Elicit Review: Google's Chief Scientist Bet on This Tool

Picture the worst part of any research project. Not the experiment, not the writing, just the reading. Hundreds of PDFs, forty browser tabs, and a spreadsheet nobody updates properly. Elicit is a bet that an AI can do a big chunk of that reading for you, and do it carefully enough that a scientist would actually trust the answer.

Elicit is an AI research assistant built to search, read, and synthesize academic papers, with citations you can check down to the sentence. It has raised $31 million from investors that include Jeff Dean, Google's own chief scientist, and it has quietly become one of the fastest-growing tools in a category most AI blogs never cover. This review breaks down what Elicit actually does, what it costs today, and whether it earns a place in your workflow if you are not already living inside Google Scholar.

Quick Verdict

What works well:

• Every answer links back to the exact sentence it came from, not just a source list at the bottom of the page

• The free plan is genuinely usable: unlimited search across more than 138 million papers, unlimited summaries, and unlimited chat with full-text papers

• The Systematic Review workflow follows PRISMA 2020 reporting standards and comes with published accuracy numbers, not just marketing claims

• SOC 2 Type II compliant, which matters if you're working with anything sensitive, like unpublished clinical or internal data

• Built by a team that started as a nonprofit research lab and still operates as a Public Benefit Corporation, so getting the evidence right is genuinely the incentive, not just the pitch

What to watch:

• It does one job. If you want a general-purpose AI assistant for everyday writing or a broad web search, Elicit will feel narrow and slow by comparison

• The free plan caps usage on Research Agent and Research Reports, the two most powerful features, fast

• Pro jumps to $49/month, a real price for a solo researcher or student next to cheaper competitors

• We could not find a public affiliate or referral program for Elicit, more on that at the end of this review

What Is Elicit, and Who Built It?

Elicit exists to help researchers be "10x more evidence-based," in the company's own words. It was built by cofounders Andreas Stuhlmüller and Jungwon Byun, and it started life as a research effort inside Ought, a nonprofit dedicated to scaling up good reasoning with machine learning.

In September 2023, Elicit spun out of Ought as an independent Public Benefit Corporation, raising a $9 million seed round led by Fifty Years. The investor list is worth pausing on: it included GitHub co-founder Tom Preston-Werner, Dropbox co-founder Arash Ferdowsi, and Jeff Dean, Google's chief scientist. At the time, Elicit said more than 200,000 researchers were using it every month, including teams at Harvard, MIT, Stanford, Genentech, Novartis, and OpenAI.

By February 2025, Elicit had grown enough to raise a $22 million Series A, co-led by Spark Capital and Footwork, with Jeff Dean returning as a backer. That round valued the company at $100 million and brought total funding to $31 million. Monthly usage had roughly doubled to more than 400,000 researchers, and Elicit said it was expanding beyond pure academia into pharma, consulting, and medtech, citing new customers like MicroGenDX and the German engineering association VDI/VDE.

Third-party estimates, not numbers Elicit has confirmed directly, put 2025 revenue around $18 to $22 million in annual recurring revenue, growing 60 to 75 percent year over year, with tens of thousands of paying subscribers. Treat those as informed estimates rather than confirmed company figures. But they line up with a company that just doubled its user base and its funding within about 18 months, so the trajectory checks out even if the exact dollar figure doesn't come from Elicit itself.

What Can You Actually Do With Elicit?

Elicit isn't one feature, it's a stack of them, each aimed at a different stage of a literature review.

• Search and summarize. Type a research question in plain English and get back a ranked list of relevant papers from an index of more than 138 million, each with a short, sourced summary.

• Chat with papers. Select specific papers and ask direct questions. Answers are grounded in the source text, with a citation you can click to verify sentence by sentence, so you aren't taking the AI's word for it.

• Extract. Build a structured data table across dozens of papers at once, pulling out sample sizes, methods, or any custom column you define, instead of copying numbers into a spreadsheet by hand.

• Systematic Review. A guided workflow built to PRISMA 2020 reporting standards. It screens papers against your inclusion and exclusion criteria, tracks the reason for every exclusion, and generates the flow diagram journals expect to see.

• Research Reports. Point it at a question and it drafts a structured report synthesizing findings across a set of papers, with every claim traceable back to its source.

• Research Agent. A more autonomous mode that runs multi-step research on its own, searching, reading, and drafting without a prompt for every single step.

• Research Alerts. Personalized alerts, up to 10 on the Pro plan, that flag new matching papers so you don't have to keep re-running the same search.

• API and MCP server. Lets teams plug Elicit's search and extraction straight into their own tools, rather than working through the web app by hand.

How Well Does It Actually Perform?

We did not run a paid Elicit subscription for this review, so what follows is based on Elicit's own published benchmarks and documented features, checked against the company's technical write-ups rather than its marketing copy.

On the numbers Elicit has published for its Systematic Review workflow, search recall against Cochrane benchmark reviews sits at 95 percent. Abstract screening reaches 97 percent sensitivity and 93 percent specificity. Full-text screening lands at 99.5 percent sensitivity, ahead of the 98 percent sensitivity Elicit measured for human dual reviewers doing the same task on the same papers. Data extraction reportedly hits 96 percent accuracy on core fields like methods and participants, with a 1.0 percent hallucination rate measured across more than 5,100 data points, a rate the company says matches experienced human reviewers.

That's a different pitch from most tools calling themselves "AI research assistants." Most promise a fast summary and stop there. Elicit is trying to prove its summary holds up against a trained human reviewer, and it has put the benchmark numbers in writing rather than hiding behind a "trusted by X researchers" badge. Whether the accuracy holds up in your specific field is worth testing on the free plan before you pay for anything, but publishing recall and hallucination rates at all is still unusual in this category.

Elicit Pricing: What You Actually Get

(Verified directly on elicit.com/pricing.)

Is it worth the price? For casual use, the free plan is genuinely enough, most people underestimate how much of Elicit you can use without paying a cent. For anyone running a systematic review as part of a job, a grant, or a publication, $49/month is a rounding error next to the hours it can save on screening and extraction. Where it gets harder to justify is the jump from Pro to Scale: $120 a month more, mostly for team seats and a wider extraction table, which only pays off once you actually have a team using it.

Elicit vs. Consensus vs. Perplexity vs. NotebookLM

Not every tool that gets called an "AI research assistant" is built for the same job, and the differences matter more than the category label suggests.

Our Consensus review covers the closest direct competitor to Elicit: both search academic papers specifically, but Consensus is built for a fast, single sourced answer, while Elicit is built for the full review process, screening, extraction, and reporting included.

Our Perplexity review is the better read if you want one AI assistant for everything, news, general web search, and the occasional paper, rather than a tool that only knows academic literature.

Our NotebookLM review is worth a look if your problem is a pile of PDFs and notes you already have, since NotebookLM is built to organize and summarize your own sources rather than search millions of new ones for you.

Who Should Actually Use Elicit?

Best for:

• Grad students and PhD candidates working through a literature review

• Research teams running a systematic review for a publication or a grant

• Pharma, medtech, and consulting teams that need PRISMA-grade rigor and an audit trail

Skip it if:

• You want a general-purpose AI assistant for everyday writing, coding, or planning, a broader chat tool will be cheaper and more flexible

• You only need one or two sourced facts occasionally, Consensus or Perplexity will get you there faster

Final Verdict: Is Elicit Worth It?

If your work or your degree involves reading dozens of academic papers a month, yes, Elicit earns its $49. The free plan alone is worth trying before you judge it, and the published accuracy numbers behind the Systematic Review workflow are hard to find anywhere else in this category.

One honest note before you go looking for a discount code. We could not find a public affiliate or referral program for Elicit on elicit.com itself. A similarly named "Elicit Research" affiliate listing does exist on affiliate networks like Admitad, but it belongs to an unrelated survey company called Opinio Quest, not the AI research assistant this review is about. So there is no commission link anywhere in this post. We are recommending Elicit because the funding, the published benchmarks, and the pricing held up under a close look, not because a link would pay us for it.

Try the free plan at elicit.com and see for yourself whether the Systematic Review workflow saves you the hours it claims to.