How to Use Consensus

Direct answer

To use Consensus well, start with a narrow research question, search for relevant papers, inspect the original studies, and record what the evidence supports. Treat the generated explanation as a route into the literature. Your conclusion should come from the papers and your appraisal, rather than the fluency of the summary.

Consensus documents natural-language and keyword searching, including queries built from titles, authors and DOIs. Its official explanation of how search works describes literature retrieval before synthesis. This tutorial turns that documented model into an editorial workflow; it does not report results from a logged-in test account.

Who this is for

This guide is for students, analysts and working researchers who need to move from an unfamiliar topic to an organized set of papers. It also helps a team evaluate whether a research assistant makes its existing process clearer. You can follow the method with a small assignment before committing a large project to the tool.

Who should avoid it? Anyone expecting a search summary to settle an important decision without checking the underlying evidence should change that expectation first. A useful literature assistant still needs a reader who can separate a research finding from an attractive explanation.

For the product overview, use our Consensus features guide . Here the focus is the sequence of work: question, retrieval, screening, appraisal and a defensible answer.

Step 1: Define the question before opening the search

A broad topic gives you too many ways to be accidentally satisfied. “Remote work productivity” might retrieve studies about employee preferences, office attendance, output, stress or management practices. These are related subjects, but they do not answer the same question.

Try an illustrative question such as: “What research examines written asynchronous updates and coordination in geographically distributed software teams?” This is an example search brief, not a claim that a particular effect has been established.

Write down four boundaries: the group you care about, the practice being examined, the outcome you want to understand, and the kinds of studies you will include. If you cannot decide the outcome, your first search may be exploratory. Label it that way so the exploration does not quietly become a final evidence review.

Create a short note containing the question and today’s date. Add the decision it will inform, such as designing an internal pilot. A reader returning next month should understand what you intended to learn and what the search was never meant to establish.

Step 2: Search in more than one way

Start with the question as a sentence. Then try a keyword version using the main concepts. The point is to investigate whether your own wording narrows the literature too much. Save both query strings exactly, even if the second search appears less helpful.

For the example above, one version might emphasize “asynchronous communication,” while another emphasizes “distributed teams” and “coordination.” A paper can use a neighboring term without studying the exact practice you have in mind. Read before treating it as a match.

Consensus also documents known-item searches. If a lecturer or colleague has provided a specific paper, try its title or DOI. Finding a known relevant item is a useful check; failing to find it is a reason to investigate coverage or query interpretation, not evidence that the paper is unimportant.

Keep searches separate while you learn. If you change several concepts at once, you cannot tell which change improved relevance. A simple search log is often more valuable than a long prompt with many competing instructions.

Step 3: Screen relevance before reading summaries closely

The first screening question is whether a paper studies your question. A sophisticated summary of the wrong population remains the wrong evidence. Check the title and abstract, then record why the study belongs in or outside your set.

Look for the unit of analysis. An individual worker, a project team and an entire organization are different units. A study about individual satisfaction may be relevant context, but it should not automatically become evidence about project delivery.

Next check what was measured. Self-reported productivity, completed tasks and independently assessed quality are different outcomes. Combining them under one label can create a stronger-sounding conclusion than the studies justify.

Use a brief exclusion reason, such as “different setting,” “outcome does not match,” or “commentary rather than empirical study.” Consistent exclusions make the final evidence set easier to explain and help another reviewer spot arbitrary choices.

Step 4: Read the original paper behind an important claim

Open the paper or its publisher record when a finding could change your conclusion. Check the methods and results before relying on an AI interpretation. If full text is unavailable, distinguish an abstract-based observation from a full-paper appraisal.

Ask how participants were selected, how the outcome was measured, what comparison was made, and which explanations remain possible. An association does not establish that one workplace practice caused an outcome. A useful summary may still omit that distinction.

Also identify the version. A preprint, accepted manuscript and published article may differ. Record the version you actually read and any access limitation. Avoid presenting a secondary summary as though you inspected the study yourself.

When the generated explanation makes a numerical claim, locate that number in the source and check its denominator, timeframe and uncertainty. If you cannot trace it, leave it out of your report until you can. You do not need every interesting sentence to produce a useful answer.

Step 5: Build a small evidence log

Use a spreadsheet or research note with one row per included paper. The table below is a template for your own work, not a report of studies reviewed for this article.

Write the finding in your own words after checking the paper. If an AI tool suggested the wording, review every substantive element before placing it in the source-checked column. Keep an unresolved question in a separate field rather than smoothing it into a confident sentence.

Step 6: Investigate disagreement instead of averaging it away

Suppose two papers seem to disagree. First compare their questions. One may examine employees choosing remote work while another examines a mandatory policy. Their findings can differ without either paper being an error.

Then compare duration, outcomes and implementation. A short transition period can reveal disruption; a longer study may examine a settled routine. Do not merge these settings simply because both abstracts contain the same phrase.

Your synthesis might therefore say that evidence varies by setting and that a particular condition needs further investigation. That is a useful answer. A decisive recommendation unsupported by comparable studies is not an improvement.

If you need citation context or broader discovery, consider the workflows in our Consensus alternatives guide . Switching tools should solve a specific evidence problem, such as inspecting later discussion of a paper, rather than simply producing another summary.

Step 7: Write an answer with an explicit boundary

A defensible research note has three parts: what the included evidence suggests, where the evidence is uncertain, and what you would do next. Cite the papers supporting each important point. Keep your operational recommendation distinct from the research finding.

For example, a team might choose to test a written update format internally because its current handoffs are unclear. That is a management decision based on a local problem. It should not be described as a universally proven result unless the literature actually supports that scope.

End the note with the search date, major search terms and coverage limitations. These details help future readers decide whether the review needs refreshing. They also make it easier to add a new paper without rebuilding the reasoning from memory.

Troubleshooting an unhelpful search

When results are sparse, remove a boundary temporarily and observe what changes. Try a neighboring term found in an already relevant paper. Keep the original question in view so broader searching does not quietly change the topic.

When results are numerous but weak, add the population or setting and tighten your screening rule. A date restriction can serve a clearly defined purpose, but automatically favoring recent work can hide foundational evidence. Explain the choice in your log.

When a summary cannot be checked, return to the paper record. If access remains limited, state that limitation. Do not fill the gap with an invented methodological detail or infer that a source supports a statement because its title sounds relevant.

When the interface differs from an online tutorial, consult the current help page and check your account’s available features. This guide avoids promises about exact button locations or paid feature allowances because those can change independently of the research method.

Practical buying criteria

Before paying, choose a small recurring research job and decide what a successful evaluation means. Useful measures include time spent finding relevant papers, time spent verifying summaries, and how easily a colleague can reproduce the evidence trail. Do not evaluate only how quickly the tool returns text.

Check your current plan’s limits directly. This article does not reproduce subscription prices or guarantee a particular export, upload or analysis entitlement. A feature visible in documentation may not be available in every account configuration.

The main pros of a literature-focused assistant are a clearer starting point and a route from a question to papers. The cons and limitations include incomplete retrieval, interpretation errors and verification work that remains with the reader. Our Consensus pros and cons discusses those tradeoffs more broadly.

A final review before you share

Keep a brief refresh rule with your research note. State what would justify revisiting the question: a newly published study, a changed decision scope or a discovery that an important source was excluded. This makes the note maintainable without implying that every search must be repeated continuously.

When another person reviews the note, invite them to challenge one inclusion and one conclusion. Ask whether the paper actually fits the stated scope and whether the conclusion preserves its limitations. Record the response alongside your evidence log. A small independent reading check can expose a misunderstanding that another round of generated prose would leave untouched.

Read your note without looking at the generated summary. Can every substantive conclusion be traced to papers you inspected? Can you explain exclusions, disagreement and missing evidence? Would another reader understand where your recommendation goes beyond the literature?

If those answers are clear, Consensus has supported a research workflow rather than merely shortened the writing stage. The result worth keeping is the checked evidence trail, because that is what another person can examine, challenge and update.