So ResearchRabbit decided to partner with Scite to sprinkle some "credibility" on their citation maps. I’m sure the marketing copy makes it sound like a silver bullet for dodgy references.
Let’s be real. Scite’s whole premise is classifying citations as supporting, contrasting, or merely mentioning. Useful in theory, but have you actually tried to rely on it for a complex, nuanced paper? The AI isn’t reading for context—it’s pattern-matching on citation sentences. I’ve seen it flag a paper as "contrasting" because the authors used a phrase like "in contrast to the method of X, we propose…" which was actually a setup for their own improvement, not a dismissal. Garbage in, slightly sorted garbage out.
Integrating this into ResearchRabbit just gives a veneer of authority to their discovery loop. You’ll get a little badge saying a paper is "supported by 12 citations." Does that tell you if the methodology was sound? If the data wasn’t p-hacked? If the finding has been replicated? Of course not. It’s a metric, and we all know how well the academic community handles the seduction of a nice, tidy metric.
Try it on a topic you know well. Pull up a seminal paper and see what Scite says. Then actually read some of those citing papers. I’d waste a good hour you’ll find at least one classification that’s laughably wrong. So, is it useful? As a very crude, first-pass filter maybe. As something to base a literature review on? You’d be better off actually reading the abstracts.
prove it to me