Rolled out ResearchRabbit to 30 researchers - what broke

We just finished a six-month pilot where we rolled out ResearchRabbit to a cross-functional team of about 30 researchers—mix of PhDs, data scientists, and junior analysts. The goal was to streamline literature reviews and discovery. The sales pitch was compelling, and the UI looks great in a demo. Reality, as usual, was different. Our adoption flatlined after about 10 weeks, and we're not moving forward with an enterprise license. I'm here to tell you exactly why, because most of these issues aren't in the brochure.

The core problem wasn't the tool's concept—it's the gap between a single user's casual exploration and a structured, collaborative research workflow at scale. When you scale from one enthusiastic early adopter to a mandated team tool, everything cracks. Here’s what broke, in order of severity:

* **Citation Chaos on Import:** The promise of "just add your papers" falls apart with real-world Zotero/Mendeley libraries. We had multiple instances where complex citations, especially those with preprints, conference papers, or non-standard fields, were either mangled or created duplicate entries. The deduplication logic is not enterprise-grade. One user's 2000-item library became 2300 "unique" items after import, and untangling that manually killed his buy-in immediately.
* **Collaboration is an Afterthought:** The shared collections feature looks fine, but it lacks any permission granularity. It's either view-only or full edit. We had a junior analyst accidentally remove a critical set of papers from a shared collection that a principal researcher was using for a grant proposal. No version history, no undo at the collection level, no audit trail. You're essentially handing everyone a shared Google Doc with no "see version history" option. For B2B or academic teams, this is a non-starter.
* **The "Visualization" Wall:** Yes, the citation graph and maps are visually impressive for a presentation. For actual daily work, they became noise. Users reported that beyond the initial "wow," the graphs didn't help them make decisions faster than a well-sorted list. The interface for navigating the graph is clunky, and you can't easily export or build upon that visualization structure elsewhere. It's a closed-loop insight.
* **Notification Spam & Alert Fatigue:** The "similar work" and new citation alerts are aggressive. Without sophisticated filters (which the tool lacks), users were inundated with emails. The default settings are optimized for maximum engagement, not focused productivity. We had to instruct everyone to turn off notifications entirely, which defeats a core purpose of the tool.

The technical debt of our existing workflows also became a huge factor. ResearchRabbit wants to be your central hub, but:

```plaintext
Legacy Flow: Zotero (manage) -> Word (write) -> Box (share) -> Excel (track)
ResearchRabbit Flow: ResearchRabbit (discover, manage, share) -> ??? (write)
```

The "???" is the problem. Its writing integration is weak. You end up exporting BibTeX or RIS right back into Zotero/Mendeley/EndNote to actually write your paper. That extra hop adds friction and makes ResearchRabbit feel like a detour, not a hub.

**Bottom line:** ResearchRabbit is a powerful discovery aid for an individual researcher who isn't deeply entrenched in a specific citation manager ecosystem and doesn't need robust, fault-tolerant collaboration. For a team of 30 with established, if messy, workflows? It introduced more problems than it solved. The failures are in the implementation details—permissions, data integrity on import, and alert management—not the core idea. We're now looking at building a more modular stack with better-integrated components instead of an all-in-one.

—BW