A platform built on citations, not generation
ResearchRabbit, a literature discovery platform based in Wellington, New Zealand, has built its entire product around mapping citation networks rather than generating text. The company says more than 4.5 million people have used the platform since it launched in August 2021, and its underlying database now spans more than 320 million academic papers, a scale that puts it well into mainstream research-tool territory rather than a niche academic experiment. On November 3, the company introduced Analyse, a new addition that visualizes relationships between papers, authors, and concepts rather than summarizing them for the user directly.
What makes ResearchRabbit notable in November 2026 is what it is not doing. The platform does not currently use generative AI anywhere in its recommendation engine. Its suggestions come from citation networks and the specific papers each researcher selects, with the system adapting to individual reading patterns over time rather than generating new text to describe what it finds. In a product category where nearly every competitor has rushed to bolt a chatbot onto an existing feature set, holding that line deliberately is itself the story, and it is one the company is telling loudly rather than quietly.
What Analyse actually does, and does not do
Analyse is designed to push researchers toward the edges of their current work, surfacing connections and threads a straightforward keyword search or a citation list alone would not reveal. The interface is visual, letting users examine relationships between papers, authors, and concepts directly rather than reading through a generated summary of them. The explicit goal, according to the company, is helping researchers follow unexpected threads without losing track of the question that sent them there in the first place.
ResearchRabbit's head of product, Digl Dixon, drew a sharp line around the company's approach, saying the team is not building a tool that generates answers. Dixon frames an instant answer as an endpoint that can quietly undermine curiosity and the deeper exploration a real literature review requires. That is a direct, public rejection of the dominant product pattern across nearly every AI tool launched in the last two years, where speed to an answer is treated as the primary metric worth optimizing.
An academic voice backing the bet
Mahsa McCauley, associate professor and head of computer and information sciences at Auckland University of Technology, offered the clearest articulation of why this design choice matters pedagogically. She argues that the literature review is where research actually starts, not a formality to get through on the way to the real work, and that exploration needs to be built into the process rather than treated as a box to check before the real analysis begins.
McCauley's broader point is sharper still: because generative AI can now produce plausible answers almost instantly, the scarce skill is no longer finding information, it is knowing which questions are worth asking in the first place. That framing gives ResearchRabbit's product decision an intellectual backbone beyond simple differentiation. It positions Analyse as a tool for building the judgment that makes AI-generated answers usable, rather than a competitor trying to out-answer the AI tools racing past it.
The context: a research population already leaning on AI
ResearchRabbit's launch lands against evidence that researchers are adopting generative AI quickly elsewhere. The article accompanying the Analyse launch cites the AI for Science 2026 study, a joint effort from Fudan University, the Shanghai Academy of AI for Science, and Nature Research Intelligence, which drew on more than 10,480 qualified survey responses across 117 countries and economies. That scale of adoption data suggests researchers are not resisting generative AI broadly, they are using it heavily for some tasks while apparently still wanting a separate, non-generative tool for the specific work of literature discovery.
That split is the real signal for enterprise knowledge management teams. Researchers appear to be reaching for generative AI heavily for synthesis and drafting while still wanting a separate, discovery-oriented tool for the earlier work of finding and connecting sources. Any organization building internal research tools for analysts, scientists, or strategists should treat that as a hypothesis worth testing rather than assuming one AI interface can serve both needs equally well, since the two jobs reward almost opposite interface designs.
What a CTO should take from a research tool's product bet
ResearchRabbit is a free-to-use platform with a paid tier, not an enterprise software vendor, and its total user base, however large, is a fraction of what a mainstream generative AI product claims. But the clarity of its positioning is instructive. The company had every incentive to bolt generative AI onto Analyse, since nearly every competing tool in adjacent categories has done exactly that over the past eighteen months, and it chose not to, publicly, with a named product leader willing to say so on record.
For enterprise technology leaders evaluating research and knowledge tools for their own organizations, the real lesson is that discovery and synthesis are different jobs, and a tool optimized for one will underperform at the other regardless of how much AI gets added to the interface to make it look more capable. Before standardizing on a single AI-powered platform for every stage of research, from literature discovery through final synthesis, it is worth asking whether that platform was actually designed for the stage you are asking it to handle, or whether it simply added a chat window to an existing search tool and called the upgrade complete.
