What it is
Elicit is an AI tool for working with scientific literature. It searches more than 138 million academic papers, summarises them, and pulls specific details (sample size, methods, outcomes) into a table you can sort and export, with the supporting quote behind each cell.
Paid plans add a Research Agent, Research Reports and a guided Systematic Literature Review workflow that screens thousands of papers. Since mid-2026 these draw on a single monthly usage pool rather than fixed counts of reports. A REST API launched in preview on 3 March 2026, and Elicit now also runs a hosted MCP server for agents such as Claude Code.
The cheaper Plus plan that many university library guides still mention no longer appears on Elicit's pricing page; paid plans now start at Pro.
Our review
Elicit is the strongest tool in its class for the dullest part of evidence synthesis, and it is also the only one with an independent accuracy study to argue with. In a feasibility study published in Research Synthesis Methods on 29 May 2026, researchers tested it on studies drawn from seven environmental and life-science reviews. Overall extraction accuracy was 86.6%, against 98.6% for two human extractors working together. Only 69% of the variables tested (48 of 70) held the authors' 87% threshold on new papers. Their recommendation is to use Elicit as a second extractor, with a third reviewer settling disagreements.
The price stops being simple the moment you leave the free plan. Pro costs $49 per user per month billed annually ($588 a year) and Scale $169 ($2,028 a year); the $12 Plus tier that suited students has gone. Since mid-2026 the Research Agent, Reports and reviews share one monthly usage pool, and each task costs what its size demands: long PDFs, many columns, figure extraction and full-text screening all eat more. Elicit does not publish how big each plan's pool is, so a team can budget the seats but not the work.
For engineers the API is more capable than most research tools offer. Version 2 covers paper and ClinicalTrials.gov search, reports, systematic reviews and the agent, and a hosted MCP server uses OAuth instead of keys. The limits are tight, though. Every endpoint shares 100 requests a minute per IP, polling included, and breaching it blocks the IP for five minutes. Reports take 5 to 15 minutes, so a script that polls too eagerly can lock itself out.
On data, Elicit holds a SOC 2 Type II report and keeps uploaded PDFs private to the account and out of its shared corpus. The explicit promise not to train on customer data is published for Enterprise only, as are SSO and single-tenant hosting. The study also flagged a quieter risk: switching to high-accuracy mode changed almost a quarter of the extracted values. A review protocol should record the Elicit mode and run date as carefully as it records a database search string.
Where it earns its place
A two-person evidence-synthesis team at a public-health agency
Screen and extract data for a systematic review of about 3,000 records
Pro screens up to 5,000 papers per review with 20 extraction columns, at $588 per seat a year. Following the 2026 Research Synthesis Methods study, Elicit acts as the second extractor and a third reviewer settles disagreements, which keeps the PRISMA record defensible.
A data engineer at a biotech building an internal literature monitor
Pull new papers and trials on a drug target into a weekly digest
API v2 with a Bearer key: paper search returns up to 300 results per request on Pro, and a separate endpoint queries ClinicalTrials.gov. All calls share 100 requests a minute per IP, and a 429 blocks that IP for five minutes, so polling has to be throttled.
The head of research at a 10-analyst policy consultancy
Give analysts a shared review workspace with usage tracking
Scale at $169 per user per month billed annually adds live collaboration, figure extraction and an admin panel; the pool grows with seats, so a four-person team gets four times an individual's usage. Ten seats come to $20,280 a year before extra usage.
A PhD student writing a literature chapter
Find and summarise papers on a narrow question
Basic is free with unlimited search and summaries and limited agent and report runs. Extraction tables at scale and systematic reviews need Pro at $49 a month billed annually, which is a lot for occasional use.
Before you commit
- Setup
- Sign up and search at once; existing libraries import from Zotero. API keys are created in account settings on Pro and above, and the MCP server connects with one command in Claude Code.
- Integrations
- Zotero import, REST API v2 (papers, trials, reports, systematic reviews, Research Agent, library and projects) and a hosted MCP server at elicit.com/api/mcp using OAuth.
- Data and privacy
- SOC 2 Type II; uploads are encrypted with AES-256 at rest and are not added to Elicit's corpus. A no-training commitment, SSO and SAML, and single-tenant AWS hosting are published for Enterprise only.
- Cost at scale
- Ten seats on Scale cost $20,280 a year; overage is billed in $100 increments up to a monthly cap you set, and unused usage does not roll over. Pool sizes per plan are not published.
- Limits
- Reviews cap at 5,000 papers and 20 columns on Pro, 20,000 and 30 on Scale, 40,000 and 40 on Enterprise. Edits and re-runs draw on the pool again; the API allows 100 requests a minute per IP.
What's good
- Every extracted cell links to its supporting quote
- Systematic reviews screen 5,000 papers on Pro and 40,000 on Enterprise
- Free unlimited search across 138 million papers
- REST API and hosted MCP server from Pro upwards
- SOC 2 Type II; uploaded PDFs stay private to your account
What's not
- No plan between free and $49 a month billed annually
- Usage pool sizes are not published, so heavy reviews are hard to budget
- 86.6% extraction accuracy in a 2026 study, against 98.6% for paired humans
- No-training commitment and SSO published for Enterprise only
- Covers academic papers and ClinicalTrials.gov, not the general web
Best for
Researchers, analysts and review teams who need to extract structured data from many academic papers.
Alternatives
Elicit: common questions
- Is Elicit worth it?
- Elicit turns hundreds of papers into a cited data table, but a peer-reviewed test put its extraction accuracy at 86.6%, so it belongs beside a human extractor rather than in place of one. We rate it 4.3 out of 5.
- How much does Elicit cost?
- Basic free (unlimited search and summaries, limited agent and reports); Pro $49/user/month billed annually ($588/year); Scale $169/user/month billed annually ($2,028/year); Enterprise custom. Extra usage on Pro and Scale in $100 increments up to a cap you set.
- What are the downsides of Elicit?
- No plan between free and $49 a month billed annually; usage pool sizes are not published, so heavy reviews are hard to budget; 86.6% extraction accuracy in a 2026 study, against 98.6% for paired humans; no-training commitment and SSO published for Enterprise only; covers academic papers and ClinicalTrials.gov, not the general web.
- Who is Elicit best for?
- Researchers, analysts and review teams who need to extract structured data from many academic papers.
- How much does Elicit cost for a team or in production?
- Ten seats on Scale cost $20,280 a year; overage is billed in $100 increments up to a monthly cap you set, and unused usage does not roll over. Pool sizes per plan are not published.
- Is Elicit safe for company data?
- SOC 2 Type II; uploads are encrypted with AES-256 at rest and are not added to Elicit's corpus. A no-training commitment, SSO and SAML, and single-tenant AWS hosting are published for Enterprise only.
- What are the best Elicit alternatives?
- The closest alternatives we have reviewed are Consensus, Undermind and Scite.
Answers checked Oct 22, 2026.
Make Elicit? Show this rating on your site
