Elicit

Elicit is a specialist AI research assistant built around scientific literature rather than general web chat. The current product combines semantic and keyword search across more than 138 million academic papers with Research Agent, cited Research Reports, a dedicated Systematic Review workflow, screening and structured extraction. Its API v2 also exposes search, reports and end-to-end systematic-review sessions, while MCP makes the same research capabilities available inside compatible agent clients. ¹²³

The systematic-review surface is unusually deep for a general-access AI research product. Current API limits reach 5,000 screened results on Pro, 20,000 on Scale and 40,000 on Enterprise, with configurable abstract screening, full-text screening, extraction and report generation. The product can search academic literature and ClinicalTrials.gov, excludes retracted papers by default in API search, and supports Zotero and uploaded-document workflows. ²⁴

The main constraint is that research automation is not equivalent to methodological completeness. Elicit can accelerate search and extraction, but the reliability of each stage depends on the question, corpus and extraction schema, so high-stakes evidence synthesis still benefits from explicit human checking. ⁵⁶

Elicit has a meaningful free tier. The current standard yearly pricing surface lists Basic at $0, Plus at $11 per user/month billed annually, Pro at $39 and Scale at $89, with Enterprise custom. Basic includes unlimited search across the academic-paper corpus plus limited Research Agent/Reports usage; Plus expands research usage and exports; Pro adds the dedicated Systematic Review workflow and API access; Scale increases usage and collaboration limits. ⁷

The same pricing page also exposes an Industry pricing variant in which Pro is $49 per user/month billed annually and Scale is $169, so buyers should confirm which pricing surface applies to their account and organization. Enterprise adds larger review limits, SSO/SAML, 2FA, domain controls, custom deployment options and stronger data commitments. ⁷

Independent evidence is mixed in a useful way. A 2026 Research Synthesis Methods study tested Elicit extraction across seven systematic reviews in environmental and life sciences. Seventy of 90 prompt designs exceeded 87% accuracy during prompt development, but performance tended to fall on new articles; repeated runs showed 90% agreement on extracted values while supporting quotes and reasoning were much less stable. The authors concluded that Elicit can complement rather than replace human extraction and that prompts should be independently validated. ⁵

A 2025 Cochrane Evidence Synthesis and Methods study found a different weakness in literature search: across four evidence syntheses, Elicit averaged 39.5% sensitivity versus 94.5% for the original searches, although its 41.8% precision was much higher than the originals' 7.55%. The evaluated product predates Elicit's major 2026 search and systematic-review changes, so those figures should not be transferred directly to the current release, but they remain a reason not to treat automated search as a drop-in replacement for expert information retrieval. ⁶

Extraction evidence is more encouraging. A 2025 proof-of-concept across 43 studies and 602 data points reported 81.4% Elicit accuracy versus 86.7% for human reviewers, with no statistically significant difference; another study reported 92% precision, recall and F1 for Elicit extraction and recommended an AI-assisted second-reviewer workflow rather than unverified automation. ⁸⁹

Security and provenance are comparatively well documented. Elicit reports SOC 2 Type II compliance, encryption in transit and at rest, regular penetration testing and vulnerability scanning, MFA plus SAML for larger organizations, and no training on Enterprise customer data; its privacy policy says servers are located in the United States and describes access, deletion and other privacy rights. Individual Library deletions remain recoverable for 30 days before permanent deletion. These controls are meaningful, but enterprise-specific no-training and zero-retention commitments should not be silently assumed for every consumer plan. ¹⁰¹¹¹²

Current panel coverage is limited. TechCrunch documented Elicit's early literature-review architecture and source-linked approach in 2023, while TechRadar later described it as an academic literature search and synthesis tool. Both predate the current Research Agent/API/Systematic Review surface, so they are useful historical corroboration rather than current performance tests. A G2 page contains one 2022 research-user review but has conflicting product-profile metadata, making the numeric rating unsuitable for this product. ¹³¹⁴

Elicit is best suited to researchers, academics, evidence teams, life-science and healthcare R&D groups, and other knowledge workers whose work begins with published research. It is particularly useful when the workflow needs traceable paper discovery, structured screening/extraction, cited reports or API/MCP integration rather than free-form general assistance.

It is less suitable as the sole search method for publication-grade systematic reviews without validation, and it is more specialized than a general-purpose assistant for coding, writing, everyday web research or broad business automation.