CONSENSUS
// The search engine that read the studies first
The little blue meter that decided to settle arguments with peer review instead of opinion. Photographed here mid-thought, somewhere around its 200-millionth paper.
The little blue meter that decided to settle arguments with peer review instead of opinion. Photographed here mid-thought, somewhere around its 200-millionth paper.
Type "does intermittent fasting actually work?" into most search bars and you get a wall of links, half of them selling you something. Type it into Consensus and you get a different kind of answer: a tally of what the published research found, sorted by study quality, topped with a meter that tells you whether scientists mostly agree, mostly disagree, or are still arguing about it.
That is the whole company in one sentence. Consensus is an AI search engine wired directly into the body of human research - more than 200 million peer-reviewed papers - and built to hand back evidence instead of vibes. It is used by students writing their first lit review, clinicians checking a guideline, and skeptics who just want to win a dinner-table argument with a citation.
Here is the inconvenient truth Consensus was built around: the best answers to most important questions already exist. They are sitting in journals, written in dense academic prose, hidden behind subscriptions, and effectively unreadable to the people who need them most.
So most of us reach for a search engine and get the internet's loudest voice rather than its most rigorous one. A single cherry-picked study gets shared a million times; the ten studies that contradict it stay quiet in a database nobody opens. The information exists. The access does not.
The founders, who grew up in families of researchers and teachers, found this genuinely annoying - the polite word for the kind of annoyance that turns into a company.
Eric Olson and Christian Salem met as Division 1 college football teammates - not the usual origin story for an academic search engine. Olson went on to data science at DraftKings, where he learned how to make large messy datasets behave. Salem brought the operating discipline. In 2021 they made a bet that language models, then still a curiosity to most people, were about to get good enough to read science the way a careful researcher does.
It was an early bet. They partnered with OpenAI before that was a fashionable sentence, raised a $3M seed from believers including Nat Friedman and Daniel Gross, and started teaching machines to extract findings - not keywords - from papers.
Co-founder & CEO. Ex-DraftKings data scientist. Former Northwestern football starter who now wrangles 200 million papers instead of playbooks.
Co-founder & COO. The other half of the teammate-turned-cofounder story, keeping the operation moving while the meter spins.
Consensus is not a chatbot that confidently makes things up. Every answer points back to real papers, with the study design, sample size and quality flagged so you can judge for yourself. The signature feature, the Consensus Meter, does something refreshingly honest for a piece of software: it admits when the science is mixed.
A one-glance read on how much of the literature agrees, disagrees, or is genuinely undecided. The rare AI feature that says "it's complicated" out loud.
An assistant inside search that drafts, formats, and synthesizes referenced answers - footnotes included.
Per-paper summaries surfacing methodology, sample size and quality indicators before you commit to reading.
Automates a full search strategy across up to 1,000 papers for systematic-review-style work.
Lets developers ground their own apps in scientific evidence, priced per call.
Filters to clinical guidelines and top medical journals for people whose answers have real stakes.
Olson and Salem found Consensus to make peer-reviewed knowledge searchable in plain English.
Backers include Draper Associates, Nat Friedman and Daniel Gross. The product opens to the public.
Led by Union Square Ventures, to scale evidence-based search and the team.
Automated multi-step search across up to 1,000 papers, edging toward systematic-review territory.
Clinical-grade filtering launches, and Consensus shows up in OpenAI's ChatGPT apps ecosystem.
A company built on evidence should be willing to show some. Here is the scale Consensus is reasoning over and the audience that keeps coming back.
Bars scaled for drama, not for a peer-reviewed journal. Consensus would, fittingly, flag that.
Behind the figures sits a freemium model: search is free, the heavier AI features live in an ~$8.99/month Premium tier, teams and institutions pay for seats, and developers pay per API call. Universities have plugged it into their libraries. OpenAI features it. The validation is less about hype and more about the boring, durable fact that people keep using it to do actual work.
Consensus has stopped describing itself as merely a search box. The stated ambition now is bigger and a little cheeky: an AI operating system for researchers - the place where a question becomes a search, a search becomes a synthesis, and a synthesis becomes something you can actually cite.
The throughline never changes. Take the world's most trustworthy knowledge, which has always been technically public and practically locked away, and make it answerable in a sentence. Not dumbed down. Not hallucinated. Just findable.
Remember the question we started with - the one about fasting, or sleep, or whatever you last argued about and lost. The old way to settle it was to shout louder or Google until you found a headline that agreed with you. Both work about equally well, which is to say not at all.
Consensus changes the terms of that argument. Now the question goes in, and the science comes out: weighed, sorted, and honest about its own uncertainty. As AI gets better at reading, and as more of the world starts asking before asserting, that little blue meter looks less like a search feature and more like a habit worth building.
Type the question. Read the research. Then, if you must, win the argument.