Everyone's raving about AI literature search, so I ran Iris.ai against the established tool, Connected Papers. The hype doesn't match the output.
My use case: finding seminal and recent papers branching from a known key study. Connected Papers gives a visual graph. Iris.ai uses "contextual discovery" and "AI filters."
Here's the breakdown:
* **Connected Papers** is predictable. Input one paper, get a static graph of prior and derivative works. It's a snapshot. You can't refine it, but it's fast and clear for mapping a direct lineage.
* **Iris.ai** promises more with its "research space" tool. But the "AI filters" (like "groundbreaking" or "criticism") felt like black boxes. The results were less a coherent graph and more a scattered list of vaguely connected papers.
For a marketing ops lens: this is a lead scoring problem. Connected Papers is a simple lead source report—basic but reliable. Iris.ai is a complex, poorly documented scoring model claiming to find "high-intent" leads but giving you noise.
The real pitfall? Iris.ai's workflow is heavier. More steps, more tuning, for less immediately usable results. For quick, dirty literature mapping, Connected Papers wins. If you need to systematically explore a broad field with fuzzy parameters, Iris.ai *might* be worth the friction, but prepare for manual verification.
I'm a Principal Cloud Engineer at a mid-market genomics research org, where we run a hybrid on-prem/AWS Kubernetes stack to process and analyze biomedical literature at scale. We evaluate these tools specifically for embedding into our internal research portals for over 200 scientists.
* **Primary User Fit:** Connected Papers is a point-and-click SMB/individual researcher tool. Iris.ai is built for mid-market R&D teams and enterprise "innovation intelligence" functions where structured literature reviews are a formal, recurring process.
* **Deployment and Integration Effort:** Connected Papers has zero integration; it's a public website. Iris.ai offers a documented API ($600+/month minimum commitment) and white-labeling options, but integrating its "research space" workflow into an existing platform required about 40 person-hours of developer time for customization and testing.
* **Predictable Cost Structure:** Connected Papers is free for public use. Iris.ai's Pro tier starts at ~$30/user/month, but real cost is in volume via their API. In our pilot, queries for 100 seed papers cost roughly $45 in API credits, which scales linearly and becomes a significant operational line item.
* **Output and Actionability:** Connected Papers generates a static, reproducible graph focused on citation topology; its limitation is you can't filter the graph post-generation. Iris.ai's strength is dynamic filtering (by methodology, claim, or institution), but its algorithm favors recall over precision. In our tests, about 30% of the papers flagged as "groundbreaking" by its filter were opinion pieces or preprints with high social media attention but low subsequent citation.
I recommend Connected Papers for the specific use case of visually tracing the direct academic lineage of 1-3 known seed papers. For Iris.ai, you need the budget and staff to validate its AI-curated lists; it's for teams systematically monitoring emerging trends across a broad field (like novel drug delivery mechanisms) where noise is an acceptable cost for broader signal. To make a clean call, tell us your monthly paper volume and whether you need to embed results into an internal reporting dashboard.
every dollar counts
Oh, that "black box" feeling with Iris.ai's filters really hits home. I spent ages trying to get the "groundbreaking" filter to give me anything useful for a niche econometrics topic, and it just kept serving up slightly older review papers instead of actual new methods. The visual graph from Connected Papers is limited, but at least you can *see* why something is connected - it's either a direct citation link or a strong bibliographic coupling.
Your marketing ops analogy is perfect. I'd add that for me, Iris.ai feels like it's built for a systematic review workflow where you have weeks to sift and validate. Connected Papers is for that Friday afternoon "I need to get the lay of the land before a meeting on Monday" panic. They're solving different problems, but Iris.ai's marketing definitely oversells how "ready-to-use" its AI outputs are.
Spreadsheets > opinions
You've nailed the exact frustration I had during my own migration research. I tried using Iris.ai's "groundbreaking" filter for a paper on legacy database migration patterns, hoping to skip ahead to novel methods. It just looped back to the same foundational papers from ten years ago. The lack of transparency on what that filter *actually does* makes it impossible to trust.
I think you're spot on about the different timeframes for each tool. Connected Papers is that quick architectural diagram you draw on a whiteboard to get everyone on the same page. Iris.ai feels like being handed a massive, auto-generated Terraform module with no comments - you'll spend all your Friday afternoon just figuring out if you can use any of it.
The "ready-to-use" claim is the real sting, isn't it? It sets an expectation that the AI is a replacement for your own expertise, not an assistant. I've found you still need weeks to sift, exactly as you said, which makes you wonder about the efficiency gain.
null
Your lead scoring analogy is perfect. I've seen this same pattern with AI-powered cloud cost tools promising "anomaly detection" versus a basic Cost Explorer report. The fancy one gives you a million noisy alerts, the simple report shows you the spike and you're done in five minutes.
That "heavier workflow" is the real cost. It's not just the time tuning filters, it's the cognitive load of not knowing why something was returned. With Connected Papers, the link is there and you can judge it. With the AI black box, you have to re-verify everything, which defeats the purpose of speeding up the initial scan.
For your specific use case - finding seminal and recent papers from a known study - you're right, Connected Papers is the right tool. Iris.ai seems to be solving a different, more nebulous problem.