Skip to content
Notifications
Clear all

Guide: Getting high-quality lit reviews with Elicit prompts, screenshots

2 Posts
2 Users
0 Reactions
61 Views
(@data_pipeline_rookie_42)
Reputable Member
Joined: 7 months ago
Posts: 237
Topic starter   [#25442]

Hi everyone. I've been tasked with using Elicit to help with literature reviews for a new analytics project. I'm a bit nervous about this because if I miss key papers or misunderstand a consensus, it could send our data modeling work in a completely wrong direction.

I've done a few basic searches, but the results feel a bit scattered. I know the quality of the output depends heavily on how you frame the prompt. Could you share some specific prompt patterns or templates that have worked well for you? I'm looking for prompts that go beyond a simple keyword search and help map out the actual methodologies or findings in a field.

For example, I tried this for a project on customer lifetime value prediction:

```
What are the most cited machine learning models for predicting customer lifetime value in B2C e-commerce from 2018-2023?
```

But I feel like I'm just getting a list of papers, not a synthesis. What's a better way to structure this? Should I be asking for "contrasting views" or "key assumptions" explicitly?

Also, any tips on using the "follow-up questions" or "search filters" effectively would be really appreciated. Screenshots of your workflow would be incredibly helpful for a newcomer like me to visualize the process.



   
Quote
(@alexc)
Reputable Member
Joined: 4 months ago
Posts: 341
 

Yeah, the key is to ask Elicit to *do* something with the papers, not just list them. Try prompts that ask for synthesis directly.

For your CLV example, something like "Compare the primary modeling approaches for B2C e-commerce CLV prediction from 2018-2023, highlighting key differences in their assumptions and reported accuracy metrics" forces a more analytical output. It works better for spotting trends.

Also, don't ignore the "Summarize" button on each paper result. I'll often pick the top 5 from a search, summarize each into the table, and then ask a follow-up question based on the patterns I see in the summaries. The follow-up questions are way more powerful for building a map than the first prompt. Got a screenshot of that workflow if you want it.


Automate everything.


   
ReplyQuote