Consensus

Consensus (Consensus NLP, Inc.) is an AI-powered academic search engine purpose-built to query a corpus of 200M+ peer-reviewed papers and return synthesized, citation-backed answers. It is not a general-purpose chatbot — the architecture retrieves real papers first, then uses LLMs to summarize only what was retrieved, structurally eliminating fabricated citations. Used by researchers, clinicians, graduate students, policy analysts, and increasingly by technical due diligence practitioners who need fast evidence synthesis on scientific or engineering claims.

Natural-language research queries across 200M+ papers sourced from OpenAlex, Semantic Scholar, PubMed, Crossref, and full-text publishing partnerships with Taylor & Francis, Sage, and the American Chemical Society. Three-tier hybrid retrieval: keyword + semantic (BM25 + embedding) search returning ~1,500 candidates, reranked by query relevance plus research-quality signals (citation counts, study design, journal prestige via SciScore), narrowed to top 20 displayed. Signature Consensus Meter visualizes yes/no/possibly/mixed distribution across top papers for binary empirical questions. Pro Analysis generates LLM-synthesized summaries with inline citations. Study Snapshots extract population, sample size, methods, outcomes per paper. Deep Search reviews up to 1,000 papers for systematic-review-adjacent work. Medical Mode filters to ~8M articles from the top 1,000 medical journals plus ~50,000 clinical guidelines. Threaded conversation allows iterative query refinement. Export to CSV/RIS; citation manager integration with Zotero, Mendeley, EndNote, and RefWorks.

Search-first architecture: LLM summarization runs only after real papers are retrieved, eliminating fake-citation hallucinations; residual risk is misinterpretation of a real paper. Summarization layer uses GPT-4o and custom fine-tuned models for specialized tasks (study design classification, argument tallying). Scholar Agent partnership with OpenAI deployed in Pro and Deep modes. Official MCP server live at https://mcp.consensus.app/mcp (published March 2026, 99.92% uptime on Smithery); supports Claude, Claude Code, ChatGPT, Cursor, and any MCP-compatible client via HTTP transport with OAuth. With an API key, returns up to 20 papers per query with filters for year, human studies, sample size, and journal quality (SJR). Also available as a native ChatGPT app and compatible with ChatGPT Deep Research. REST API documented at docs.consensus.app. MCP server is read-only for search; no write capabilities on the academic search product.

Free tier: 25 Pro Searches/month (top 20 papers each) and 3 Deep Searches/month; access to all features with search volume caps. Premium: $8.99/month (or $108/year) — unlimited Pro Analyses and Study Snapshots. Pro: $15/month (or $120/year) — unlimited Pro Searches plus 15 Deep Searches/month. Deep plan: includes everything in Pro plus 200 Deep Searches/month (pricing not publicly listed; email for quote). Teams: $9.99/seat/month — centralized billing for research groups. Enterprise (200+ users): custom quote via sales@consensus.app. Student discount: 40% off. Institutional access available via LibKey integration for university site licenses. Pricing as of mid-2026 per Consensus help center and third-party review sources.

Citation managers: Zotero, Mendeley, EndNote, RefWorks (auto-citation export). Library systems: LibKey integration for university site license access to paywalled articles (2025–26 academic year rollout, 170+ university partnerships). ChatGPT: native app + Deep Research compatibility. MCP: Claude, Claude Code, ChatGPT, Cursor, Codex, Windsurf/Cascade, and any HTTP MCP client. REST API available with API key. Export formats: CSV, RIS. No native CRM, Slack, or productivity suite integrations documented.

Anonymized data model: only query text is stored, fully anonymized and inaccessible to external parties; no PII collected or tied to user identity. Consensus explicitly does not use user data to train its own or any third-party AI models. Data is not sold or shared with advertisers. Encryption in transit via SSL/TLS. Organization-level data isolation available on enterprise contracts; on-demand deletion of org search data contractually documentable. SOC 2 certification not publicly confirmed on the vendor's own security page — treat as not verified. GDPR and CCPA alignment implied by privacy policy but formal certifications not disclosed. No ISO 27001 reference found.

Founded 2021 in Boston (now headquartered in San Francisco) by Eric Olson (CEO) and Christian Salem (CPO), two Division I college athlete teammates from families of researchers and teachers. Legal entity: Consensus NLP, Inc. Total funding: $45M per the company's own About Us page (as of mid-2026), confirmed by PitchBook at $45.6M. Funding rounds include: pre-seed backed by Winklevoss Capital; seed led by Draper Associates ($3M announced April 2023); Series A of $11.5M led by Union Square Ventures with participation from Nat Friedman, Daniel Gross, Draper Associates, Path Ventures, and Alumni Ventures (August 2024); additional rounds bringing total to $45M. Notable investors: Union Square Ventures (lead Series A), Draper Associates, Winklevoss Capital, Nat Friedman (former GitHub CEO), Daniel Gross. 10M+ users; 170+ university library partnerships. ~48 employees as of April 2026.

Moderate. For a firm's core diligence workflow, Consensus is the fastest way to stress-test scientific claims embedded in a target company's pitch — e.g., validating whether a quantum error-correction approach, a novel energy storage chemistry, or a semiconductor process node claim is actually supported by peer-reviewed literature, or finding the state of the empirical literature on an emerging technology market. The MCP server plugs directly into Claude/ChatGPT workflows, enabling AI-native analysts to pull grounded citations mid-conversation. Sensitivity is Low: only public scientific literature flows through the tool; no firm deal data, LP information, or internal documents are exposed. Fit is not Strong because the tool is fundamentally a research synthesizer, not a deal-flow, portfolio, or relationship management tool — its value is narrow but real for technical diligence on deep-tech targets.

Consensus Meter forces binary yes/no framing on questions that are not empirically binary; less reliable for humanities, qualitative social science, or highly paywalled fields where only abstracts are available. AI-generated study design classifications and quality tags can be wrong. Not a replacement for full systematic review (PRISMA discipline still required). Deep Search (up to 1,000 papers) is paywalled. SOC 2 and formal compliance certifications not publicly confirmed — a gap for institutional procurement. No native integrations with VC workflow tools (CRM, Slack, Notion). Free tier search volume caps are restrictive for heavy users. Journal-quality scoring (SciScore) is optimized for biomedical research and may not translate reliably to CS, economics, or engineering fields central to a firm's thesis.