Consensus: our review

Consensus is a scientific search engine that answers a question with a summary of the relevant papers, not an opinion. You ask it "does meditation reduce anxiety?" and it returns what the literature says, study by study, with the methodology and citation count of each.

What it actually does

The way it works reads on three levels. The first is simple search: you ask a question in plain language, and Consensus queries a database of more than 200 million scientific papers. It returns a list of results ranked by relevance, each paper accompanied by an excerpt of its abstract, its study design (randomized controlled trial, meta-analysis, observational study...), its citation count, and a badge showing whether the study supports or contradicts the starting hypothesis.

The second level is the Pro summary: the tool aggregates up to 20 papers and writes a summary of what the literature says as a whole, with the matching references. This is the level that replaces an hour of reading abstracts with a thirty-second answer.

The third level is the deep-dive analysis (Deep): up to 50 papers reviewed for a complex question. The tool extracts the key data — sample size, study duration, population studied — and lays it out in a comparison table. This is the level suited to a thesis, a publication, or a clinical decision.

Why it interests us (or not)

Consensus's strongest point in the AI-tool landscape is its reliability. It doesn't generate text: it indexes, searches, and summarizes. The difference with a general-purpose chatbot is fundamental: when ChatGPT confidently answers you with something untrue, you don't know it. When Consensus gives you a conclusion, you can click through to the paper and check for yourself. For a lawyer looking for the scientific case law behind a medical argument, or an accountant who wants to source a tax position, that traceability is decisive.

The no-training commitment is also written in black and white: "We do not use your data to train large language models. Your queries, content, and personal information are never used to train AI models — neither our own nor those of any third party." That's not a deduction, it's an explicit statement.

The main downside is location: all servers are in the United States. For a medical or legal practice that must document European hosting of its data, that's a dealbreaker. The nature of the data processed — search queries, not client files — reduces the risk, but doesn't remove it.

Where it falls short

Pricing is per user, with no real team plan before reaching quoted tiers. The Pro tier at $20 a month is individual: a team of five researchers costs $100 a month. The Deep tier at $65 a month includes 200 deep-dive analyses, but stays individual. For an SMB that wants to equip its R&D or monitoring team, the bill adds up fast with no visible economy of scale.

The lack of a French-language interface is another obstacle. Questions can be asked in French, but the summaries and paper excerpts are in English. A professional user who publishes in French has to systematically rephrase the results.

Lastly, the tool specializes in English-language scientific literature. For very French topics — tax law, local regulation, public policy — coverage is thinner, and a general-purpose tool paired with a national documentary database may be more relevant.

Our recommendation

Consensus is a remarkable tool for anyone who needs to quickly check what the science says on a topic, and to be able to prove it. We recommend it for individual use or for a pair, in a context where US hosting isn't a blocker. For team use or in a regulated sector, weigh the cumulative cost and the location constraint before committing. If you're looking for a more general-purpose monitoring and research tool, explore the alternatives we've reviewed.