AI-powered research assistant that finds answers in scientific papers.
Research
Consensus is the AI search engine for scientific literature — a Perplexity-shaped product built exclusively on top of the peer-reviewed corpus. Founded in 2021 by Eric Olson and Christian Salem out of Wharton, Consensus indexes 200+ million research papers from Semantic Scholar, PubMed, and adjacent scholarly sources, and answers natural-language research questions with citations that all point back to real, verifiable peer-reviewed studies. It has become the default "actual literature" tool for researchers, clinicians, graduate students, and any knowledge worker who has to answer "what does the evidence say about X" and cannot afford to accept whatever a general chatbot hallucinates back.
What Consensus actually does, cleanly stated, is turn the peer-reviewed literature into a conversational research assistant. You ask a real evidence question — "does intermittent fasting improve insulin sensitivity in adults with prediabetes," "how does semaglutide compare to tirzepatide for weight loss maintenance," "what is the current evidence for creatine supplementation in cognitive aging" — and Consensus returns a synthesized answer with a Consensus Meter showing how the evidence distributes across yes / no / possibly, backed by individual paper cards with study type, sample size, and one-sentence findings. Every claim is anchored to a specific paper you can click into and read. Every paper card tells you whether it is a randomized trial, a meta-analysis, an observational study, or an animal model — the details that separate real evidence from evidence-shaped noise.
The one-line positioning: Consensus is the AI research tool for people who have to answer questions correctly, not just quickly. Clinicians deciding on a treatment. Grad students writing a lit review. Health journalists fact-checking a claim. Researchers scoping a new project. If your job includes the phrase "what does the evidence actually show," Consensus is the tool that puts real, cited evidence in front of you in five seconds instead of five hours of PubMed queries. It is not a general assistant. It is a specialist that beats every generalist on the specific task.
Consensus has widened its product carefully since 2024 — enough to justify a subscription, not so much that it lost its specialist edge. The features that earn the subscription are the Consensus Meter, GPT-powered synthesis, and study-type filters.
Consensus Meter. For yes/no research questions, a visual signal that shows how the top 20 studies distribute across "yes," "possibly," and "no." Not a substitute for reading the literature, but a fast triage tool that tells you whether a claim is broadly supported, actively contested, or barely studied.
Peer-reviewed corpus only. The index is 200+ million research papers from Semantic Scholar and adjacent sources — no SEO blog posts, no Medium articles, no AI-generated summaries pretending to be journalism. Every citation is a real paper you can pull the DOI on.
Study Snapshots. Every paper card in a search result shows the study type (RCT, meta-analysis, systematic review, observational, animal model), the population, the sample size, and a one-sentence AI-generated finding. This is the feature clinicians and researchers rely on to triage — you can skim 20 studies in a minute and know which two to read in full.
AI-generated syntheses. Above the paper cards, Consensus produces a cited paragraph synthesis of the evidence, powered by GPT-class models running on the retrieved corpus. This is genuinely useful for lit-review scaffolding — a first-draft "what does the evidence say" paragraph you refine yourself.
Study quality signals. Consensus flags high-quality journal indicators (h-index, Scimago quartile) so you can tell at a glance whether a paper is from a top-tier venue or a predatory publisher. Not perfect, but useful triage.
Filters that matter. Filter by study type (RCT, meta-analysis, review), publication year, human vs. animal population, and specific journals. This is the "let me narrow to the last five years of human RCTs" workflow that PubMed makes possible and slow, and Consensus makes fast.
Paper deep-dive. Click into any paper for a longer AI-summary of methods, findings, and limitations, plus a chat interface to ask specific questions about the study. This is where Consensus becomes a real reading assistant, not just a search engine.
Bookmark and export. Save papers, tag them into projects, export lists to Zotero or Endnote for later citation work.
Consensus is freemium and deliberately priced accessible to individual researchers.
| Plan | Monthly | Annual (per month) | Included |
|---|---|---|---|
| Free | $0 | — | Unlimited basic search, ~10 AI syntheses per month |
| Premium | $9.99 | $8.99 ($107.88/year) | Unlimited syntheses, Consensus Meter, GPT-4 |
| Teams | Custom | Custom | Shared workspaces, admin, higher limits |
| Enterprise | Custom | Custom | SSO, data privacy, institutional access |
Premium at $9.99/month (or $8.99/month effective on annual) is the tier working researchers land on. It unlocks unlimited AI-generated syntheses, unlimited Consensus Meter runs on yes/no questions, access to GPT-4-class synthesis (versus a lighter model on free), higher search-result depth, and priority speed on the corpus. The pricing is genuinely aggressive — under $10/month is a real bargain for anyone whose day involves literature review.
Teams and Enterprise are quote-only and add shared research workspaces (like Perplexity Spaces, but tuned for the scholarly workflow), admin controls, institutional single-sign-on, and contractual data privacy guarantees. Universities and hospital systems have been signing these since 2024.
The honest read: Consensus Premium at $9/month is the cheapest useful academic research subscription on the market. If your work involves any recurring "what does the evidence say" query — clinicians, journalists, grad students, health researchers, product managers in regulated industries — the subscription pays for itself in the first hour of avoided PubMed labor.
Pros
Cons
Clinicians answering point-of-care questions. "What is the current evidence for X treatment in Y population" is exactly the query Consensus handles better than any general chatbot. Study type filters and quality signals make triage fast enough for real clinical workflow.
Graduate students writing literature reviews. The AI-synthesis-with-citations output is genuinely useful as first-draft scaffolding. Consensus does not write your lit review — but it collapses the "find 30 relevant papers" phase from a week to an afternoon.
Health and science journalists fact-checking claims. When a source claims "studies show X," Consensus tells you in a minute whether studies actually show X, are mixed on X, or barely touch X.
Researchers scoping new projects. Before committing to a research question, Consensus shows you the shape of the existing literature — is this well-studied, contested, or a gap. Prevents six months on a settled question or, worse, a question no one is asking.
Product managers in regulated industries. Healthtech, biotech, and nutrition PMs need real evidence for product claims. Consensus lets a non-clinical PM triage the literature well enough to know when to bring in a domain expert.
Curious professionals with expensive time. Anyone whose "am I healthy" or "should I take this supplement" or "does this therapy work" question deserves a real answer, not a wellness-blog answer. Consensus is the answer engine for that use case.
Consensus's real competition is the tools it replaces: PubMed and Google Scholar, general-web chatbots, and specialist academic research tools.
Perplexity with Academic Focus — the closest general-purpose competitor. Perplexity's Academic mode filters to Semantic Scholar and arXiv, and the answer synthesis is strong. Consensus has cleaner triage UX, better study-type filters, and a peer-review-only guarantee. Many researchers pay for both.
Elicit — the other big AI academic tool. More focused on structured literature review workflow — extracting specific data across dozens of papers into a table. Consensus is faster for triage; Elicit is deeper for systematic review work. Complementary.
SciSpace — the AI paper-reading tool. Consensus finds papers; SciSpace helps you understand them once open. Complementary — many working researchers use both.
Semantic Scholar — the free academic search engine Consensus is built on top of. No AI synthesis, no Consensus Meter, but free and comprehensive. If you already know exactly what you are looking for, Semantic Scholar is enough.
PubMed and Google Scholar — the incumbents. Free, comprehensive, and requires the researcher to do all the synthesis work themselves. Fine for narrow searches; painful for evidence questions.
Scite.ai — the "smart citations" tool that shows whether papers support, mention, or contrast a claim. Different angle on the same problem. Also worth stacking.
Sign up free at consensus.app. No credit card required. Ask a real evidence question you actually care about — a health, nutrition, therapy, or research question you have wondered about but never fully answered.
Watch the Consensus Meter on a yes/no question. Ask something contested — "does daily coffee reduce Alzheimer's risk" — and note how the meter breaks down. This is the feature you are actually evaluating.
Use the filters. Rerun the same query filtered to "meta-analysis" or "human RCT only" and see how the picture sharpens. This is how working researchers actually use the tool.
Upgrade to Premium if the tool sticks. At consensus.app/pricing you get unlimited AI syntheses and Consensus Meter for $9.99/month or $107.88/year. Annual is the right choice — at under $9/month, the subscription is a no-brainer if you use it weekly.
Stack it with a paper-reading tool. Consensus finds the papers; use SciSpace or Elicit to actually work through the ones that matter. The two-tool workflow is the mainstream setup for real literature work.
Is Consensus more accurate than ChatGPT for research? For evidence questions grounded in peer-reviewed literature, yes — and the difference is not close. ChatGPT hallucinates citations. Consensus cites real papers you can click. For non-literature questions, ChatGPT is broader.
Which model powers Consensus? The AI synthesis is powered by GPT-4-class and Claude-class models running on top of Consensus's retrieved paper corpus. Model providers rotate. The Consensus Meter uses in-house classifiers trained on scientific paper structure.
Does the corpus cover all fields? Coverage is strongest in health, biology, psychology, and adjacent life sciences — the fields Semantic Scholar and PubMed index deeply. Coverage is thinner in humanities and some social sciences. Physical sciences (physics, chemistry) are well covered but arXiv-heavy areas can lag.
Are the AI syntheses trustworthy? Usually. Not always. Consensus occasionally misrepresents studies with counter-intuitive findings or oversimplifies contested evidence. Treat the synthesis as a starting point; click through and read the actual paper when the stakes matter.
Can I access the full text of the papers? Consensus surfaces abstracts, metadata, and study snapshots on every paper. Full-text access depends on whether the paper is open access or whether you have institutional subscriptions. Many top journals are still paywalled.
Can I use Consensus for commercial or clinical decision-making? Consensus is a research and triage tool, not a clinical decision support system. Nothing in its terms replaces professional judgment or systematic review methodology for high-stakes decisions. Use it to accelerate the work, not to replace expertise.
Consensus Premium at $9/month is one of the highest-value AI subscriptions on the market, and the only serious answer for "AI that actually cites real peer-reviewed evidence." For clinicians, researchers, grad students, journalists, and anyone whose day includes real evidence questions, Consensus collapses hours of PubMed labor into minutes and puts real study-type-and-sample-size context in front of you before you commit to reading the paper. The peer-reviewed-only guarantee is the feature every general chatbot lacks — and the reason Consensus stays the daily driver for professionals who cannot afford a made-up citation.
Where Consensus stops being the right answer: research questions outside the peer-reviewed literature (market data, business news, tech topics, humanities), casual chat and drafting tasks (any general chatbot is broader), and full-text deep reading (SciSpace and Elicit are stronger for the reading phase). Consensus is the finder, not the reader — the "here is the evidence, ranked, filtered, and triaged" tool.
The honest recommendation: try the free tier on three real evidence questions from your actual work. If the answers save you meaningful PubMed time and the Consensus Meter tightens your thinking on contested questions, upgrade at consensus.app/pricing. Stack it with Perplexity Pro for general research ($20/month) or SciSpace for paper reading ($8/month). At under $10, Consensus is the specialist that earns its subscription in the first working week.