Connected Papers
A visual graph of the papers related to any seed paper
Our verdict
A fast, cheap way to map a field, but it never tells you what the papers actually claim.
Feed it one paper and it draws a similarity graph of related work, highlighting the influential classics and the recent edge papers you would miss by following citation chains. Free for a handful of graphs a month, with a cheap subscription for heavier use.
How we scored it
What we liked
- ✓Instant related-work map
- ✓Highlights prior and derivative work
- ✓Very simple to use
What to watch
- ✕Needs a seed paper to start
- ✕Shows what exists, not what papers say
Key features
- ◆Similarity graph
- ◆Prior and derivative works
- ◆Influential paper highlights
- ◆Year and citation colouring
- ◆BibTeX export
Best for
Alternatives to Connected Papers
Perplexity
AI search that answers with citations you can actually check
Perplexity searches the live web and writes an answer with every claim linked to its source, so follow-ups stay grounded in what's actually online. Pro unlocks stronger models and longer Deep Research reports, but the free tier already covers most day-to-day questions.
NotebookLM
Grounded AI notebook that answers only from your own sources
NotebookLM loads your documents, sites, videos and slides into a model that answers strictly from them, with every sentence linked back to the passage it came from. Audio Overviews turn a stack of papers into a podcast-style discussion, which is genuinely good for skimming.
Elicit
Literature-review assistant that screens papers into structured tables
Elicit reads academic papers and pulls methods, sample sizes and findings into a sortable table, which makes screening hundreds of titles for a review dramatically faster. It works on peer-reviewed corpora, so it is the wrong tool for general web questions.
Consensus
Ask a research question, get an evidence-backed yes or no
Consensus turns a question into a synthesis of what peer-reviewed studies conclude, with a consensus meter and quotes from each paper. It is built for answerable claims ('does X work?'), not open-ended exploration.