Elicit

Elicit#

Overview#

Elicit is an AI-powered research assistant that goes beyond Consensus’s quick synthesis by extracting custom data from papers and helping you screen large volumes of literature. While Consensus answers “what’s the consensus?”, Elicit answers “what specific data do I need from these papers?”

Website: https://elicit.com/

Key Difference from Consensus:

  • Consensus: Quick synthesis → “Do X interventions work?” → Yes/No with evidence

  • Elicit: Data extraction → “Extract sample sizes, methods, outcomes from 100 papers” → Custom table ready for analysis

Key Features#

  • 📊 Custom Data Extraction: Define any column, AI extracts from papers

  • 🔍 Bulk Screening: Process 50-200+ papers efficiently

  • 📈 Analysis Tables: Structured data ready for meta-analysis

  • 📝 Paper Summaries: AI-generated abstracts and key findings

  • 🎯 Question Answering: Chat with individual papers

  • 💾 Export: CSV/BibTeX with all extracted data

  • 🆓 Free Tier: Limited monthly usage; paid plans for heavy users

Getting Started#

Visit https://elicit.com/ and sign up. Plans and usage limits change often - check the pricing page for the current free allowance before planning a large extraction.

Elicit Homepage

Main Use Case: Custom Data Extraction#

Perfect for: Meta-analysis, systematic reviews, literature synthesis requiring structured data.

Example: Extract RCT Details#

  1. Search: “randomized controlled trials cognitive behavioral therapy depression”

  2. Add Custom Columns: Click “+ Add column”

    • “Sample size?”

    • “Intervention duration?”

    • “Control condition?”

    • “Primary outcome measure?”

    • “Effect size (Cohen’s d)?”

    • “Dropout rate?”

  3. AI Extracts: Elicit reads papers and fills columns automatically

  4. Review & Export: Check accuracy, download CSV for meta-analysis

Common Extraction Templates:

Medical Research:
- "Inclusion/exclusion criteria?"
- "Primary/secondary outcomes?"
- "Adverse events?"
- "Funding source?"

ML/AI Research:
- "Dataset used?"
- "Model architecture?"
- "Evaluation metrics?"
- "Code available?"

Social Science:
- "Sample demographics?"
- "Data collection method?"
- "Statistical approach?"
- "Effect sizes?"

Integration with Review Buddy#

Recommended Workflow:

  1. Comprehensive search with Review Buddy: python main.py --skip-download writes every paper to results/papers.csv, and the papers that survived filtering to results/papers_filtered.csv (or papers_filtered_ai.csv).

  2. Extract the DOIs for Elicit:

    import pandas as pd

papers = pd.read_csv("results/papers_filtered.csv")   # or papers_filtered_ai.csv
dois = papers["DOI"].dropna()
dois.to_csv("dois_for_elicit.txt", index=False, header=False)
print(f"{len(dois)} DOIs written ({len(papers) - len(dois)} papers have no DOI)")

    Alternatively, upload results/references_filtered.bib directly, or the PDFs Review Buddy downloaded to results/pdfs/.

  3. Upload to Elicit and define your extraction columns (manual step in the web interface).

  4. Download the structured CSV - ready for analysis!

When to Use: Elicit vs Consensus#

Use Case

Tool

Why

“What’s the consensus on X?”

Consensus

Quick yes/no synthesis

“Extract sample sizes from 100 RCTs”

Elicit

Custom data extraction

“Does treatment A work?”

Consensus

Evidence overview

“Compare methods across 50 papers”

Elicit

Structured comparison

“Quick background research”

Consensus

Fast synthesis

“Meta-analysis data collection”

Elicit

Detailed extraction

“Teaching/learning”

Both

Consensus for overview, Elicit for depth

Combined Workflow:

1. Consensus (15 min): Get overview and consensus
2. Review Buddy (30 min): Comprehensive search
3. Elicit (2-4 hours): Extract data and screen papers
4. Traditional reading: Verify and deep read

Tips for Success#

Crafting Extraction Questions:

  • ✅ Specific: “What was the mean age of participants?”

  • ❌ Vague: “Participant demographics?”

  • ✅ Closed: “Was the study randomized? (Yes/No)”

  • ❌ Open: “Describe the study design”

Best Practices:

  • Always verify critical data (sample sizes, effect sizes, p-values)

  • Check AI citations (click to see source text)

  • Start with 10-20 papers to test extraction quality

  • Define all columns before processing to save credits

Warning

AI extractions are not error-free, and their accuracy varies with the field, the question and the paper. Always verify extracted values against the source text for publication-quality work - and report that you used an AI tool and how you checked it.

Resources#

Next Steps

  • Compare with Consensus for synthesis vs extraction

  • See Overview for tool selection guide

  • Explore complete workflows in documentation