🛠️ AI Tool Lab Updated daily

What Elicit Actually Is (From Someone Who Reads Papers for a Living)

I have spent the last two years doing literature reviews for clients - mostly busy clinicians, founders raising in health tech, and consultancies that need evidence-backed answers but would rather pay someone else to read the papers. Elicit is the tool that made me stop dreading that work.

Elicit is a research assistant built for academic literature, not general web search. It searches a corpus of 138 million papers and 545,000 clinical trials, then pulls out what matters: you ask a question, it finds the relevant studies, and instead of making you open forty PDFs it extracts the sample sizes, methods, outcomes and effect sizes into a structured table. Every single claim links to the exact sentence in the source paper - click through and you land on the sentence that supports it. That sentence-level citation system is the whole point. ChatGPT summarises vaguely; Elicit shows you the receipt.

It grew out of Ought, the non-profit AI research lab, and went through a serious product shift in 2025-2026. The old 'ask Elicit anything' assistant became a full research workflow: Research Agents run automated multi-step jobs like competitive landscape mapping, strict screening criteria (added December 2025) tightened the systematic review workflow, an API launched in March 2026, and automated reports can now synthesise up to 80 papers at once. Over 2 million researchers use it. My honest take after two years: it is the only AI tool I trust to hand a client a citation for.

Pricing, Plainly

• Free (Basic): unlimited search across 138M+ papers, unlimited summaries, unlimited chat with papers, Zotero import, 2 automated reports per month, 2 table columns.
• Pro: $49/user/month, billed $588/year (35% off). Standard usage for Research Agents, reports and systematic reviews; screening up to 5,000 papers; 20 columns; extraction from up to 135 data sources; 10 personalized research alerts; custom extractions from your uploaded PDFs; API access.
• Scale: $169/user/month, billed $2,028/year. 5x the usage, figure extraction and interpretation, live team collaboration, 30 columns, 200 data sources, admin panel with usage tracking.
• Enterprise: custom. Screen 40,000 papers, 40 columns, no training on your data by default, PRISMA-grade screening accuracy, SSO/SAML/2FA, unlimited API.

There is no per-seat minimum drama on the free plan - you can run a real project on it, which is more than most research tools offer.

Why It Pays (The Real Income Angle)

Elicit's buyers fall into two buckets: people who must read papers (academics, clinicians) and people who need what is in the papers but refuse to read them (founders, consultants, investors). The second bucket is where the money is.

1. Literature review as a service. This is the cleanest entry point, and I know half a dozen people doing it. A PhD student drowning in 200 papers, a clinic director who needs to know what the evidence says about a treatment, a startup team writing a tech due-diligence doc - they all pay $150-$500 for a scoping review they do not have time to do. On the free plan you can run 2 automated reports a month; on Pro at $49 the ceiling disappears (144 reports a year). Four to six reviews a month at $200-$400 each clears $1,000-$2,500 at a tool cost of $0-$49. The margin is basically the invoice.

2. Systematic review support for health and med-tech. This is the big-ticket play. Systematic reviews are the gold standard in medicine and they normally take a team of reviewers months - screening thousands of abstracts by hand. Elicit's PRISMA-style screening workflow (5,000 papers on Pro, 40,000 on Enterprise) compresses the screening phase to days. Research consultancies and med-tech companies pay $2K-$10K per systematic review engagement, and your tool cost is one Pro seat. If you have any domain credibility in health, this is the highest-margin service I know that a solo operator can run.

3. Evidence-backed market and competitive research. The Research Agents feature turned Elicit into more than a literature tool - you can run multi-step jobs like 'map every published study on AI in dental imaging, extract the authors and funders, flag the gaps.' Funds, VCs and consultancies pay $500-$2K for a well-cited landscape report. My last one took two days and billed at $1,500; the entire Elicit cost was the $49 Pro seat.

4. Grant-writing evidence packs. Funder applications want cited evidence that the problem exists and your approach is feasible. Assemble that evidence pack with Elicit - search, extraction, citation exports - and charge $500-$2,000 per grant package. Non-profits and research startups are chronically undersold on this service.

5. Research-alert digest subscriptions. Pro gives 10 personalized research alerts. Productise it: a monthly 'what's new in X' briefing for clinicians, product managers or investors at $200-$500 per month per client. Five clients at $300 clears $1,500 a month for an hour of curation work - Elicit does the watching, you do the editing.

6. Cited course and CE content. If you create training content - continuing education, clinical training, science communication - Elicit lets you build courses with real citations instead of vibes, which is exactly what CE reviewers check. A cited course package sells for $1K-$5K to professional bodies.

The unit economics that sell the story: a $300 literature review burns roughly $0-2 of Pro usage and a few hours of your time. The tool cost inside a $2K systematic review engagement is one $49 seat.

How I Actually Run It (A Real Workflow)

A recent job: a telehealth startup needed to know whether remote monitoring reduced hospital readmissions for heart failure patients, and what the evidence said about the effect sizes. The old way was two weeks of manual screening. The Elicit way: one research question, it returned the relevant trials, I added extraction columns for sample size, follow-up duration and readmission effect, exported the table, cross-checked the top five studies by opening the cited sentences, and wrote the client a 12-page memo with 40 citations. Total hands-on time: about six hours. Billed: $900.

The rule I live by: always open the cited sentences for any number you plan to put in front of a client. Elicit is anchored to real papers and rarely fabricates a source, but it can still misread a metric inside a messy PDF - catching that is your job, not its.

What Elicit Is Bad At (Honest)

• It is an academic tool, full stop. No general web search, no news, no product data. If your question lives outside peer-reviewed literature - 'what are the best Shopify apps' - Elicit is the wrong tool and returns nothing useful.
• The free tier is a teaser for report work. Unlimited search and summaries are genuinely free, but the 2-reports-per-month and 2-column limits push any real paid service onto Pro at $49/month. Fair price, not a cheap one.
• Extraction needs human verification. The 99.4% accuracy claim is marketing-grade; in practice complex tables, figures and ambiguous methods still get misread. Never hand a client an extracted number you have not checked against the source sentence.
• There is a learning curve. The column-and-table workflow is a spreadsheet, not a chat box. First-timers spend an afternoon fighting it before it clicks.
• Index gaps are real. Coverage is strongest in biomedicine; niche fields, very recent papers and non-English literature come up short because the corpus is built on Semantic Scholar data.
• Collaboration is gated. Real-time team editing only arrives at Scale, $169/month. A two-person review operation fits in one Pro seat, but if you need live collaboration, the jump is steep.

Elicit vs The Alternatives

The honest summary: Consensus is faster for binary questions, ResearchRabbit is better for finding connected papers, NotebookLM is better when the sources are yours. Elicit wins when you need to extract structured evidence across many papers and hand a client a verifiable citation - nobody else in the category does that.

Who Should Use It (And Who Shouldn't)

Use it if: you sell research, literature reviews or systematic review support to health, academia or evidence-hungry businesses; you are a consultant who needs cited answers fast; you write grants, courses or reports that require real citations; or you are a researcher who wants to cut the screening phase of a review from weeks to days.

Skip it if: your work is general web research, news, product or market data - you want Perplexity or Exa instead; you are a casual user who just wants a quick answer (Consensus is faster); or you refuse to verify extracted numbers manually, because trusting Elicit blindly will eventually embarrass you in front of a client.

Getting Started (6 Tips)

• Start free and run a real project on it. Unlimited search and summaries mean you can test whether it covers your niche before spending a cent. If your field returns thin results, you learned that for free.
• Learn the extraction table before anything else. That is the killer feature - add columns for sample size, outcome, effect size - and it is where the value for clients lives.
• Click every citation you plan to use. The sentence-level links are the product's superpower and your professional safety net. Check, then quote.
• Upgrade to Pro only when you have a paying use for reports. Two free reports a month is enough to prove the service; the $49 pays for itself on the first $300 review.
• Use Research Agents for landscape jobs. For 'map the evidence in X' client work, the multi-step agents save a full day versus doing it query by query.
• Export to RIS/CSV and keep your own library. Import from Zotero, export to your reference manager - client handoff documents land cleaner and the citations stay portable.

The bottom line: Elicit is not a chatbot that guesses, it is an evidence machine that shows its work. For anyone billing for research - literature reviews, systematic reviews, landscape reports, grant support - it is currently the best tool in the category, and at $0-49 a month it costs almost nothing next to what the service built on it can charge.