Consensus

Consensus is an AI search engine built around a specific question type: what does the published evidence actually say about this claim? You ask something answerable — does intermittent fasting reduce blood pressure, does this pathway drive that phenotype — and it returns findings extracted from relevant papers, each tied to its source, with an indication of how the literature distributes across yes, no and mixed.

What it does

It draws on a corpus of over 200 million peer-reviewed papers assembled from Semantic Scholar, OpenAlex and direct publisher partnerships. The distinguishing feature is the consensus meter, which summarises the balance of findings across the retrieved studies rather than presenting the first plausible answer. Results carry study type and journal quality signals, so a claim supported only by small observational work is visibly different from one supported by trials.

Where it fits

This is a triage and orientation tool. It is fast, it is good at telling you whether a question is settled or contested, and it is useful when you are working just outside your own field and need to know the shape of the evidence before committing time. It is not a systematic review tool — for structured screening and extraction, Elicit is the better instrument.

Strengths

• Large corpus with credible provenance rather than an opaque index.
• Every finding links to the paper it came from.
• The consensus view surfaces disagreement instead of hiding it behind a confident summary.
• Study-type and quality signals attached to results.
• Fast, and genuinely usable on the free tier.

Limitations

The consensus meter is a convenience, not a meta-analysis — it counts papers rather than weighting them by quality, sample size or risk of bias, and a well-populated bar can look more authoritative than the underlying evidence deserves. It works best on questions phrased as testable claims and poorly on open exploratory queries. Extraction can flatten the caveats that make a finding conditional. As with every tool of this kind, the retrieved paper is the evidence; the summary is a pointer to it.

Verdict

Excellent for orientation and for checking whether a claim you have encountered is actually supported. Treat the consensus bar as a prompt to go and read, never as a result you can cite.