An AI research platform that has evolved from a literature search assistant into a Research Agent for high-stakes evidence synthesis. It searches 138M+ academic papers, 545K+ clinical trials, plus patents, regulatory databases, biology databases, and web sources, and produces fully cited reports, tables, figures, slides, and documents. Its Systematic Review workflow supports PRISMA 2020 and reports 95% search recall, 97% abstract screening, and 96% data extraction across 994 Cochrane reviews. The Research Agent is available via web, API, and MCP server (Pro and above), allowing it to run inside Claude, ChatGPT, and other MCP clients. Pricing: Free (unlimited paper search/summaries/chat, limited Agent usage), Pro $49/user/mo annual (API/MCP access), Scale (collaboration, Shared Projects, Skills), Enterprise (SSO, single-tenant, no training on customer data). Customers include NASA, NIH, Harvard, Takeda, and Unilever. SOC 2 Type 2. Caveat: independent evaluation found extraction reasoning matches only 30–46% across repeated runs, so it complements rather than replaces human reviewers.

Pros & Cons:

Pros: The breadth of sources is unusually wide for an academic tool—138M papers plus clinical trials, patents, regulatory data, and user uploads in one Agent session, covering evidence that a single PubMed search would miss; the Systematic Review workflow with PRISMA 2020 support is a genuine time-saver for a task that normally takes months, and the published Cochrane benchmark numbers (95% recall, 96% extraction) provide a quantitative reference point; API and MCP server availability from Pro ($49) means research workflows can run inside Claude, ChatGPT, or custom pipelines, which is rare for a specialized academic tool; the customer list (NASA, NIH, Harvard, Takeda, Unilever) and SOC 2 Type 2 report provide enterprise validation.

Cons: **Independent verification of extraction quality is sobering**—UBC testing found supporting quotes and reasoning matched only 46% and 30% of the time across repeated runs, and high-accuracy mode dropped to 10% and 0%, meaning the output requires substantial human validation before use in high-stakes decisions; **the product has restructured around the Agent**, with older Find Papers/Extract Data/Chat entry points retired in September 2026, which may disorient existing users and tutorials; **usage is a single monthly pool that does not roll over**, and larger Agent sessions, reviews, and premium PDF parsing consume it faster, making cost forecasting difficult; **API and MCP access require Pro at $49/user/mo annual**, and SSO/SAML/single-tenant are Enterprise-only; English-only design—Elicit states non-English quality varies widely and consumes usage faster; Enterprise has a no-training guarantee, but lower tiers do not, which matters for teams handling sensitive research.

Who is it for?:Researchers, academics, and analysts conducting systematic reviews or evidence synthesis who want to reduce manual search and screening time while keeping human verification in the loop; life sciences and pharma teams doing indication prioritization, trial design, or patent research that requires cross-referencing literature, trials, and regulatory data; organizations that want research Agent capabilities accessible via API or MCP inside Claude/ChatGPT workflows; enterprises requiring SOC 2 and a no-training guarantee (Enterprise tier). Not suited for non-English primary research (quality degrades), users seeking a simple paper search tool without the Agent-centric workflow, or teams that cannot dedicate human reviewers to validate extraction outputs.