Elicit is an AI research assistant for . Its core value lies in integrating paper search, data extraction, systematic reviews and research reports into a traceable and auditable evidence comprehensive pipeline, reducing the time cost of literature research for scientific researchers.

In-depth review of Elicit: an AI evidence synthesis engine for rigorous scientific research

Core parameters and statistics of Elicit

Project	Current Public Information
Product positioning	AI-driven academic literature retrieval and evidence synthesis platform
Delivery form	Web application + REST API v2 + MCP Server
Paper database size	138 million+ academic papers (including conference papers), updated weekly
Clinical Trials Database	545,000+ ClinicalTrials.gov registered trials, updated in real time
Maximum single analysis volume	Search returns up to 1,000 papers, systematic review supports 40,000 papers
Supported platforms	Web (main entrance), API (developers), MCP (clients such as Claude/ChatGPT)
Place of Residence	United States (US)
Core team background	CEO Andreas Stuhlmüller (MIT PhD/Stanford postdoc), COO Jungwon Byun (former Upstart growth leader)
Latest version	API v2.0.0 (2026-07-15), including MCP Server

Product Boundary: Elicit focuses on the automation of the entire process from "research questions" to "structured evidence reports", covering search, screening, extraction and synthesis. It does not provide general conversational AI services, nor is it suitable for casual Q&A in non-academic scenarios. For teams that need to deploy completely offline and customize the underlying embedding model, they need to evaluate the Enterprise solution or use it in combination with internal systems.

Users and market recognition of Elicit

User scale: The official website states "Trusted by over 5 million researchers", covering multiple vertical fields such as pharmaceuticals, academia, medical devices, policy research, consumer goods and industry. Public customer cases include medical researchers working on the Cochrane review system of pharmaceutical companies such as Ultragenyx Pharmaceuticals, as well as teams from universities such as Stanford.

Industry Verification:

• BioASQ Benchmark Lead: Elicit achieves the highest recall on 5,486 biomedical questions, @50 recall of 60.3%, 27% ahead of second-place OpenAlex keyword search (47.4%). This advantage is consistent across all result windows of 10, 20, 50, 100, 200.
• Cochrane Systematic Review Evaluation: An evaluation based on 994 Cochrane reviews shows that Elicit reaches 95%+ level at all stages - search recall rate 95.0%, abstract screening sensitivity 96.9% (close to the level of two-person review), full-text screening sensitivity 99.5%, and data extraction accuracy 95.6%.
• PRISMA 2020 Compliant: The systematic review process is certified according to PRISMA 2020 guidelines, and every decision is traceable and auditable.

Enterprise customer coverage: Public information shows that it has entered large pharmaceutical companies, medical device companies, government policy research institutions and consumer product R&D departments. The Enterprise plan provides enterprise-level capabilities such as SSO/SAML, 2FA, domain authentication, and single-tenant deployment.

Boundary Statement: Specific ARR, number of active paying users and other business indicators have not been publicly disclosed. If the procurement process relies on a clear Tier 1 pharmaceutical company customer list or industry certification (such as SOC 2), you need to request supporting materials directly from the Elicit sales team in the business section.

Elicit’s cost advantage: three-tier billing adapts to different research scales

Elicit adopts a tiered pricing strategy of "free trial → professional subscription → enterprise contract". The core difference lies in system review scale API access and collaboration capabilities.

Price Plan Comparison

Plan	Price	Core differences	Applicable objects
Basic (free)	$0	Limited Research Agent and Report quota, unlimited search/abstract/paper conversation, 1 Alert	Personal exploration, initial research on student thesis
Pro	$49/month (pay $588 annually, save 35%)	Standard Research Agent/Report/SLR quota, screen 5,000 papers, 20 table columns, 135 data sources, 10 Alerts, API access	Independent researcher, graduate student, systematic review performer
Scale	$169/month (pay $2,028 annually, save 39%)	5x standard limit, supports chart extraction, real-time collaboration, 200 data sources, 30 table columns, management panel	Research teams, laboratories, small and medium-sized R&D departments
Enterprise	Business pricing	Custom quota, filter 40,000 papers, 40 extraction columns, SSO/SAML, 2FA, domain verification, single tenant, exclusive CS, unlimited API	Pharmaceutical companies, large academic institutions, medical device companies

C-end/Personal Cost: The free version can meet the needs of paper search and basic abstracts. The unlimited number of searches is its key advantage different from competing products. The Pro plan is the lowest threshold for personal systematic review.

Developer/API Cost: API access is included in Pro plans and above, and monthly consumption caps can be set for additional usage. Search requests have rate limits based on packages (subject to real-time data on the official pricing page).

Enterprise/Privatization Hidden Costs: The Enterprise solution involves integrated development (custom data source access, custom template creation), team training (workflow changes to get started with system review), governance (authority audit, compliance document preparation) and continuous operation and maintenance (single-tenant context management). These human investments often exceed the licensing fees themselves.

Elicit’s main features

• Intelligent Paper Search: Paper retrieval based on semantic embedding, covering 138 million+ academic papers and 545,000+ clinical trial records. Supports keyword + semantic mixed search, automatically excludes retracted papers, and can filter by year, journal section, and research type. Use a custom embedding model to sort content by content similarity rather than pure keyword match frequency.
• Research Report: Enter a research question, and Elicit automatically executes a search → filter → extraction → comprehensive system review-like process to generate a structured in-depth report with sentence-by-sentence citations. Supports three export formats: Markdown, PDF, and DOCX, and each claim can be traced back to the original paragraph.
• Systematic Review: End-to-end review workflow, supporting PRISMA 2020 standards, covering search → abstract screening → full text screening → data extraction → report synthesis, all in one section. Supports dual-track operation of Boolean + MeSH keywords and semantic search, and can be introduced into PubMed, ClinicalTrials.gov and company-owned databases. Reasons for exclusion and original article citation were recorded for each screening decision.
• Research Agent: After the upgrade in June 2026, it supports multi-modal input (RNA-seq, FCS flow cytometry data, microscopic image PDB structure files, etc.), can perform computational tasks such as differential expression analysis, phenotype quantification, and test result interpretation, and return visual results.
• Library: Automatically stores and manages all sources discovered during the search process, and supports Zotero import to facilitate reuse of literature collections across projects.
• Alerts (Research Updates): Automatically push newly published relevant papers based on saved search strategies to avoid manual repeated searches.
• API + MCP Server: Officially released in July 2026, it encapsulates the three core capabilities of search, reporting, and system review into REST API and MCP protocol endpoints, supporting direct calls from clients such as Claude Desktop, Claude Code, and ChatGPT.

Expert View: The core synergy of Elicit lies in the four-stage data recovery of "Search→Filter→Extract→Report" - the output of the previous step automatically becomes the input of the next step, and all decision-making records are uniformly saved in the session, making evidence synthesis no longer a separate manual operation, but an end-to-end auditable pipeline. The opening of API/MCP goes a step further, allowing this pipeline to be automatically orchestrated and called by AI Agents and internal systems, instead of being limited to Web UI operations.

Elicit’s model and version evolution

Main line historical version

Version ID	Date	Core changes
Public Beta	~2023-01	Elicit public beta, focusing on paper search and abstraction
Systematic Review + PRISMA 2020	2026-05-06	Released end-to-end system review workflow, supports PRISMA 2020, and launched Systematic Review API
Research Agent + monthly quota pool	2026-06-30	Research Agent supports multi-modal data analysis, and billing is switched from workflow restrictions to monthly unified quota pool
API v2 + MCP Server	2026-07-15	Open the three REST APIs of Search/Report/SLR, launch MCP Server, and support client integration such as Claude Desktop/ChatGPT
Paper Search Eval	2026-07-17	Released BioASQ evaluation results, Elicit search surpassed 5 competing systems on all windows

Version evolution insights

Elicit's version history clearly reflects its evolution path from "single search tool" to "evidence synthesis platform" to "programmable research infrastructure". 2026 is a year of explosive functional density - a PRISMA-compliant system review was launched in May, the Research Agent multi-modal upgrade and billing model reconstruction were completed in June, and the API and MCP were opened in July. This cadence indicates that Elicit is moving from a SaaS product to a "platform + protocol" architecture that allows AI Agents and internal systems to invoke its evidence synthesis capabilities in a standardized way.

Elicit’s technical advantages

Mechanism: Factored Cognition (decomposed cognition) architecture. Elicit has adopted the "decomposed cognition" route since 2018 - comprehensively dismantling evidence into discrete building blocks such as search, screening, extraction, reporting, etc., and then combining each block after independent verification. This is essentially different from the "end-to-end black box" approach of general-purpose AI tools: general-purpose LLM is prone to problems such as summary bias (amplification of supporting evidence, weakening of contradictory findings), hallucination citations, long-context accuracy attenuation, etc. when doing literature research. However, Elicit's decomposition architecture allows each step to be independently audited and optimized.

Search layer: Use a self-developed dedicated embedding model for semantic retrieval instead of relying on the conversational capabilities of a general-purpose LLM. In the multi-source retrieval strategy, semantic search is responsible for "coverage" and keyword search is responsible for "reproducibility". The two complement each other. The @50 recall rate of 60.3% in the BioASQ evaluation is supported by this hybrid strategy.

Screening layer: In abstract screening (96.9% sensitivity) and full-text screening (99.5% sensitivity), Elicit independently judges each inclusion/exclusion criterion, and adopts a "yes/maybe" collegial mechanism - as long as one criterion is judged as "no", it will be excluded, and the rest will be included in the next stage. It is better to screen multiple times than to miss the screen.

Extraction layer: Achieve 95.6% accuracy in the three major fields of Methods, Participants, and Interventions. Each extraction is marked with the citation position (quote + section) in the original text, rather than just the conclusion. Elicit's hallucination rate was on par with experienced human reviewers in the randomized noninferiority trial (1.0%).

Effectiveness: For end users, it means that the output of Elicit can be directly used for high-stakes decisions (regulatory submissions, drug approvals, clinical guideline development), without having to manually verify each piece of information twice like using general AI tools. In a randomized non-inferiority trial, Elicit achieved accuracy comparable to human reviewers on 5,100 data points.

Applicable Scenarios: Most suitable for systematic review scenarios that have rigid requirements for evidence completeness and auditability. For scenarios where you only need to quickly understand a general overview of a field and do not care about reference granularity, Elicit's capabilities may exceed actual needs, and there is a risk of over-configuration of functions.

How to use Elicit

Entrance	Applicable objects	Typical actions
Web console (elicit.com)	Academic researchers, students, corporate researchers	Register an account → Enter research question → Perform search/generate report/launch system review
REST API (docs.elicit.com)	Engineering Team, Data Scientist	Get API Key → Call /search/papers /sessions/reports /sessions/systematic-reviews endpoint
MCP Server (https://elicit.com/api/mcp)	Claude/ChatGPT User AI Agent Developer	Claude Desktop: Settings → Connectors → Add custom connector → Enter the name "Elicit" and URL
Claude Code	Programming researcher	claude mcp add --transport http elicit https://elicit.com/api/mcp → /mcp → select elicit → Authenticate

API quick call example (Search):

curl -X POST "https://elicit.com/api/v2/search/papers" \
  -H "Authorization: Bearer elk_live_<YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What is the effect of GLP-1 receptor agonists on weight loss,",
    "filters": {
      "year_min": 2020,
      "journal_quartile": "Q1"
    },
    "max_results": 50
  }'

Implementation path: It is recommended to pilot 1-2 review questions first, and use the Basic/Pro solution to verify the recall rate and extraction accuracy; then compare the Elicit output with the team's existing review process in two tracks (the old and the new run simultaneously); after confirming that the quality is up to standard, lock in the cost with an annual subscription, and upgrade Scale/Enterprise to unlock collaboration and API capabilities when needed.

Product Pricing for Elicit

• Free Tier (Basic): Unlimited search, unlimited abstracts, unlimited paper conversations, 1 Alert, limited Research Agent/Report quota. Suitable for personal trials and students.
• Professional level (Pro): $49/month (annual payment $588), new API access to 10 Alerts, system review workflow (5,000 articles), 135 data sources and 20 tables.
• Team Level (Scale): $169/month (annual payment $2,028), 5 times the quota, chart extraction, real-time collaboration 200 data sources, 30 tables, and management panel.
• Enterprise: Business pricing, 40,000 article filtering capability, 40 column extraction, SSO/SAML, 2FA, single tenant, unlimited API, dedicated CS team.

Pricing Features: The unlimited search of the Basic free version is very competitive among similar tools; Pro is the lowest entry threshold for enterprise-level users (API access is bound to Pro and above); starting from July 2026, Research Agent, Report, and SLR will share a monthly quota pool, avoiding the flexibility issue caused by separate billing based on workflow.

Application scenarios of Elicit

• Pharmaceutical and Biotechnology R&D: Use Elicit to perform systematic reviews or generate in-depth reports during new drug target validation, competitive product pipeline analysis, and preclinical safety assessment. Pharmaceutical company researchers report that the upfront research time per asset has been reduced from about 20 hours to a few hours, and the work that originally required a multi-person team to complete in two weeks can be significantly reduced.
• Academic Systematic Reviews and Meta-Analyses: Use Elicit's PRISMA compliance process to complete Cochrane-level reviews, automatically generating PRISMA flowcharts and included study characteristics tables. In a validation of 994 Cochrane reviews, a 95% search recall rate meant researchers rarely missed key documents.
• Medical Devices and Clinical Evaluation: In the preparation of CE mark technical documents and the preparation of clinical evaluation reports (CER), Elicit can automatically extract clinical evidence and generate structured reports, significantly shortening the document preparation cycle.
• Policy research and evidence-based decision-making: Government agencies and think tanks can use Elicit to quickly collect all relevant research under a certain topic (such as "The impact of AI on employment" and "Comparison of carbon emission policies") and generate a summary report with cited evidence.
• Enterprise Competitive Intelligence and Technology Scanning: R&D teams can set up Alerts to continuously track academic progress in specific technical fields, and connect APIs to internal knowledge bases to build automated competitive intelligence pipelines.

Applicable groups of Elicit

• Academic researcher (Master/Ph.D./Postdoc/Professor): Literature review, systematic review, and evidence collection before writing the paper. Elicit's citation transparency and PRISMA compliance enable its output to be embedded directly into the methods section of an academic paper.
• Pharmaceutical company and biotechnology R&D teams: Teams that need to frequently perform evidence synthesis during the drug discovery, clinical development, and medical affairs stages. The audit capabilities and custom templates of the Enterprise solution can be connected to internal SOPs.
• Medical Devices and Diagnostics Company Regulatory Affairs Team: Prepare regulatory documents such as clinical evaluation reports, safety and performance summaries (SSPs). Elicit’s extraction accuracy and audit trail reduce compliance risk.
• Policy Analysts and Think Tank Researchers: It is necessary to cover a large amount of literature and formulate citable policy recommendations in a short period of time. Elicit's Report function can compress multiple people's weeks of research into days.
• AI Agent Developer and Research Infrastructure Team: Embed Elicit’s evidence synthesis capabilities into autonomous research agents and internal analysis pipelines through APIs and MCP Server.

Not Fitting the Boundary: If a team only needs a general Q&A tool to answer simple factual questions ("How many parameters does a Transformer have?"), Elicit is overloaded. If the research scenario does not require citation traceability, or the user's working language is not English (Elicit's semantic search mainly targets English papers), the effect will be significantly reduced. Organizations that require completely offline operations and do not allow data to leave the local network will need to confirm the feasibility of a single-tenant deployment with the Enterprise sales team.

Summary and Outlook of Elicit

Core Competencies: Elicit is one of the few AI research tools on the market that uses "evidence synthesis" rather than "dialogue question and answer" as its core interaction paradigm. Its factored cognition architecture PRISMA compliance process and end-to-end accuracy verified by Cochrane make it irreplaceable in high-stakes scientific research decision-making scenarios. The opening of API and MCP further extends this capability from Web UI to the AI ​​Agent ecosystem.

Current limitations: Semantic search coverage is highly dependent on its internal paper index (~138 million articles), and some preprints in newer or niche fields may not be indexed (but users can upload their own PDFs through BYOData). Data extraction is currently mainly for English papers, and the ability to process Chinese/Japanese/German and other multilingual documents has not been publicly stated. Full-text screening of systematic reviews is limited by PDF accessibility—in the Cochrane review, approximately 42% of included studies were filtered because full text was not available.

Procurement/Adoption Risk Assessment: For teams considering purchasing the Enterprise solution, it is recommended to verify the following terms before signing: ① Whether the data processing terms allow the content of cited papers to be used for model secondary training (officially, Enterprise does not use customer data for training by default); ② Whether the availability commitment and fault response time in the SLA meet the production needs; ③ Whether the rate limit of the API and the billing upper limit of the quota pool are set in line with the expected growth pace; ④ Whether the update frequency and version management strategy for single-tenant deployments are compatible with internal change management processes.

Related tools: khanmigo, quizlet

Version Info

• Elicit API v2 + MCP Server ：Officially released API v2 and MCP Server, opening three programmatic interfaces for search, reporting and system review, and supporting MCP client access such as Claude Desktop, Claude Code and ChatGPT. （2026-07-15）
• Elicit Systematic Review PRISMA 2020 ：The systematic review workflow fully supports the PRISMA 2020 guidelines and releases the Systematic Review API, making the entire process of search, filtering, extraction, and reporting traceable and auditable. （2026-05-06）
• Research Agent + monthly quota pool ：The Research Agent upgrade supports multi-modal data analysis (gene sequences, omics data sets, microscopic images, etc.), and billing is switched from workflow limits to a monthly unified quota pool. （2026-06-30）
• Elicit Public Beta ：The public beta version of Elicit is online, providing AI-driven paper search and abstract services for academic users. There is no official precise date yet. （~2023-01）