Elicit Review: AI Research, Systematic Reviews and Pricing

Elicit has moved beyond helping researchers find and summarize papers. Its August 2026 Research Agent launch positions the platform as an environment for gathering evidence, reasoning through scientific tradeoffs, and producing materials that can inform high-stakes decisions. The distinction is important. A literature assistant helps someone understand existing research. A decision-support system can influence trial design, clinical strategy, indication prioritization, laboratory methods, patents, market entry, policy, and investment. The potential value is much larger, but so is the cost of being confidently wrong.

Elicit announced the Research Agent in a blog post dated August 4, 2026, and recorded it in the public changelog on August 5. The company says the Agent can gather information from scientific literature, clinical trials, patents, regulatory databases, structured biology databases, commercial sources, the web, and uploaded internal data. It can then produce cited reports, data analyses, tables, calculations, figures, visualizations, slides, and documents. Projects keep sources, context, instructions, and outputs connected across sessions, while the API allows organizations to bring Elicit's research functions into their own workflows.

That is a meaningful expansion from Elicit's original strength in evidence synthesis. It also changes how buyers should evaluate the product. Search recall and extraction accuracy remain important, but they are no longer enough. Teams now need to examine how the Agent handles uncertainty, conflicting evidence, assumptions, missing data, calculations, source quality, internal information, and decisions where several options are defensible. A polished slide deck is not proof that the reasoning behind it is complete.

### From paper workflow to decision environment

Elicit's core product is still recognizable. A researcher can search more than 138 million academic papers using semantic or keyword queries, explore clinical trial records, chat with papers, add custom extraction columns, create reports, organize a source library, and monitor topics with Alerts. The dedicated Systematic Review workflow guides users through protocol setup, paper gathering, title and abstract screening, optional full-text screening, data extraction, and report generation. Supporting quotes, exclusion reasons, overrides, audit information, and PRISMA-oriented methods make the process more inspectable than a general-purpose chat interface.

Research Agent builds on that foundation but broadens the unit of work. The question is no longer only, what does the literature say? It can become, which development path has the strongest evidence, what uncertainty could change the choice, what data should be collected next, and how should the evidence be communicated to a team? Elicit presents the Agent as a system that can move from evidence gathering through analysis to decision-ready materials in one environment.

The Projects feature is central to that model. Important research decisions rarely happen in one prompt or one afternoon. A team may begin with a landscape review, continue into a trial comparison, update the work with internal results, explore an alternative mechanism, and prepare a presentation weeks later. A Project can preserve relevant sources, instructions, intermediate outputs, and context across those sessions. That continuity can reduce repeated searching and help teams build a reusable evidence base. It also creates a governance responsibility: project context needs owners, permissions, version control, review dates, retention rules, and a clear way to remove obsolete or unauthorized material.

### Elicit is trying to measure reasoning, not only answers

Alongside Research Agent, Elicit introduced BioDecisionBench, a benchmark focused on reasoning failures in high-stakes pharmaceutical decisions. The company argues that benchmarks measuring only final answers are insufficient because a conclusion can look plausible while missing a decisive consideration. Its benchmark covers reasoning about areas such as preclinical study design and observational study validity. In Elicit's published result, its Smartest mode identified 76.7% of key considerations compared with 68.8% for Claude Opus 5 Max.

The focus on reasoning process is sensible. Scientific decisions often involve incomplete evidence, causal assumptions, external validity, competing risks, operational constraints, and uncertainty that cannot be reduced to one correct sentence. Evaluating whether a system surfaces the right considerations can be more useful than grading the elegance of its final recommendation. Transparent methods and sentence-level citations also give reviewers a practical route back to the evidence.

Buyers should still treat these as vendor-reported benchmark results. A life-sciences benchmark may not predict performance in education policy, materials science, consumer safety, engineering, or a company's proprietary research process. Even within pharma, a benchmark score does not reveal every failure mode that matters to a particular team. Organizations should create their own evaluation set from completed decisions, difficult edge cases, known contradictions, missing-data situations, and examples where a convincing analysis led to the wrong outcome. They should score source selection, factual accuracy, assumption visibility, calculation correctness, uncertainty, reasoning coverage, and whether the system knows when to stop.

### The API makes research quality an infrastructure issue

Elicit's July 2026 API and MCP launch is another important part of the new direction. The platform now offers Search, Reports, and Systematic Review endpoints on Pro or higher plans. The Search API covers academic papers and clinical trials. The Reports API produces cited research reports. The Systematic Review API exposes an end-to-end process with control over search, screening, extraction, and reporting. The same endpoints can be used through MCP-compatible clients.

This can help teams put evidence directly into internal agents, research portals, competitive-intelligence systems, product workflows, or decision records. It can also scale mistakes beyond the Elicit interface. If an API-generated claim loses its source ID when it enters a database, the reviewer may see a conclusion without its evidence. If a downstream agent summarizes the report again, qualifiers can disappear. If a system automatically creates a recommendation when evidence is sparse, uncertainty can turn into a business action without a responsible owner.

A production integration should preserve the query, search strategy, source identifiers, passages, extraction definitions, methods, model or mode, timestamps, usage, edits, and approval status. It should distinguish retrieved facts from calculations, Elicit inferences, and human judgments. High-impact workflows need stop conditions for missing or conflicting evidence, plus a named reviewer before results affect patients, studies, regulatory submissions, safety claims, policy, intellectual property, or capital allocation.

### Systematic review rigor still depends on the researcher

The Research Agent launch does not remove the limitations that already apply to Elicit's paper workflows. Elicit's corpus is broad, but no single search system guarantees complete retrieval. Specialist databases, regional indexes, dissertations, books, conference material, grey literature, hand searching, citation chasing, trial registries, and direct author contact may still be necessary. Semantic search can find conceptually related work that keyword queries miss, but reproducibility requires that researchers record queries, filters, imports, dates, sources, and manual additions.

Full-text access is another boundary. Elicit can use open-access papers, uploaded PDFs, Zotero imports, and an extension that retrieves eligible papers through a user's active institutional access. When full text is unavailable, some analysis depends on the title and abstract. That may be adequate for initial screening, but abstracts frequently omit subgroup definitions, intervention details, outcome timing, attrition, analysis choices, adverse events, and limitations. Missing full text should be recorded as missing evidence, not silently converted into an exclusion or a confident extraction.

Automated screening and extraction also require carefully designed criteria and fields. A strict criterion can wrongly exclude a paper when the relevant information is not present in the available text. An extraction can confuse treatment arms, combine time points, change units, or report an unadjusted estimate when the protocol requires an adjusted one. Elicit's supporting quotes make these errors easier to detect, but a citation trail only helps when someone follows it. Critical cells, figures, calculations, and exclusion decisions still need human verification.

### Pricing now reflects several kinds of researcher

Elicit's official pricing page displays audience-specific annual rates. Basic is free and includes unlimited paper search, summaries, and paper chat, with limited Research Agent and Report usage. The academic annual view lists Plus at $11 per user per month, Pro at $39, and Scale at $89. Another audience view lists Pro at $49 and Scale at $169 per user per month when billed annually. Enterprise is custom. Because the same page changes by audience selection, buyers should confirm the correct segment, currency, tax treatment, billing period, and usage allocation inside the live checkout or sales process.

The plan decision should be based on workflow, not only seat price. Plus adds exports and standard usage. Pro is the practical starting point for dedicated systematic reviews, higher report limits, custom extraction, Alerts, API, and MCP. Scale adds more usage, figure interpretation, real-time collaboration, up to 200 report sources, larger column batches, and administration. Enterprise adds much greater review scale, custom data sources, security controls, deployment options, customer success, and custom usage. Teams should run representative projects and record how the shared usage pool behaves before forecasting annual cost.

### A strong direction with a higher burden of proof

Research Agent is a logical next step for Elicit. The company already built tools for finding evidence, structuring it, showing source passages, and preserving a research process. Extending those capabilities into analysis and decision-support deliverables could remove a large amount of repetitive work from scientific teams. The combination of Projects, diverse sources, cited outputs, API access, and explicit attention to reasoning makes the launch more substantial than adding another chat mode.

The phrase high-stakes decisions must carry a high burden of proof. Elicit should be judged not only by how much information it can gather or how quickly it can create a presentation, but by whether it helps experts notice decisive evidence, expose uncertainty, challenge assumptions, and defend the path from source to conclusion. The safest use is not to ask the Agent to decide. It is to use the Agent to assemble evidence, test options, reveal gaps, prepare transparent materials, and make expert review more effective.

For research teams with defined methods and responsible reviewers, Elicit is becoming a flexible environment that connects literature work to the decisions that literature is meant to inform. For students or casual users, the free search and paper tools remain useful. For clinical, regulatory, policy, legal, safety, or investment work, the new Agent should be treated as decision-support infrastructure with documented oversight, not an autonomous authority.

WhatAI Decision Box

Elicit is an evidence-focused AI platform for finding papers, screening studies, extracting structured data, creating cited reports, and running systematic review workflows. Its 2026 Research Agent expands that scope into broader scientific analysis and decision-support work.

Elicit is strongest when a research question needs traceable sources, structured comparisons, explicit screening criteria, linked supporting quotes, and an auditable path from search to synthesis. It offers more research-specific controls than a general chat assistant and can preserve work across reports, reviews, tables, and Projects.

Search coverage, full-text access, extraction accuracy, study quality, statistical interpretation, reporting standards, and applicability still require human judgment. Researchers should verify every important claim against the original source and document any supplementary databases, manual searches, corrections, and review decisions.

Use Cases

Pricing varies by plan and region — see current pricing.

Plan features change — last updated: 2026-08-16.

Elicit Community Discussions

Explore community discussions. Ask and answer questions on Elicit to grow and learn together.

Elicit Pros & Cons

How to Get Results with Elicit: Step-by-Step Workflow

1. Define the review question

   Write the population, intervention or exposure, comparator, outcomes, context, study designs, date range, languages, and decision the evidence will support. Register a protocol when the project requires one.

2. Set the evidence boundaries

   Choose eligible publication types, sources, databases, full-text requirements, grey literature, preprints, regional indexes, and quality thresholds. Decide what Elicit will cover and which supplementary searches remain manual.

3. Choose the correct plan

   Estimate review count, papers per review, extraction columns, report sources, Agent usage, exports, alerts, collaboration, figure extraction, API access, and security needs. Confirm the correct academic or industry pricing view.

4. Build reproducible searches

   Use both semantic and keyword searches, record every query and filter, import results from specialist databases, and deduplicate before screening. Test known relevant papers to identify likely gaps.

5. Refine screening criteria

   Turn the protocol into clear, independent criteria. Pilot them on representative papers, inspect Elicit's supporting quotes and reasons, adjust ambiguous rules, and define how missing information should be handled.

6. Screen with human oversight

   Review title and abstract recommendations, apply strict criteria only where failure truly requires exclusion, record overrides, resolve uncertainty, and use dual human review where the protocol or stakes require it.

7. Retrieve and screen full text

   Use open access, uploads, Zotero, and approved institutional access through the extension. Track missing papers and apply full-text criteria before extraction when eligibility depends on details absent from abstracts.

8. Design extraction columns

   Define each variable, unit, time point, population, missing-data rule, and acceptable source location. Pilot the schema, inspect supporting passages, and manually verify critical numbers, methods, tables, and figures.

9. Synthesize and challenge

   Generate a report or Agent analysis, then test competing explanations, heterogeneity, study quality, bias, uncertainty, applicability, and contradictions. Separate source facts, calculations, model inferences, and expert judgments.

10. Document and maintain

    Export the search strategy, screening decisions, exclusions, extractions, sources, methods, corrections, and final outputs. Store the approved work in a Project, set Alerts, and schedule updates when new evidence could change the conclusion.

Elicit Gotchas and Limits to Know Before You Start

• Elicit is an evidence assistant, not an authority that makes research findings true.
• Sentence-level citations can still be attached to a claim that overstates or misreads the source.
• A source passage may support one detail while the surrounding paper changes its interpretation.
• Full-text access is not available for every indexed paper.
• When full text is missing, extraction may rely on titles and abstracts with limited detail.
• The browser extension depends on lawful access through the user's active institutional subscriptions.
• The extension does not give access to papers that the user or institution is not entitled to retrieve.
• The corpus includes preprints, which may not have completed peer review.
• Books, dissertations, and some forms of grey literature are outside Elicit's main paper corpus.
• Regional, specialist, non-English, and unpublished evidence may require supplementary databases.
• A large source corpus does not guarantee complete recall for a specific systematic review.
• Semantic search is useful for discovery but may be harder to reproduce than a documented keyword strategy.
• Search results can change as source indexes, models, ranking, and corpus updates change.
• Screening recommendations depend on the clarity and independence of the criteria.
• Strict criteria can exclude relevant papers when the available text is incomplete or ambiguous.
• AI as a second reviewer may not satisfy every protocol, journal, regulator, or evidence standard.
• Dual review still requires conflict resolution and clear accountability for the final decision.
• Automated extraction can confuse arms, populations, time points, units, or adjusted and unadjusted values.
• Figures and tables can contain information that text-only extraction misses or misinterprets.
• Study quality and risk of bias require domain-specific judgment beyond simple relevance screening.
• A narrative report can hide heterogeneity, null findings, publication bias, and incompatible outcome definitions.
• Research Agent can use web and internal data, so source quality varies more than in a paper-only workflow.
• Internal uploads may contain confidential, personal, regulated, or commercially sensitive information.
• High-stakes scientific advice needs qualified human review, independent checks, and documented uncertainty.
• Elicit's BioDecisionBench results are vendor-reported and should not replace testing on the buyer's tasks.
• Decision-support outputs can appear polished even when assumptions, calculations, or missing evidence are weak.
• Projects preserve context over time, which also increases the need for access, retention, and deletion rules.
• API and MCP use can scale unsupported conclusions unless source and approval requirements are enforced downstream.
• The pricing page displays different annual prices depending on the selected audience.
• Usage for Agent, Reports, and reviews is pooled and varies by plan, so teams should test representative work.
• Exports are plan-dependent and the Elicit Library itself may not support a complete export.
• Enterprise security claims must be checked against the signed contract and configured deployment.
• Pricing, corpus size, usage limits, plan names, and feature availability can change after verification.

Which Elicit Feature Fits Your Use Case

Starter Prompts for Elicit

What is Elicit?

Elicit is an AI research platform for finding scientific papers, screening studies, extracting structured evidence, generating cited reports, organizing sources, running systematic review workflows, and supporting broader research decisions.

Is Elicit free?

Yes. Elicit's Basic plan is free and includes unlimited paper search, summaries, and chat with papers, plus limited usage for Research Agent and Research Reports. Exports, systematic reviews, higher usage, collaboration, and API access require paid plans.

How much does Elicit cost?

As verified on August 16, 2026, the annual pricing view lists Plus at $11 per user per month, Pro at $39 or $49, and Scale at $89 or $169 depending on the audience selected. Enterprise is custom. Confirm the correct academic or industry selector before purchasing.

What is Elicit Research Agent?

Research Agent is Elicit's environment for gathering evidence across scientific and public sources, analyzing data, exploring tradeoffs, and producing cited reports, tables, calculations, charts, slides, and documents for scientific decision support.

Can Elicit perform a systematic review?

Elicit provides a dedicated workflow covering protocol setup, paper gathering, title and abstract screening, optional full-text screening, structured extraction, and report generation. Researchers remain responsible for the final search strategy, manual decisions, methods, quality assessment, and reporting.

Does Elicit support PRISMA 2020?

Yes. Elicit added PRISMA 2020 support in May 2026 and provides methods trails, screening reasons, audit information, and flow diagrams. Using these features does not automatically make a review publication-ready or compliant with every protocol requirement.

How many papers can Elicit search?

Elicit says its search covers more than 138 million academic papers from sources including Semantic Scholar, OpenAlex, and PubMed. It also supports searching more than 500,000 ClinicalTrials.gov records. Corpus size does not guarantee complete retrieval for every review.

Does Elicit read full papers?

Elicit can use available open-access full texts, uploaded PDFs, Zotero imports, and eligible papers retrieved through its browser extension. When full text is unavailable, some analysis may rely only on titles and abstracts, which can omit decisive details.

Does Elicit provide citations?

Yes. Elicit links AI-generated claims and extraction cells to supporting source passages. Users should still open the original paper, check that the passage supports the claim, and verify context, methods, figures, numbers, and limitations.

Can Elicit replace PubMed, Scopus, or specialist databases?

Not for every review. Elicit combines broad sources and supports reproducible search, but specialist databases, grey literature, regional indexes, hand searching, citation chasing, and protocol-specific queries may still be needed for comprehensive retrieval.

Can teams collaborate in Elicit?

Scale and Enterprise support real-time collaboration and administration. Elicit also offers dual review, conflict resolution, pooled usage, view-only sharing, and audit information, with exact availability depending on the plan and workflow.

Does Elicit have an API?

Yes. Pro and higher plans can access Search, Reports, and Systematic Review APIs, with the same endpoints available through MCP for compatible clients. Organizations should apply authentication, rate, data, provenance, and review controls around automated use.

Is Elicit suitable for medical or high-stakes decisions?

It can support evidence gathering and analysis, but it should not be the sole basis for clinical, regulatory, policy, legal, safety, or investment decisions. Qualified experts must assess source quality, applicability, uncertainty, conflicts, and consequences.

Is Elicit secure?

Elicit maintains a Trust Center, publishes a data-processing addendum, references SOC 2, and offers enterprise controls such as SSO, SAML, 2FA, domain verification, and custom deployments. Buyers must verify the controls, contract, data location, retention, model use, and access design required for their work.

Related Research & Knowledge Work Tools

Explore the Network