior Works lists papers commonly cited across the graph’s paper set — typically the shared foundational or seminal literature. Derivative Works lists papers that commonly cite the graph’s paper set — typically more recent work, including reviews and meta-analyses, that builds on the same foundation. Both are curated subsets of the same similarity computation, not an independent, exhaustive citation trace.
Is Connected Papers the same as Research Rabbit?
No. Both are graph-based literature-discovery tools, but Connected Papers generates a one-shot similarity graph per origin paper (refreshed manually), while Research Rabbit builds persistent, growing Collections with separate Similar Work, Earlier Work, and Later Work modes and an active new-paper Monitor feature. See CASRAI’s Research Rabbit guide for the full comparison.
Can I export Connected Papers results to Zotero or Mendeley?
Yes. Papers found in a Connected Papers graph, including its Prior Works and Derivative Works lists, can be exported to reference managers such as Zotero, Mendeley, and EndNote. Connected Papers also has a dedicated Paperpile browser-extension integration for one-click saving.
Is Connected Papers good for a systematic review?
It is useful for the early scoping stage — quickly mapping a topic’s foundational and derivative literature from a known paper — but it does not by itself meet the documented, reproducible, multi-database search requirements of a PRISMA 2020-reported systematic review. Treat it as a supplement to a formal, database-specific search strategy, not a replacement for one.
Related CASRAI Resources
Connected Papers is one of several citation-graph tools researchers and research administrators use for literature discovery. See CASRAI’s Research Rabbit guide for the closest comparable tool, the Semantic Scholar guide for the AI-powered search engine and open corpus that Connected Papers’ own graphs are built on, and the Persistent Identifiers & Research Information Systems pillar for the broader discovery and identifier landscape. For how subscription citation databases compare on coverage and cost, see the Scopus vs. Web of Science vs. OpenAlex comparison. Related Dictionary entries: DOI, ORCID iD, preprint, CRIS interoperability, and h-index.
