Spend your first graph on a paper you know well
Don't start with the paper you're unsure about. Paste in a paper you have already read and could argue about, and look hard at what comes back. You will be able to tell straight away whether the map is placing things sensibly.
Why this one: Circle size here is a citation count and circle colour is a publication year — two fixed encodings you have to learn to read, and the only way to learn them is on papers whose vintage and standing you already know.
Tweak: If the result looks wrong rather than unfamiliar, check you clicked the right paper in the autocomplete list — two papers can share a title.
Read the Prior Works list before you read the graph
Open the Prior Works tab. It lists the papers most often cited *by* the papers on your map, which in practice means the foundational work everyone in that corner of the field builds on. Click one and the map highlights every paper on it that cites that work. The list can be downloaded if you want it out of the browser.
Why this one: Prior Works is computed from the graph you just built rather than from a global citation count, so it surfaces what is foundational for this specific cluster — not what is famous overall, which is what a Google Scholar 'cited by' number gives you.
Use Derivative Works to find the review you should have read first
Open the Derivative Works tab. It lists papers that cite many of the papers on your map at once, which usually means either a review of the whole field or recent work at its edge. If you are new to the topic, one good review here saves you a week of reading.
Why this one: Because it ranks on how many of your graph's papers a later work cites, it reaches field-level reviews that a keyword search for 'review' or 'survey' will miss — the useful ones frequently don't say so in the title.
Build a second graph from the best paper the first one found
Pick the paper on your map that you most wish you had known about, and start a fresh graph from it. Where the two maps overlap, you have found the papers you can be most confident actually matter to your question.
Why this one: Each graph is anchored to its starting paper and recomputed from scratch rather than filtered, so moving the starting paper produces a genuinely different map — and the overlap between two of them is information no single graph contains.
Tweak: Connected Papers also builds multi-origin graphs from several papers at once, on every plan including free, if you would rather combine than compare.
Check the build date before you trust a graph someone sent you
Every graph carries the date it was built, with a Graph History panel listing earlier versions of it. If someone shares a Connected Papers link with you, look at that date first. If the site offers you a rebuilt version, take it — the database behind the map moves, and so do the citation counts that set circle sizes.
Why this one: Graphs here are saved objects with their own URLs and version histories rather than live queries. That is what makes them shareable, and it is also why a stale one looks exactly as authoritative as a fresh one.
