Elicit AI

Elicit AI is not trying to be another general-purpose chatbot. Its focus is narrower and potentially more useful for researchers: finding academic literature, organizing papers, extracting structured evidence, screening studies, and building research reports around scientific sources.

That specialization is the main reason Elicit stands out. Instead of asking an AI model to answer a research question from general knowledge, users can search academic literature, inspect individual papers, extract information into structured tables, and trace generated claims back to supporting sources.

The platform has also changed significantly in 2026. Its current experience centers on Research Agent, Research Report, Systematic Review, and the Library. Older standalone workflows such as Find Papers, Extract Data, and Chat with Papers have been folded into the newer experience rather than remaining separate starting tools.

This makes Elicit more ambitious than a simple semantic search engine. But it also creates an important expectation to manage: AI-assisted research is not the same as automated scientific truth. Search coverage can be incomplete, full text is not available for every indexed paper, extracted data still needs verification, and systematic-review methodology remains a human responsibility.

What Is Elicit AI?

Elicit is an AI research assistant developed by Elicit Research, PBC. The project originated at the nonprofit Ought before becoming an independent public benefit corporation in 2023.

Its core purpose is helping people work with research evidence. That includes discovering papers, asking questions about academic literature, extracting study characteristics and results, screening papers, organizing collections, and generating cited research reports.

The platform is primarily browser-based. It also offers a browser extension for Chrome and Edge that can help researchers retrieve full-text papers through institutional or publisher access they already have. Eligible paid plans can access API and MCP capabilities as well.

How Elicit Works in 2026

The current product is organized around several connected workflows rather than a collection of isolated AI tools.

Research Agent is the flexible research workspace. It can search for academic papers, work with uploaded material, investigate questions, extract information, and use saved Library collections as context.

Research Report is designed for producing a more automated literature-based report. It gathers relevant sources and synthesizes them into a cited output, making it useful when you want an initial overview of a research question without manually constructing a complete evidence table.

Systematic Review provides the more structured workflow. It supports stages such as gathering papers, screening studies against criteria, extracting data, and preparing a report. This workflow is aimed at users who need more control and repeatability than an ordinary AI answer provides.

Finally, the Library acts as the research collection layer. Papers can be saved, imported, organized into collections, reused as context, and exported for use elsewhere.

Academic Paper Search

Elicit says its research corpus contains more than 138 million academic records, drawing from sources that include OpenAlex, PubMed, Semantic Scholar, and publisher material. ClinicalTrials.gov registrations are also available through relevant research workflows.

That number needs an important qualification: 138 million discoverable records does not mean 138 million full-text papers. It also does not mean every record is peer reviewed or that the corpus contains every scholarly work relevant to every field.

The Research Agent supports natural-language research questions, making semantic discovery easier for users who do not know the exact keywords used in the literature.

For more controlled systematic-review work, Elicit also supports keyword and Boolean searching, multiple search tabs, PubMed, ClinicalTrials.gov, uploaded papers, and Library sources. Advanced search syntax can narrow results using fields such as title, author, journal, DOI, PubMed ID, open-access status, and other criteria.

This combination is useful because semantic and traditional searching solve slightly different problems. Semantic search can surface papers that discuss a concept using unexpected terminology, while Boolean and database-specific strategies are more reproducible for formal evidence reviews.

Can Elicit Replace Google Scholar or PubMed?

Not safely for every research workflow.

A 2025 independent evidence-synthesis study using an earlier version of Elicit found substantially lower search sensitivity than the original searches used in four review cases, although Elicit also found some unique relevant studies and showed higher precision. The researchers recommended using it as a supplement rather than a replacement for traditional searching.

Elicit has continued evolving since that evaluation and has published stronger results from its own later evaluations. However, those studies use different software versions, datasets, and methodologies, so they should not be combined into a simple claim that the current system always achieves a particular level of recall.

For exploratory research, semantic discovery can be extremely convenient. For an exhaustive systematic review, medical evidence search, or other work where missing a study can materially affect the conclusion, researchers should still build a reproducible search strategy and consider the appropriate specialist databases.

Full-Text Papers and PDF Access

A paper appearing in search results does not guarantee that Elicit can access its full text.

When full text is available, the system can use more detailed evidence than it can obtain from metadata or an abstract alone. Users can also upload PDFs they are legally entitled to access.

The browser extension can help retrieve additional full-text papers using institutional or publisher access that the researcher already has. According to Elicit, authentication credentials remain on the user’s device.

This does not provide a legal workaround for publisher paywalls. The researcher still needs legitimate access to the material.

It is also important to distinguish an answer based on an abstract from one based on a complete paper. Methods, limitations, subgroup details, numerical results, and qualifications may not appear in the abstract.

Working With Uploaded Files

The Research Agent can work with up to 20 uploaded resources in a session, including supported formats such as PDFs, documents, spreadsheets, presentations, and images.

For research papers with complex tables, mathematical notation, or difficult layouts, eligible paid users can enable premium PDF parsing. That can improve extraction from challenging documents, but it also consumes additional usage and does not guarantee that every scanned, malformed, or visually complex PDF will be interpreted correctly.

This is particularly important when extracting numerical results. A model may identify the right number but associate it with the wrong study arm, unit, denominator, or time point. Important values should therefore be checked against the original paper.

Language Support

Elicit is optimized for English-language literature. It may handle queries and papers in other languages, but accuracy varies and non-English use can consume usage faster. It is not intended as a translation service. Test your discipline and language before relying on it for multilingual evidence reviews.

Data Extraction and Custom Columns

Structured extraction is one of the strongest reasons to choose Elicit over a general AI assistant.

Instead of summarizing every paper into a paragraph, researchers can define columns representing the information they need. Examples might include:

• Sample size
• Study design
• Population
• Intervention
• Comparator
• Primary outcome
• Follow-up period
• Reported limitations
• Effect estimates

The system then attempts to populate those fields across papers and can provide supporting quotations so the researcher can inspect where an extracted answer came from.

But the quotation is as important as the generated cell. An AI extraction that looks plausible may still misunderstand context. Researchers should verify units, definitions, study arms, outcomes, time points, and missing-data cases before relying on the table.

Current plan limits also affect the number of columns that can be added at a time: Plus supports five, Pro 20, Scale 30, and Enterprise 40 according to the pricing information reviewed for this article.

Research Reports

Research Reports provide a faster route from a question to a cited overview.

Rather than manually building a full systematic-review workflow, users can ask the platform to investigate a research topic, identify relevant literature, and organize findings into a report supported by sources.

This is useful for getting oriented in an unfamiliar field, preparing an early literature overview, identifying major themes, or finding papers worth reading in full.

It should not be confused with a completed systematic review. A polished report does not prove that the search was exhaustive, that every included study met formal eligibility criteria, or that risk of bias and evidence quality were assessed according to a recognized methodology.

Systematic Reviews

Eligible plans can use a guided process to gather studies, define inclusion and exclusion criteria, screen papers, extract structured information, and prepare review outputs.

Search can combine semantic discovery with more precise keyword and Boolean strategies. Researchers can add multiple search tabs and bring in papers from Elicit’s corpus, PubMed, ClinicalTrials.gov, uploaded PDFs, and saved Library collections.

The workflow can then assist with screening and data extraction rather than forcing users to move everything manually between unrelated AI chats.

However, “systematic review workflow” should not be interpreted as “automatic PRISMA compliance.” Elicit can assist with stages of evidence synthesis, but methodological decisions remain with the researcher.

Humans still need to determine whether the search strategy is adequate, whether screening decisions are defensible, how duplicate reports are handled, whether study quality or risk of bias has been evaluated appropriately, and whether the final synthesis supports the conclusion.

Screening and Human Verification

AI-assisted screening can be valuable when a literature search produces hundreds or thousands of candidate papers.

Elicit can help apply inclusion and exclusion criteria and prioritize the review process. Higher organizational tiers also support more advanced collaborative workflows, including features for independent human screening.

The danger is false exclusion. Saving time by automatically discarding irrelevant papers is useful only if relevant studies are not silently removed.

For high-stakes reviews, AI should therefore reduce repetitive work while preserving human oversight rather than replacing it.

Library, Zotero, and Exports

The Library lets researchers save papers, organize collections, import references from Zotero, and reuse collections as context for future work.

Shared collections and linked research projects make it more useful for ongoing evidence work than a temporary chat history.

The Library also includes AI-assisted duplicate detection. Elicit itself warns that duplicate detection can miss cases, so manual confirmation remains important.

Library references export as RIS. Research Reports export as PDF or Word on every plan, including Basic. Agent table exports require Plus or higher; Systematic Review table exports require Pro or higher. Tables support CSV or Excel, while suitable source tables also support RIS or BIB.

Even with these features, the Library should not automatically be treated as a complete replacement for a mature citation manager such as Zotero or EndNote.

API, MCP, and Research Automation

Elicit expanded beyond its browser interface in 2026 with API and MCP access for eligible paid subscriptions.

These capabilities can allow developers and organizations to integrate scholarly search, reports, and research workflows into other systems.

API usage is subject to plan allowances, rate limits, and specific terms. It is not simply an unlimited external version of the website.

The service also uses multiple AI models behind its research workflows. Although Elicit has publicly discussed particular Anthropic Claude models at different points, there is no stable public mapping showing that every current feature always uses one specific model.

For that reason, choosing Elicit should be based primarily on its research workflow rather than assuming the product is equivalent to whichever underlying language model was mentioned in a past announcement.

Elicit Free Plan

The Basic plan is free and provides a useful way to evaluate the platform before paying.

It includes limited Research Agent and Research Report usage, while Elicit advertises unlimited academic-paper search, paper summaries, paper chat, source viewing, and Zotero import.

The free tier excludes the dedicated Systematic Review workflow, Agent table exports, figure extraction, and API access. Research Report PDF and Word exports remain available.

“Unlimited” does not mean every Agent action is unmetered. When monthly usage runs out, the Agent remains available for Find Papers searches and tables; other Agent work waits for a reset or eligible extra usage.

Elicit Pricing in 2026

Elicit currently offers Basic, Plus, Pro, Scale, and Enterprise tiers. Pricing requires more caution than usual because the official pricing page reviewed for this article exposed different annual price sets, apparently related to customer type, while the exact month-to-month checkout prices could not be verified reliably from the published page.

Plan	Verified Annual Pricing Display	Main Positioning
Basic	Free	Trying paper search, Agent, and reports with limited usage
Plus	One verified annual display: $132/year ($11/month effective)	Light research with exports and up to 5 extraction columns
Pro	Verified annual displays included $468/year and $588/year	Guided systematic reviews, API/Routines, up to 20 columns
Scale	Verified annual displays included $1,068/year and $2,028/year	Larger reviews, collaboration, figure extraction, higher limits
Enterprise	Custom	Institutions and organizations needing negotiated controls and limits

Pricing verified: October 14, 2026. The official pricing page displayed multiple annual card sets, apparently connected to customer categories such as academic and industry pricing, but the extracted page did not establish the mapping clearly enough to present it as fact. Actual checkout pricing, taxes, discounts, currency, and billing options should be checked before purchasing.

These are annual commitments, not verified month-to-month subscription prices. The help center confirms discounted Academic pricing, selected from the pricing dropdown; nonprofits can use these academic plans rather than receiving a separate nonprofit discount. Confirm the applicable category and allowance before paying.

How Elicit Usage Limits Work

Elicit has moved away from the older simple credit model toward a monthly usage pool shared across Research Agent, Research Reports, and eligible Systematic Review work.

Usage varies with research complexity, file size, screening depth, and extraction settings. Premium PDF parsing is available on paid plans and increases usage for every PDF processed while enabled, not only difficult papers. Turn it off after the task if you no longer need it.

The allowance refreshes monthly and unused usage does not roll over.

Elicit does not publish a stable universal conversion such as “one report equals X credits,” so it would be misleading to estimate a guaranteed number of reports from a subscription.

Pro and Scale users can enable paid extra usage with a monthly spending cap. Charges normally occur in $100 increments, with smaller remaining amounts settled at the cycle end. Scale team usage is pooled across seats. Monitor representative work before estimating recurring costs.

Which Elicit Plan Is Best?

Basic makes sense for students and researchers who want to test semantic discovery, ask questions about papers, and experiment with research reports.

Plus is more relevant to lighter users who need structured extraction and exports but do not require the full guided review workflow.

Pro is the first clearly verified tier for the dedicated guided Systematic Review workflow. It supports reviews with up to 5,000 papers, up to 20 extraction columns, and access to capabilities such as Routines and the API.

Scale adds higher usage, collaboration, figure extraction, and up to 30 columns added at a time. Confirm the current screening-paper limit before committing to a large project; the retrieved pricing page did not clearly establish a separate Scale ceiling.

Enterprise is intended for organizations that need negotiated limits, institutional controls, security features, and larger review capacity.

Privacy and Confidential Research

Elicit’s public privacy policy, individual terms, team agreements, API terms, and Enterprise commitments do not all use identical language.

For example, Enterprise materials provide stronger no-training commitments for covered customer data, and API terms state that submitted input data is not used to train AI models unless the customer opts in. Other contractual language can grant broader rights for service development and improvement.

That means it would be inaccurate to state broadly that “Elicit never trains on user data” for every user and every account type.

Researchers handling unpublished studies, confidential pharmaceutical information, personal data, or proprietary evidence should check the agreement governing their specific account and their institution’s policies before uploading sensitive material.

Security and Research Integrity

Security controls and research reliability are separate questions.

A service can implement strong technical security while an AI extraction can still be scientifically wrong. Researchers should evaluate both.

For evidence work, the most useful safeguard is traceability: important generated answers should lead back to the underlying paper, quotation, table, or figure so the researcher can inspect the original evidence.

Retractions, duplicate publications, preprints, corrections, and conflicting studies also require attention. An AI system finding a paper does not establish that the paper is trustworthy.

Elicit AI Pros

Elicit AI Cons

Elicit vs Google Scholar and PubMed

Google Scholar is primarily a broad scholarly discovery and citation-exploration system. PubMed is a specialist biomedical literature database with established indexing and database-specific search controls.

Elicit adds something different: AI-assisted semantic discovery, structured extraction, screening, and evidence synthesis.

Those capabilities make it easier to move from “find papers” to “compare what these papers report.” They do not make the underlying academic databases obsolete.

A rigorous researcher may use both: traditional databases for controlled retrieval and Elicit for discovery, organization, extraction, and synthesis.

Elicit vs NotebookLM

The two products overlap in document analysis but start from different research problems.

Elicit is designed to find academic evidence and then help screen and extract it.

NotebookLM is particularly useful when you already have a controlled collection of sources and want to question, synthesize, study, or transform that collection.

If the problem is “find the relevant scientific literature about this research question,” Elicit is usually the more natural starting point. If the problem is “help me understand these 30 sources I already selected,” NotebookLM may be the more natural fit.

Elicit vs ChatGPT and Claude

ChatGPT and Claude are much broader general-purpose AI assistants. They can analyze documents, write, reason, code, and support many workflows outside academic literature.

Elicit trades that breadth for a more specialized research structure.

A researcher who needs repeatable paper extraction, screening stages, and evidence tables may benefit from Elicit’s dedicated workflow. Someone who needs one assistant for research plus coding, writing, brainstorming, data analysis, and other general tasks may prefer ChatGPT or Claude.

These tools can also complement each other. Elicit can help construct and verify an evidence base, while a general assistant can support later communication or analysis—provided important claims remain tied to their original sources.

Elicit vs Consensus

Consensus is one of the more direct alternatives because it also focuses on research-grounded answers and academic search.

Elicit places particular emphasis on configurable extraction tables, screening, and the guided systematic-review process. Consensus emphasizes its own question-to-evidence search and synthesis experience.

The better choice depends on whether your priority is quickly investigating research questions or building a structured evidence-review workflow. Comparing corpus-size marketing numbers alone does not establish which tool will retrieve the best literature for a particular topic.

Who Is Elicit Best For?

Elicit is particularly compelling for graduate students, PhD researchers, academics, evidence-synthesis teams, biomedical researchers, and professionals who repeatedly need to search and compare scientific literature.

It can also be useful for undergraduate students who need a faster way to discover and understand research papers, provided they still evaluate source quality and follow their institution’s academic-integrity rules.

Its value is highest when research involves repeated structured questions across many papers. If you need to extract the same population, intervention, sample size, outcome, or limitation from dozens or hundreds of studies, the evidence-table workflow can be much more efficient than ordinary conversational AI.

Who Should Consider an Alternative?

Users who mainly want general web research may be better served by Perplexity or a general AI research tool. Those who already possess their source collection and want deep source-grounded learning may prefer NotebookLM.

Researchers primarily interested in citation-network exploration can complement or replace parts of the discovery workflow with tools such as ResearchRabbit or Connected Papers.

And anyone conducting an exhaustive formal review should consider Elicit part of a larger research methodology rather than assume one AI platform can replace specialist databases, librarians, reference managers, statistical tools, and human methodological review.

Sources: Official pricing, academic pricing, usage limits, exports, and API terms.