How to Use Consensus to Find Evidence for a Claim
Most people who try to back up a claim with research hit the same wall: they find one study that supports their point, feel satisfied, and stop looking. The problem is that single studies are often wrong, or right in very specific contexts that do not apply to the claim at hand. Consensus was built to address exactly this. Instead of giving you one paper, it tells you what the body of literature actually says.
The headline feature is a "consensus meter" that shows the rough percentage of relevant studies that support, partially support, or contradict a given claim. That meter is not magic and it has real limits, but as a first-pass check on whether evidence tends to support your claim, it is genuinely useful in a way that Google Scholar is not.
How to Write a Query That Gets Good Results
Consensus works on research questions, not keyword searches. The search bar is expecting a claim or a question about a specific relationship: "Does intermittent fasting improve insulin sensitivity?" or "Is mindfulness effective for anxiety in adults?"
The more specific you are, the more reliable the consensus meter becomes. "Does caffeine affect performance?" will produce a noisy result because the literature covers wildly different doses, populations, and performance metrics. "Does 200mg of caffeine improve reaction time in sleep-deprived adults?" is much more answerable.
A few query formats that work well:
- "Does X cause Y in [population]?"
- "Is [intervention] effective for [outcome]?"
- "Does [behavior] increase [risk]?"
If you paste in a vague claim and the meter shows 40% yes / 30% partially / 30% no, that is often a sign to narrow the question rather than a genuine split in the literature.
Reading the Consensus Meter
After you search, the meter appears at the top with a color-coded arc and a percentage breakdown. Here is how to read it honestly:
| Meter reading | What it actually means |
|---|---|
| 85%+ yes | Strong signal. Most indexed papers lean in this direction. Worth citing. |
| 60-85% yes | Moderate consensus. Evidence leans positive but there is meaningful heterogeneity. |
| 40-60% mixed | Genuine debate or heterogeneous findings. Investigate sub-questions. |
| Below 40% yes | Evidence leans negative or conflicting. Do not assume your claim holds. |
The meter is calculated from the papers Consensus can index and understand, which skews toward English-language, open-access, peer-reviewed journals. This is a non-trivial selection effect. If most research on your topic is published in specialist journals that are not open access, the meter could be misleading.
Also worth knowing: Consensus does not weight studies by quality. A poorly designed study with 30 participants counts the same as a meta-analysis of 10,000. The quality filter (see below) is how you correct for this.
Using Study Snapshots
Below the meter, you will see a list of study cards. Each card is a "snapshot" that shows:
- Paper title and journal
- Year of publication
- A one-sentence AI-generated finding summary
- A color dot indicating whether the study supports, partially supports, or contradicts the claim
The one-sentence summary is the most useful thing on the card for quick scanning. It saves you from reading abstracts. But treat it as a pointer, not the final word. The summaries can miss important caveats, especially when a paper's conclusion is nuanced or conditional.
Click any card to expand the full abstract and see the original citation. If you want the full paper, the DOI link is there. This is where you do the actual reading on the papers that will end up in whatever you are writing.
A practical habit: after scanning the list of snapshots, sort by "Most cited" using the filter dropdown. Highly cited papers are not always right, but they are the ones other researchers have found worth referencing. If a highly cited paper contradicts the consensus meter, that is a flag to dig deeper before claiming strong support for your claim.
Applying Quality Filters
The quality filter is the feature most casual users skip, and it should not be skipped. Click "Filters" above the results and you will see several options:
- Study design: Filter for only RCTs, meta-analyses, or systematic reviews. This is the single most useful filter for evaluating claims about causation.
- Sample size: Set a minimum. I typically use 100+ for any health-related claim.
- Journal tier: Consensus integrates with some journal-ranking data. Filtering for higher-ranked journals reduces noise, but also reduces results.
- Year range: Essential for fast-moving fields. Nutrition research from 2005 is often outdated.
After applying filters for meta-analyses and systematic reviews only, the result count typically drops from 50 or 100 papers to 5 to 15. But those 5 to 15 papers are worth far more than 100 mixed-quality studies, and the consensus meter re-calculates based on the filtered set. That recalculated meter is usually the number I trust.
Step-by-Step: Checking a Specific Claim
Here is a real workflow for using Consensus to fact-check a claim before writing it up:
- Phrase the claim as a research question. "Coffee reduces the risk of type 2 diabetes" becomes "Does coffee consumption reduce type 2 diabetes risk?"
- Run the initial search. Note the unfiltered meter as a baseline.
- Apply filters: systematic reviews and meta-analyses only, last 10 years, sample size 1000+.
- Read the finding summaries on the top 5 to 8 papers.
- Click through on any paper where the summary seems surprising or where the effect size matters (large vs. small benefit).
- Check whether the effect sizes are clinically meaningful. A statistically significant effect can be negligibly small.
- Note any important moderators the papers mention: dose, population characteristics, confounders.
If after step 7 the claim still holds in the form you want to make it, you have solid evidence. If not, you now know exactly how to qualify the claim accurately.
Citing Consensus Results Properly
Consensus is a discovery tool, not a citable source. Do not cite "according to Consensus.app." Cite the underlying papers.
When writing up findings, the correct approach is:
- List the specific papers you are drawing on, with DOI or URL.
- Note the study design and sample size for any study you quote a finding from.
- If you reference the consensus meter, describe it as a summary of N indexed studies rather than as a standalone authority.
This matters especially for anything that will be reviewed: a grant application, a health article, an academic paper, a policy brief. Reviewers know what Consensus is and will check your citations directly.
When Consensus Is Not the Right Tool
Consensus indexes primarily biomedical, psychology, nutrition, and social science literature. For engineering, law, history, or niche technical fields, coverage is thin and the meter will not be reliable. For questions about recent events (anything from the last few months), the indexing lag means results will be incomplete.
It also cannot handle normative questions. "Should companies disclose executive pay?" has no consensus meter because that is not an empirical claim. Consensus only works for questions that empirical research can answer.
For what it covers, Consensus is one of the most honest research tools available. The combination of a meter and individual study snapshots, filtered to high-quality designs, gives you a quick but defensible read on whether evidence genuinely supports a claim. The key is using the quality filters and then reading the actual papers for anything you plan to stand behind.