Consensus (academic AI search)
Last reviewed
Sources
36 citations
Review status
Source-backed
Revision
v4 ยท 4,015 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
36 citations
Review status
Source-backed
Revision
v4 ยท 4,015 words
Add missing citations, update stale details, or suggest a clearer explanation.
Consensus is an AI-powered academic search engine that uses large language models to find, summarise, and synthesise findings from peer-reviewed scientific literature across a corpus of more than 200 million research papers. It is best known for its "Consensus Meter", a visual breakdown of how the literature trends "yes", "no" or "possibly" on a binary research question, and for being one of the most widely used research GPTs in OpenAI's GPT Store. By 2026 Consensus reported more than 8 million users and was used by researchers from over 10,000 universities. [17] The company is based in Boston and was founded in 2021 by former Northwestern football teammates Eric Olson and Christian Salem; the standalone web app launched in September 2022. [1][22]
Consensus indexes papers primarily aggregated from the Semantic Scholar corpus, OpenAlex and the company's own crawls. It launched a ChatGPT plugin in September 2023, and shipped a custom GPT in the GPT Store in November 2023 after OpenAI announced GPTs at DevDay. [9][10] The company has raised about $15 million in disclosed funding, including a $3 million seed in April 2023 led by Draper Associates and an $11.5 million Series A in July 2024 led by Union Square Ventures. [2][5] At the Series A the company reported "400,000+ monthly active users and nearly $2 million in annualized revenue, growing the latter by over 600% in 2024 alone", and by 2026 OpenAI reported that Consensus had "grown to more than 8 million researchers worldwide and increasing revenue by 8x". [5][17]
Consensus was founded in 2021 by Eric Olson (CEO) and Christian Salem (Chief Product Officer). The two had been teammates on the Northwestern University football team, where Olson played from 2012 to 2017 and Salem was a quarterback and wide receiver. Olson holds a B.A. in organisational change (2016) and an M.S. in predictive analytics (2017) from Northwestern, and worked as an analytics specialist at the daily fantasy sports company DraftKings before starting Consensus. Salem, also a 2016 Northwestern graduate, spent roughly three years in product management at the National Football League before joining Olson full time. [22]
In interviews Olson has framed the company around a personal limitation rather than a technical one: "We are consumer technologists who are the children of academics, teachers, and researchers. We deeply value research and evidence, but we've never had the skills nor attention span to comb through the research ourselves." [5] Salem has dated his own conviction to early experiments with language models: "I had been following the advancements of AI and became convinced of its power and utility after playing with GPT-3 in 2020." [5] The idea grew out of arguments with friends about diet, sleep and other health topics, where Olson kept finding himself sifting through Google results to back up a claim with peer-reviewed evidence. He wanted a search engine that would do the sifting for him: take a question in plain English, search across the scientific literature, and return what the studies actually say. By the end of 2021 both founders had quit their jobs to build the prototype full time. The first outside check came from Winklevoss Capital in a pre-seed round in July 2021.
Consensus opened to the public in September 2022. [1] The first version of the product was a relatively simple academic search engine: users typed a research question, and Consensus returned a ranked list of relevant papers along with a short, model-extracted finding pulled directly from each abstract. Crucially, every finding was a verbatim quote from the paper rather than a paraphrase, which made it easy to verify and harder for the model to fabricate.
The early product used in-house fine-tuned models to extract findings, but the launch happened to coincide with the rise of larger general-purpose language models. Within a few months, Consensus pivoted to using OpenAI's GPT-3.5 and then GPT-4 for extraction and synthesis, and added the now-signature Consensus Meter feature in 2023. [12][15]
On March 23, 2023 OpenAI introduced its first batch of ChatGPT plugins, with eleven launch partners including Expedia, Instacart, Kayak, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram and Zapier. [18][20] Consensus was not in that initial cohort, but it was approved as a third-party plugin shortly after and made generally available to ChatGPT Plus subscribers on September 23, 2023. [9] The Consensus plugin let users ask research questions inside ChatGPT and have the model fetch real findings, with citations, from Consensus's index of over 200 million papers. It quickly became one of the most-used research-focused plugins.
After OpenAI announced custom GPTs at its November 2023 DevDay, Consensus released ResearchGPT (later renamed simply "Consensus") as a custom GPT on November 14, 2023. [10] When OpenAI launched the GPT Store on January 10, 2024, Consensus was one of the products OpenAI featured by name. OpenAI shut down the original ChatGPT plugin system in two stages: new plugin chats were disabled on March 19, 2024, and existing plugin chats were retired on April 9, 2024. [21] The Consensus team migrated traffic to the custom GPT before the shutdown, which by then was already the more popular surface. [11]
Consensus closed an $11.5 million Series A on July 23, 2024, led by Union Square Ventures with participation from Nat Friedman and Daniel Gross (both early backers of Perplexity), as well as previous seed investors Draper Associates, Kevin Carter, Path Ventures and Alumni Ventures. [5][6] At the time of the announcement the company reported "400,000+ monthly active users and nearly $2 million in annualized revenue, growing the latter by over 600% in 2024 alone". [5][7]
Through late 2024 and into 2025 the team rolled out Pro Analysis (a Copilot-style synthesis of up to 50 papers), Study Snapshot (an automated extraction of population, sample size, methods and outcomes for a single paper) and Deep Search, a multi-agent system. By 2026 Consensus reported more than 8 million total users across the website and GPT Store, used by students, faculty and researchers from over 10,000 universities, with over 170 university libraries partnering to provide institutional access. [17] In an OpenAI customer story published in 2025, Consensus described its Scholar Agent as a multi-agent system "built on GPT-5 and the Responses API" that "mirrors how researchers actually work, helping them get from question to conclusion in minutes instead of weeks": a Planning Agent decomposes the question and decides which actions to take next, a Search Agent combs the paper index, a user's private library and the citation graph, a Reading Agent interprets papers individually or in batches, and an Analysis Agent synthesises results and composes the final output. [17] OpenAI reported that over the year to mid-2026 Consensus had "grown to more than 8 million researchers worldwide and increasing revenue by 8x". [17]
The core search pipeline has stayed roughly the same since the 2023 redesign, with new layers added on top. [14][32]
The Consensus Meter is the product's most recognisable feature. It requires at least five relevant papers to render, colour-codes each cited paper to match its stance and lets users drill into the methodology, recency, journal and impact of the underlying studies. [13] The company is open about its limits: the meter does not always honour conditions in the original question (a result about ibuprofen safety in children might still be classified as "yes" for an adult-safety query) and it is not designed for highly specific clinical decisions. A separate "medicine mode" narrows the search to roughly 8 million papers from the top 1,000 medical journals plus about 50,000 clinical guidelines for clinically focused questions. [32]
| Round | Date | Amount | Lead investor | Other notable participants |
|---|---|---|---|---|
| Pre-seed | July 2021 | undisclosed | Winklevoss Capital | |
| Seed | April 2023 | $3.0M | Draper Associates | Winklevoss Capital, Kevin Carter, Brian Pokorny, Nomad Capital, Alumni Ventures, Des Traynor (Intercom), Rob May (AI Operator Fund), Billy Draper (PathVC), Kindergarten Ventures, David Dohan (OpenAI Research) |
| Series A | July 2024 | $11.5M | Union Square Ventures | Nat Friedman, Daniel Gross, Draper Associates, Kevin Carter, Path Ventures, Alumni Ventures |
Total disclosed funding by mid-2024 was approximately $15 million. [5][24] The seed announcement noted that the round, combined with the Winklevoss pre-seed, brought total funding to roughly $4.25 million at that point. [2]
Consensus has a free tier and two paid tiers, with pricing that has shifted slightly over time as new agentic features were added.
| Tier | Approximate price (2025) | What you get |
|---|---|---|
| Free | $0 | Unlimited basic searches, basic Consensus Meter, basic study filters, access to Consensus GPT in the GPT Store |
| Premium | around $8.99/month (annual) or $11.99 monthly | Unlimited GPT-4 powered summaries, unlimited Consensus Meter, advanced filters, Pro Analysis, Study Snapshot |
| Enterprise / Teams | custom | Institutional access for universities and companies, single sign-on, admin tools, library integrations |
Key product surfaces include:
The path through OpenAI's developer surfaces is part of why Consensus is so widely known to general consumers, not just academic researchers.
| Date | Event |
|---|---|
| March 23, 2023 | OpenAI launches ChatGPT plugins in alpha with 11 launch partners. Consensus is not in the initial cohort. |
| September 23, 2023 | Consensus ChatGPT plugin is generally available to ChatGPT Plus subscribers, exposing the 200M-paper index inside ChatGPT. |
| November 14, 2023 | Consensus releases ResearchGPT, a custom GPT, days after OpenAI announces GPTs at DevDay. |
| January 10, 2024 | OpenAI launches the GPT Store. Consensus is among the featured products. |
| March 19, 2024 | OpenAI disables new chats with the original plugin system. |
| April 9, 2024 | OpenAI fully retires ChatGPT plugins. Consensus traffic has by then largely moved to the custom GPT. |
| 2025 | Consensus releases the Scholar Agent, a multi-agent deep-research system built on GPT-5 and the OpenAI Responses API. |
Consensus markets itself as a tool for "finding evidence-based answers fast". In practice the most common use cases are:
The AI-powered research tool space exploded between 2022 and 2026. The table below sketches how Consensus sits relative to the major alternatives.
| Tool | Vendor | Year | Focus | Peer-reviewed only | Free tier | LLM/GPT integration | Key feature |
|---|---|---|---|---|---|---|---|
| Consensus | Consensus (Olson, Salem) | 2022 | Plain-English Q&A over peer-reviewed papers | Yes | Yes | GPT-4 / GPT-5 via OpenAI API; custom GPT | Consensus Meter for yes/no questions |
| Elicit | Elicit (formerly Ought) | 2021 | Literature review automation, data extraction | Mostly | Limited | Multi-model, including Claude | Literature matrix / data extraction across many papers |
| Scite.ai | Scite | 2018 | Citation context ("smart citations") | Yes | Limited | Yes | Classifies citations as supporting, mentioning or contrasting |
| Perplexity | Perplexity | 2022 | General web search with citations | No | Yes | In-house and frontier models | Conversational web search with sources |
| ChatGPT Deep Research | OpenAI | 2025 | Long-form general research reports | No | ChatGPT Plus / Pro | Built into ChatGPT | Multi-step browsing agent |
| Gemini Deep Research | 2024 | Long-form general research | No | Gemini Advanced | Built into Gemini | Plans, browses and writes reports | |
| Semantic Scholar | Allen Institute for AI | 2015 | Academic paper search and metadata | Yes | Yes | TLDR summaries; corpus underlies many AI tools | Open corpus and citation graph |
| SciSpace | Pubble.ai | 2020 | Paper search, summarisation, chat with PDFs | Yes | Yes | GPT-3.5 / GPT-4 | Chat-with-PDF for any paper |
| Wolfram | Wolfram Research | 1988 (Wolfram Alpha 2009) | Computational knowledge | No | Yes | ChatGPT plugin / GPT | Symbolic computation, not paper search |
| Google Scholar | 2004 | Academic search | No (but mostly scholarly) | Yes | None | Largest free academic index, no AI synthesis | |
| PubMed | NIH / NLM | 1996 | Biomedical literature | Yes | Yes | None natively | Authoritative biomedical index |
| JSTOR / EBSCO | JSTOR / EBSCO | various | Subscription academic databases | Yes | No | Some AI features | Deep journal archives |
A frequently quoted workflow among power users is to "explore with Perplexity, validate with Consensus, synthesise with Elicit, verify citations with Scite". Consensus is rarely positioned as a replacement for any of those tools; its strongest pitch is the combination of peer-reviewed-only scope, verbatim quotes and the Consensus Meter.
No AI research tool is a substitute for reading the underlying papers, and Consensus is no exception. Common criticisms include:
Consensus reached around 200,000 registered users by the seed-round announcement in April 2023, more than 400,000 monthly active users by the Series A in July 2024, and more than 8 million total users by 2026 across the website and GPT Store. [2][5][17] Adoption has been particularly strong in academic libraries: over 170 university libraries partner with the company, and Yale, the University of Virginia, Loyola Marymount, Oklahoma State, Bentley, George Mason, Texas A&M, the University of St. Thomas, ETH Zurich and HKUST are among institutions that have written guides recommending or trialling the tool. [17][27][29][30] The Allen Institute for AI's Semantic Scholar, the underlying corpus provider, also collaborated publicly with Consensus. [33]
Coverage has been broadly positive but cautious. A 2025 systematic review in PMC concluded that Consensus is useful for early-stage literature scoping but not a substitute for human-led systematic review. [31] Independent reviewers including Aaron Tay (a librarian who writes widely about academic search tools) have praised the product's transparency and the Consensus Meter while warning that the binary framing can mislead in subtle clinical or policy questions. [32] In TechCrunch, VentureBeat, MobiHealthNews and Inside Higher Ed coverage of the funding rounds, the dominant framing is that Consensus is one of the more credible attempts to apply LLMs to research without falling into the hallucination trap that has plagued general chatbots in academic settings. [4][8]
Key developments since the Series A include:
The broader competitive context has shifted as well. ChatGPT Deep Research (OpenAI, February 2025) and Gemini Deep Research (Google, December 2024) both produce long, agentic research reports across the open web, but neither is restricted to peer-reviewed sources. Consensus continues to bet that a corpus-curated, citation-traceable approach will remain the right one for high-stakes research questions, even as general-purpose research agents grow more capable.