Hey everyone, I've hit a weird data discrepancy in Iris.ai that's throwing off my literature review workflow and I'm hoping someone else has seen this.
I'm using Iris.ai to map out some foundational papers in my field (revenue operations, naturally), and I'm relying heavily on the citation counts to gauge impact. But I've noticed that for several key papers, the citation numbers in Iris.ai are significantly different from what I pull directly from Web of Science. We're talking differences of 20-30%, sometimes even more. For example, one seminal paper on sales process automation shows 142 citations in Iris, but Web of Science lists 187.
This is a big deal for my prioritization. I love Iris for the discovery and mapping, but I need to trust the metrics. Has anyone else run into this? I'm trying to figure out where the disconnect might be.
My best guess is that Iris might be pulling from a different source database (maybe Scopus or Crossref?) and there's a time lag or coverage difference. Or perhaps it's an issue with how they're deduplicating or counting self-citations? I'd love to hear if anyone has dug into the settings or has a reliable method for cross-checking within the tool itself. Which number should I be trusting for a true "impact" assessment?
You've put your finger on one of the more subtle but important challenges in using any research tool. Your guess is almost certainly correct - Iris.ai does not use Web of Science as its primary data source. They aggregate from several open and proprietary databases, and each has its own coverage rules, update schedule, and yes, methods for handling things like self-citations and corrections.
The 20-30% delta you're seeing is, unfortunately, common when comparing across platforms. I'd trust Web of Science for the raw citation number if that's your standard, but the value in Iris.ai is rarely the absolute count. It's the *relative* impact within the map it generates. The discrepancy becomes a real problem, as you've noted, when you're trying to use the number for an absolute threshold.
One approach I've seen others take is to use Iris for the discovery and relationship mapping, then manually verify the citation counts for your final shortlist of key papers directly in your trusted source. It's an extra step, but it reconciles the workflow. Have you noticed if the discrepancy is larger for newer publications? That often points to a time lag in the data ingestion.
Stay curious.