Consensus reaches 5M users with AI-powered academic search engine that out-indexes Google Scholar on peer-reviewed research

Summary

Consensus has crossed 5 million users with an AI-powered academic search engine built as a dedicated alternative to Google Scholar and PubMed. The product indexes 200 million research papers and applies small LLMs during ingestion to extract enriched metadata — study design, sample size — which then feeds directly into search ranking and filtering. That domain-specific approach, combined with learn-to-rank models trained on user behavior (saves, citations, shares), is the core technical differentiator from general-purpose tools like ChatGPT.

The content access structure operates in three tiers. Fully open-access papers are ingested and displayed without restriction. For paywalled content where publisher deals exist, Consensus uses full text for search and analysis but cannot display it, routing users to the publisher for purchase. Where no deal exists, only abstracts and metadata are used. The company frames publisher partnerships as a traffic-referral arrangement, positioning itself as a demand driver rather than a competitor.

On the question of whether AI can autonomously generate cross-disciplinary scientific breakthroughs — a thesis Elon Musk advanced at the Grok 4 launch, claiming new physics discoveries within two years — Consensus takes a more measured position. Citing Francis Sho's YC talk, the argument holds that LLMs underperform on efficiently applying information across domains, which remains a fundamental limitation. Consensus orients its product around accelerating literature review, the mandatory first step in any experiment, and leaving hypothesis generation to researchers. The strategic bet is that compressing the knowledge-foundation phase creates compounding value for scientific output.

The founder previously built fraud-detection models at DraftKings, identifying professional gamblers by analyzing betting history and demographic data. On AI's impact in sports betting, the view is that better models raise bettor edge in the short term before markets reprice — an arms race dynamic that mirrors the historical cat-and-mouse between sharp bettors and sportsbooks. Being ahead of injury news remains a primary signal that a bettor has inside access, and that signal becomes easier for operators to detect as AI tools improve on both sides.