Consensus turns research questions into a reviewable evidence map

Consensus is a research search and synthesis platform built around scholarly papers. Its practical value is reducing the distance between a question and a set of studies worth reading. A useful result includes more than a fluent answer: it identifies the papers behind that answer, the population and outcomes they examine, and the points where their findings do not line up. Consensus gives researchers several ways to move through that process, but the researcher still decides what the evidence supports.

01 / ProductWhat Consensus actually does

The product combines paper search with AI-assisted synthesis. Papers mode provides a results list; Pro produces a response based on a smaller group of papers; Deep Review carries out a broader sequence of searches and creates a longer report. Users can shape the question and requested output instead of accepting one fixed summary format. The search guide³ describes natural-language questions, keywords, Boolean searches and corpus choices, including a library of saved or uploaded material.

These modes suit different stages of research. A paper list is useful when you already know the terminology and want to judge the candidate studies yourself. A synthesis helps when you need an overview of how findings relate. A deeper review can expose themes and disagreements before you invest in a detailed reading programme. Moving between these stages is more useful than treating the longest generated answer as the best answer.

The unit of evidence remains the study. Several papers may reuse a dataset, report different outcomes from one experiment or cite the same earlier finding. A polished report can make those relationships look simpler than they are. Keep enough study-level information to distinguish independent evidence from repeated discussion of the same evidence.

02 / AudienceWho should use it

Consensus fits people who regularly ask questions that published research can help answer: an education researcher designing a literature review, a product researcher examining human-computer interaction, or an analyst investigating an established technical method. The strongest starting question specifies a population, intervention or comparison, an outcome and a time frame. That structure gives both the search and the reader something concrete to evaluate.

It is less useful as the only discovery tool for a brand-new product launch, unpublished industry practice or a topic whose important material lives outside scholarly publishing. For that broader public-web task, the Perplexity blueprint offers a relevant comparison. Use the tool whose source universe matches the question before comparing the quality of its prose.

For a formal evidence review, a team also needs a reproducible search plan, inclusion decisions and an extraction record. Consensus can contribute candidate papers and a working map, while the Elicit blueprint is worth considering when structured extraction and a systematic review workflow dominate the job. The decision is about the work surrounding the answer, including how another researcher will inspect the result.

03 / WorkflowA workflow for reviewing classroom research

The following is a proposed workflow for an education team examining spaced practice. Start with a question such as how spaced and massed practice compare for delayed retention among secondary-school students. Define the subject, age range, intervention duration and minimum delay before assessment. A study measuring immediate quiz performance should not silently become evidence for retention several months later.

Establish the candidate set

Run a focused Papers search first. Record terminology used by the relevant studies and look for review papers that reveal older names for the same teaching approach. Try those terms in additional searches. Save plausible inclusions with a short reason, and keep a separate list of exclusions. The aim is to understand what the search is finding before asking the model to organise it.

Then use Pro for a targeted comparison: ask for the studies grouped by outcome timing, with disagreements and unresolved questions separated. The Pro guide⁵ supports customised instructions and evidence-oriented outputs such as claims-and-evidence tables. Ask for the comparison fields that matter to your decision instead of requesting a generic account of the topic.

Inspect the evidence behind the synthesis

Use Deep Review when the first pass reveals several connected subquestions. Its documentation² describes breaking a question into searches, screening a larger candidate pool and typically analysing around 50 papers, sometimes more. These are workflow characteristics, not a promise that the report covers every relevant publication. A narrower question may produce a more useful report than a broad topic with an impressive paper count.

Open the citations that support the central comparison. A checkmark indicates that full text was used; its absence indicates an abstract-based source. That symbol does not certify study quality. For the classroom example, inspect assignment method, sample size, outcome measurement and whether the comparison groups received similar teaching time. Record the actual finding beside the interpretation you plan to use.

Consensus also provides a table view with fields such as population, methods, results and duration. The results-view guide⁴ says cells can remain blank when information cannot be extracted. Treat a blank as an unresolved extraction question. It does not mean the study had no sample, did not measure an outcome or reported a value of zero.

Finally, export the report and paper list into the team's working record. Keep the search question, filters, date and inclusion decisions alongside it. Select three disputed or influential findings for independent reading by another colleague. The useful output is a traceable evidence map that helps the team decide what to read next and which claims are ready to carry into its own work.

04 / PricingPricing and the limits that matter

Consensus separates ordinary research use from the more expensive Deep Review allowance. The figures below are in US dollars and come from the subscription guide¹, accessed on 15 September 2026. Annual prices are commitments for the year.

Deep Review demand is the clearest reason to move beyond Pro. A researcher preparing one substantial review may fit comfortably inside Pro's allowance, while a team comparing many questions each week may not. Count distinct questions and expected reruns after refining scope. One broad report followed by several narrower reviews can consume more allowance than the initial project plan suggests.

The results table has a separate limit: the current guide lists 3 papers per query on Free, 20 on Pro and 50 on Deep. That can matter if your practical output is a comparison table rather than a narrative. A higher Deep Review allowance and a larger table view solve different problems; assess both against a representative assignment.

API or MCP use should be budgeted separately from interactive research habits. An automated assistant can call a research tool repeatedly in ways that a person using the website would not. Before adding it to a shared workflow, decide which steps genuinely need another search and which can reuse an already collected paper set.

05 / DistinctionsWhat stands out in practice

The most useful design choice is the ability to move between the answer and the underlying papers. A research summary is easier to challenge when its citations are close to the claims and its evidence can also be viewed as rows. That makes the product suitable for an exploratory reading cycle: ask, inspect, refine and ask a narrower question.

Consensus also makes the distinction between abstract and full-text use visible. This matters because an abstract may omit a subgroup result, an important limitation or a measurement detail that changes the interpretation. The indicator helps readers prioritise where to open the original paper. It should prompt more careful reading of a consequential claim, rather than serve as a score for the whole report.

A Consensus Meter can provide a quick view of how selected findings are classified around a question. It becomes less informative when the question bundles several populations or outcomes. Before interpreting a stance distribution, ask whether the papers are actually answering the same question. Agreement about immediate performance and disagreement about long-term retention can coexist without the literature being internally inconsistent.

The product's value grows when the user brings a good comparison framework. A table that separates participant age, intervention dose and assessment delay is more informative than a list of positive and negative studies. The tool can accelerate the assembly of that framework, while domain knowledge determines which distinctions belong in it.

06 / QuestionsQuestions to resolve before adopting it

Start with coverage. Choose a small set of important papers already known to your team and see how they appear across keyword searches and natural-language questions. Check whether a relevant paper is absent, merely ranked lower or excluded by a filter. Those are different failure modes and call for different changes to the research process.

Next, inspect a few difficult extractions. Papers with multiple experiments, several follow-up periods or complex comparison groups are good candidates. Compare the table and synthesis with the original methods and results. A useful evaluation records the precise error or missing field, because an overall impression of accuracy does not show which work still needs manual attention.

Finally, decide how research records move between collaborators. A copied answer alone loses the reasoning behind inclusion and exclusion. Keep a shared place for the question, candidate set, source documents, extraction decisions and final interpretation. That record also makes it easier to update the review when a new study changes an important conclusion.

07 / DecisionThe decision on Consensus

Choose Consensus when your bottleneck is finding and organising relevant scholarly evidence into a first, inspectable picture. Its modes offer a sensible progression from candidate papers to targeted synthesis to a broader review, and its table view gives researchers another way to examine the material.

Choose the plan around the repeated task you actually have. Pro can suit regular research with occasional deeper reports; Deep makes more sense when broad reviews and larger comparison tables recur. For a formal review, test how well the exported material fits your team's screening and extraction process before making it the centre of that process.

The best outcome is a better reading and reasoning workflow. A useful Consensus report gives you specific papers to inspect, clearer distinctions between their findings and a sharper next question. That is a stronger basis for adopting the tool than the apparent completeness of a single generated response.