What Connected Papers is
Connected Papers is a web tool that takes one paper — a DOI, a title, an arXiv ID — and draws a graph of the papers most similar to it. Nodes are papers, size reflects citation count, colour reflects publication year, and proximity reflects similarity.
The detail almost everyone gets wrong is what “similarity” means here. The graph is not a citation network: an edge does not mean one paper cited the other. Connected Papers uses co-citation and bibliographic coupling — two papers are close if they cite many of the same works, or are cited together by many of the same works. That is why the graph surfaces papers your seed never cited and which never cited your seed, including work published simultaneously in a parallel community. Understanding this changes how you read the output: clusters are research communities, not citation chains.
Why researchers use it
- Field shape in sixty seconds — one seed paper produces a visual map of who is working on what, before you have written a single search string.
- Prior and Derivative works views — a separate pair of lists showing the foundational papers behind the cluster and the recent work building on it.
- Gap detection — generate a graph from a paper you already cite; any large node you do not recognise is a hole in your reading, and it is immediately visible.
- No account needed to look — you can build a graph and evaluate the tool before signing up for anything.
- Export to your reference manager — send selected papers to Zotero or download as BibTeX rather than retyping.
Where it fits in a research workflow
This is a first-week tool, not a systematic-review tool. Its place is the beginning of a project, when you know one or two good papers and do not yet know the vocabulary of the field well enough to search it properly. The output is orientation: which communities exist, which papers everyone shares, and which subfield you have accidentally been reading only half of.
It runs on the Semantic Scholar corpus, so Semantic Scholar is its ceiling — anything missing there is missing here. The natural next steps are different tools: Research Rabbit for iterative, session-based exploration; Litmaps if you want the map to keep monitoring for new arrivals; a real database search in Scopus, Web of Science, or PubMed once you know what to ask for. Treat Connected Papers as the thing that tells you what to search for, never as the search itself.
Getting started
Five minutes. There is genuinely nothing to configure.
- Paste the best paper you know on your topic — a well-cited, on-target one. The graph is only as good as the seed, and a marginal seed produces a marginal map.
- Read the big nodes you do not recognise first. Those are the papers your field assumes you have read.
- Open the Prior Works and Derivative Works lists — the graph gets the attention, but these two lists are where the field’s foundations and its current frontier actually are.
- The step people skip: build a second graph from a different seed in the same field. If the two maps barely overlap, you have found two communities working on your problem without citing each other, which is usually the most valuable thing the tool can tell you.
Connected Papers vs the alternatives
| Alternative | Does it better | Pick it if |
|---|---|---|
| Research Rabbit | Iterative, multi-seed exploration with a persistent collection; free | You want to keep exploring rather than take one snapshot |
| Litmaps | Monitoring — the map re-runs and alerts you to new papers | You want the field to keep reporting to you for months |
| Semantic Scholar | Actual search, TLDRs, and a free API for scripted workflows | You need querying and automation, not a picture |
| scite | Whether citing papers support or contradict a finding | You are asking whether a result held up, not who else works on it |
Connected Papers is the best single-snapshot tool of the four. If you will look more than a handful of times a month, Research Rabbit does the same job free and iteratively.
Cost, licensing, and your data
The free tier allows a small number of graphs per month — five at last check — which is enough for targeted use and not enough for an exploratory month. Paid access removes the limit at a low single-digit monthly price, with academic rates; institutional plans exist.
There is nothing sensitive to upload here: you provide a public paper identifier, and it returns public bibliometric data. That makes it one of the few AI-adjacent research tools with essentially no data-governance question attached. The dependency worth noting is different — the underlying corpus is Semantic Scholar’s, so coverage gaps and metadata errors are inherited, not introduced.
The honest review
Strengths. The co-citation approach is the right idea, and it surfaces work that citation-chasing structurally cannot: parallel literatures, contemporaneous discoveries, adjacent communities using different vocabulary for the same problem. Reading a graph takes under a minute, and the insight per minute is higher than any other discovery tool in this directory.
Limitations. One seed per graph is a hard constraint that becomes frustrating fast — you cannot say “papers like these five”, which is what you usually want. The free tier’s monthly cap is low enough to feel punitive precisely when a new project makes you want to explore. Coverage is thin outside the sciences, and very recent preprints can be absent for weeks. And the visual is persuasive in a way the data does not always earn: a tight, attractive cluster looks like a settled field even when it reflects a small, incestuous citation community.
Verdict. Use it at the start of every new project and at the end of a literature review as a coverage check. Do not pay for it unless you are exploring several fields at once — the free allowance covers occasional use, and Research Rabbit covers heavy use for nothing. The condition that flips the answer is frequency, not features.
Common questions
Is Connected Papers free?
Partly. You get a small number of graphs per month at no cost — five at last check — and unlimited graphs on a low-cost subscription. Building a graph does not require an account, so you can evaluate it before deciding.
Does Connected Papers show citations between papers?
No, and this is the most common misunderstanding. Edges represent similarity computed from co-citation and bibliographic coupling, not direct citation. Two connected papers may never have cited each other at all — which is exactly why the tool finds work that citation-chasing misses.
Which is better, Connected Papers or Research Rabbit?
They answer different questions. Connected Papers gives a better single snapshot from one seed. Research Rabbit is better for sustained exploration across multiple seeds and is free, which matters if you are mapping a field over weeks rather than checking one paper.