I've been using Elicit for a few months now, mostly for targeted literature reviews. But lately, I've been trying to use it for those early, fuzzy, exploratory phases of a project—you know, when you're just trying to map out a field? I've found my usual precise queries often give me too narrow a slice.
So, how do you structure a query when you want a *broad* overview? I've experimented a bit, and here's what seems to be working for me:
* **Lead with the core concept, not the relationship.** Instead of "effect of X on Y," I start with just "X." For example, "agentic workflows in software development" instead of "how do agentic workflows improve productivity in software development."
* **Use plural keywords and broader terms.** "Cognitive biases" instead of "confirmation bias," or "model distillation techniques" instead of "specific technique name."
* **Frame it as an open-ended question.** Questions like "What is known about [Topic]?" or "What are the current research directions for [Field]?" seem to pull in more diverse papers than hypothesis-driven queries at this stage.
* **I then use the "Suggest search terms" and "Paper recommendations" features heavily** to branch out from the initial results. It feels less like a single search and more like following a trail of breadcrumbs.
My main pitfall has been getting a list of papers that's still too specific because my first query was accidentally narrow. I'd love to hear how others approach this.
What's your strategy for casting a wide net with Elicit? Do you have a go-to query structure for exploration?
– Amanda
Show me the accuracy numbers.
That's a really solid approach. I find the shift from "effect of X on Y" to just "X" especially useful - it stops the search engine from filtering out papers that examine X from a different angle you hadn't even considered yet.
One thing I'd add to your point about open-ended questions is to sometimes state what you *don't* know. A query like "definitions of [emerging term]" or "competing frameworks for [broad concept]" can be great for mapping conceptual boundaries early on. It helps surface foundational papers that a standard question might miss.
Love that you're using the suggestions feature to branch out. That iterative, follow-the-breadcrumbs method is often how the best exploratory sessions go
Raise the signal, lower the noise.
Stating what you don't know is a clever trick. It works similarly in monitoring - you can't alert on what you haven't instrumented, so you sometimes query for "unexpected patterns" or "metrics without a clear owner" to find blind spots.
Your point about foundational papers is exactly why I avoid the phrase "best practices" in early searches. It filters out the messy, contested papers that actually explain how a field evolved.
Your fancy demo doesn't scale.