This site has run a research-summary piece on a single paper most rounds this month. The question that article format can’t answer on its own is the one that matters most to anyone deciding whether to care: is this actually new, or is it the fifth paper doing the same thing with a different acronym? reading-research-papers-tutorial-getting-started-artists covers how to read one paper closely. This is about seeing where it sits among the others.
Connected Papers does one specific thing: feed it a paper (by title search, arXiv ID, or DOI), and it builds a force-directed graph of similar work — not papers that cite your paper, but papers that share a strong citation overlap with it, which tends to surface the actual research neighborhood rather than just “things published after this and pointing back to it.”
Step 1: start from a paper you already have
Go to connectedpapers.com and paste in a paper’s title, arXiv ID, or DOI. If you’ve been reading this site’s research-summary pieces, you already have a steady supply — any arXiv ID in a “Related Reading” section works as a starting point.
Step 2: read the graph, not just the list
Each node is a paper. Node size reflects citation count — bigger nodes are more-cited work, which is a rough proxy for influence, not quality. Node color maps to publication year, usually on a gradient, so you can see at a glance whether your paper sits in a cluster of recent work or one that’s mostly citing decade-old foundations. Edge thickness reflects how strong the citation overlap is between two papers — a thick edge between two nodes means they draw on substantially the same prior work, not just that one happens to cite the other once.
Your starting paper sits at the center, highlighted. Everything radiating outward is similar by that citation-overlap measure, closer nodes being more similar.
Step 3: use the Prior Works and Derivative Works lists
Below the graph, Connected Papers generates two separate reading lists: Prior Works, the most-cited papers that your paper’s reference list draws from, and Derivative Works, newer papers that cite your starting paper. This is the fastest way to answer “what is this building on” and “has anyone done something with this since” without manually chasing citation chains through each paper’s own bibliography.
Step 4: know the free-tier limit before you rely on it daily
The free plan allows five new graphs per month — plenty for occasional spot-checks on a specific paper, not enough if literature review is a daily habit. The paid Academic tier removes that cap; pricing has moved around across reporting (figures from roughly $6–$8/month billed annually show up across third-party listings), so check the current price on Connected Papers’ own pricing page before committing rather than trusting any number repeated secondhand, including this one.
What it won’t tell you
A citation-overlap graph is not a relevance or quality filter. A paper can sit in a dense, well-connected cluster and still be wrong, retracted, or superseded — the graph shows you where a paper sits in a conversation, not whether anyone in that conversation is right. Pair it with actually reading the “Comments:” and “Journal reference:” fields on arXiv itself — a paper’s position in a citation graph and its publication status are two separate questions, and conflating them is how “well-connected” starts getting mistaken for “settled.”