I've been testing Iris.ai against my usual manual literature review process for a SaaS support topic. My goal was a seamless flow from finding papers to having them ready for a manuscript.
I start with a broad research map in Iris.ai to find initial papers, then use the context filters to narrow down. The key for me is exporting the final paper list to a CSV. I clean the data a bit (mostly removing extra columns) and import that CSV directly into Zotero. From there, Zotero handles the citations and PDF management. It cuts out hours of manual entry.
Has anyone else tried a similar pipeline? I'm curious if there are better ways to handle the data export step, or if other reference managers work more smoothly with Iris.ai's output formats.
Nice setup! I've used a similar CSV export route with Zotero, but I found the column mapping can be a bit finicky depending on Iris.ai's export template. One gotcha I ran into: sometimes the "DOI" field doesn't import cleanly if it's in a URL format instead of just the identifier, and Zotero fails to fetch metadata. I ended up adding a quick Make.com scenario to reformat that column before the import.
I'm also curious about reference managers with direct API support. I know some folks pipe Iris.ai results into a Notion database using their API, then use Zotero's web connector later. It adds a step but gives more control over the data structure. Have you ever hit the DOI import issue?
Integration Ian
Your CSV export method is fundamentally sound, but I'd recommend formalizing the data cleaning step into a reproducible script. The column structure in Iris.ai's exports can shift between versions, and a manual process becomes fragile over time. I use a small Python script with pandas to standardize the CSV, ensuring the DOI column is stripped to just the identifier and that author names are in a Zotero-friendly "Last, First" format before import. This adds a minute to the pipeline but guarantees consistency and eliminates the manual tweaking you mentioned.
While Zotero's CSV import works, I've found its direct connector for DOI/PMID/arXiv IDs to be more reliable for metadata fetching. My adapted workflow exports the list of DOIs from Iris.ai, then uses Zotero's "Add Item by Identifier" in batch. This bypasses CSV field mapping issues entirely. You do lose the abstracts and keywords from the Iris.ai export, however, unless you supplement them later.
Has your manual cleaning process identified any other problematic fields besides extra columns? I've occasionally had to handle journal name abbreviations inconsistently in the source data.
The script-based approach is the right call. I've been doing something similar with a Python workflow that normalizes the CSV and also queries the Crossref API for any missing metadata before the Zotero import. This recovers the abstracts and keywords that the DOI batch import would otherwise drop.
One field that consistently causes trouble in the Iris.ai export is the journal name. They sometimes use the full title, sometimes an abbreviation, and occasionally a different abbreviation than what Zotero expects. My script runs a simple lookup against a curated list of common journal abbreviations (e.g., "J. Am. Chem. Soc." vs. "Journal of the American Chemical Society") before writing the cleaned CSV. It's not perfect, but it catches maybe 80% of the mismatches.
Have you found a reliable way to handle the case where Iris.ai exports a single record with multiple DOIs concatenated? I've seen that in some of their topic map exports, and it breaks the batch import unless you split them.
every dollar counts