Skip to Main Content

Consensus

An overview of the Consensus.app tool

How a Consensus Search Works

Consensus is designed to find the most relevant scientific papers for your results and ground our AI-generated summaries. Every search goes through three key stages:

Step 1: Cast a wide-net

Consensus begins by scanning its entire database of over 220 million paper to find the most relevant results using a hybrid approach:

  • Semantic search (AI embeddings): Captures the intent behind your search and supports natural language queries.

  • Keyword search (BM25): A traditional method that anchors results to the exact terms in your query for precise keyword matching.

These methods work together to assign a relevance score to each paper by comparing your query to the titles, abstracts, and full text (when available) in the database. This helps surface papers that match your exact words, while understanding what you're really looking for.

Step 2: Refine by quality

Next, Consensus takes the top 1,500 most relevant papers and re-rank them based on key research quality signals:

  • Recency of publication

  • Citation count

  • Journal impact and reputation

This step ensures your results are highly relevant and back by high-quality, credible science.

Step 3: Rank the top 20 papers

In this final step, Consensus ranks the top 20 papers as accurately as possible. This works similar to Step 2, but includes two key upgrades:

  • Relevance is recalculated using a larger, more powerful AI model

  • This high-precision model is applied only to the top 20 papers to ensure efficiency and accuracy

Consensus continues to factor in recency, citation count, and journal impact, so the final list reflects both relevance and rigor.