The vendor lock-in point is absolutely valid, but I think the "semi-portable state" assessment of Paperpile's Google Drive output is overly optimistic. Having to manually reconstruct structure from a disorganized folder of PDFs and disparate note files creates its own form of lock-in: data is technically accessible, but the cost of re-integrating it into a new system is often prohibitive. You're trading a formal silo for a de facto one made of chaos.
A structured export is non-negotiable. The real question for SciSpace isn't just if they have an export function, but the fidelity of the exported relationships. Can you get a clean JSON or CSV that preserves the link between a highlighted passage, the note attached to it, and the specific paper? Or is it just a dump of notes and a separate dump of citations, leaving you to manually reassemble context? That's the difference between a migration and a data loss event.
A messy, unmanaged export can be just as paralyzing as a clean silo when you're trying to move a decade of lab annotations. The risk isn't only the platform's existence, but also the maintainability of your own data legacy.
—BJ
> "Which is more critical for a neuroscience w..."
You answered it. It's the writing. The core loop is reading, annotating, and citing. SciSpace introduces friction at two of those three points.
Your benchmark shows Paperpile wins on annotation and ties on the plugin, but simpler and faster is better for 12 people. SciSpace's library feature looks good for your lead on an org chart, but if the daily workflow is clunky, no one will keep the library updated.
The AI is a triage tool, not a writing tool. If your group uses it for discovery, they'll still need to annotate in something usable. Now you're managing two systems.
Exactly. The "proprietary work product" clause in the terms of service is a critical variable. Many platforms treat user-generated text as a license to use, not as a transferable asset. I've seen export functions that strip out all annotations, leaving only the bibliographic metadata, which defeats the purpose of testing the export's *format*.
The pragmatic test isn't just exporting a bib file. It's exporting a project, then immediately importing it into a competing tool to see what data survives the round trip. If the annotations don't make it, you're not just locked into a tool; you're locked into a workflow with no exit path for your actual intellectual work.
prove it with data