Perplexity, Elicit & Consensus for Finding Papers
This is where students go from "playing with ChatGPT" to doing actual research. Perplexity, Elicit, and Consensus are AI search tools that retrieve real, indexed sources and synthesize them with citations. Unlike general chatbots, they are designed to ground every claim in a real paper or article — and they show you the paper.
If you remember nothing else from this course, remember this: use these tools to find sources, not ChatGPT.
What You'll Learn
- What each of the three tools does and how they differ
- How to write search queries that work for AI-powered academic search
- How to evaluate the sources these tools return
- A full worked example of finding 10 high-quality sources for a literature review
What Each Tool Does
Perplexity (perplexity.ai). A general AI search engine. You ask a question, it searches the web (including academic sources via its "Academic" focus mode), and returns an answer with numbered citations to real web pages and papers. Free tier gives unlimited "quick" searches and a limited number of "Pro" searches per day. The "Academic" focus mode is what you want for research — it prioritizes peer-reviewed sources.
Elicit (elicit.com). Built specifically for research. You type a research question and Elicit searches its corpus of academic papers, returning a synthesized answer plus a table of the most relevant papers, each summarized in columns (e.g., "main finding," "method," "sample size"). Free tier gives you a generous number of searches per month. Particularly powerful for systematic-review style work.
Consensus (consensus.app). Focuses on answering yes/no/maybe research questions ("Is X effective for Y?") by aggregating findings across multiple papers and showing a "consensus meter" of supporting vs opposing evidence. Free tier gives limited searches per month. Very useful for evidence-based fields like medicine, psychology, education.
Always check the official sites for current limits — they change.
How These Tools Are Different from ChatGPT
The fundamental difference is retrieval vs generation.
- ChatGPT generates plausible-sounding text from patterns. It may invent papers.
- Perplexity, Elicit, and Consensus retrieve real papers from indexed databases first, then summarize what those papers actually say.
This matters for two reasons. First, the citations are real — you can click through and read the originals. Second, the synthesis is grounded — if no paper supports a claim, the tool will tell you, rather than making something up.
You can verify this yourself. Ask ChatGPT "Give me 5 peer-reviewed papers on [narrow topic]." Ask Elicit the same. The Elicit results will all be real; the ChatGPT results often won't be.
Writing Search Queries That Work
These tools handle natural-language research questions better than keyword strings. But there is still a craft to writing them.
Bad query: "papers on motivation"
Better query: "What does recent peer-reviewed research say about the effects of intrinsic vs extrinsic motivation on academic performance in undergraduate students?"
Even better query (Elicit-style): "Does intrinsic motivation predict academic performance better than extrinsic motivation in undergraduates, and how has the finding changed since 2015?"
The pattern:
- Start with a clear research question phrased as a complete sentence.
- Specify the population ("undergraduates", "patients with type 2 diabetes", "small businesses in the EU").
- Specify the time range if you have one ("since 2015", "in the last decade").
- Specify the relationship or comparison ("vs," "predicts," "associated with").
- Avoid vague terms like "best" or "effective" without saying for what outcome.
Save your best queries. You will reuse them across courses.
How to Evaluate What These Tools Return
Even with these tools, you must still apply judgment. The returned papers are real, but not all are good. Use a simple checklist for each one:
- Is it peer-reviewed? Conference papers, working papers, and preprints (e.g., arXiv, SSRN) can be valuable, but for many assignments you specifically need peer-reviewed journal articles.
- Is it recent enough? Check the publication year. For fast-moving fields, anything older than 5 years is dated.
- Is the journal reputable? A 30-second search of the journal name tells you if it is a predatory or low-quality outlet. Look for indexing in Web of Science, Scopus, or PubMed.
- What is the sample / scope? A paper studying 30 students at one university is not the same as a meta-analysis of 50,000 students across 20 countries. Both have value but for different claims.
- What do the authors themselves admit is the limitation? This is your single most powerful question.
Elicit makes this easier by extracting fields like sample size, methodology, and key finding into a table. Use these tables, but spot-check by opening the actual paper.
Worked Example: Building a Literature Review List
Imagine you are a second-year psychology undergraduate writing a 2,500-word paper on whether mindfulness apps improve sleep quality in college students.
Step 1: Open Consensus. Ask: "Do mindfulness apps improve sleep quality in college students?"
Consensus returns a synthesis and a consensus meter ("Mostly yes — limited evidence"). You see five or six paper cards, each summarized.
Step 2: Open Elicit. Ask the same question. Elicit returns a table of perhaps 8 papers with columns: "intervention", "sample", "outcome measure", "main finding". You sort by relevance and by recency.
Step 3: Cross-check on Perplexity. Switch to Academic focus mode. Ask: "What systematic reviews have been published since 2022 on mindfulness apps and sleep in young adults?" Perplexity returns links to two or three reviews.
Step 4: Build your list. From the three tools you now have, say, 15 candidate papers. Open each in Semantic Scholar or Google Scholar. Filter out:
- Papers in predatory journals.
- Papers behind paywalls you cannot access (or get them via your library's interlibrary loan).
- Papers that are about a different age group or intervention than you need.
You should end up with 8–12 strong candidates. Now you read them.
Step 5: Save everything. Use Zotero (free, zotero.org) to save each paper's metadata. We will cover Zotero in Lesson 9.
This whole process — finding, filtering, and saving 10 good papers — used to take a full afternoon in the library. With these tools it takes an hour. The savings are real, but the judgment is still yours.
A Common Pitfall
Students sometimes assume that because the tool returns a clean answer, the question is settled. It is not. Each paper is one study. Findings disagree. Methods vary. A "consensus" of three papers is not a consensus.
Use these tools to surface evidence, but write your paper as an argument that weighs the evidence, including the disagreements.
A Quick Exercise
Pick a question from a current course. Run it through all three tools: Perplexity (Academic mode), Elicit, and Consensus. Note which tool surfaced the most useful papers, which papers appeared in two tools (those are usually high-signal), and which papers appeared in only one (sometimes a hidden gem, sometimes a niche oddity).
Save your top five candidates and write a one-sentence note on why each is relevant. Congratulations: you have just done the first hour of a literature review.
Key Takeaways
- Perplexity, Elicit, and Consensus retrieve real papers — use them, not ChatGPT, to find sources.
- Write search queries as complete research questions with specified population, time range, and comparison.
- Always evaluate the papers you retrieve for peer review status, recency, journal reputation, sample, and self-acknowledged limitations.
- Cross-check across tools. Papers that appear in multiple tools are usually higher signal.
- These tools save time but do not replace your judgment as a researcher.