Elicit
Ask a research question; get structured answers from real peer-reviewed papers.
You are a graduate student starting a literature review. Your advisor has said “just find the relevant papers” as if that were a two-hour task. You have opened Google Scholar, clicked through fourteen abstracts, and you have not written a single sentence.
Why this tool matters
Elicit is an AI research assistant built specifically on the academic literature. It searches tens of millions of papers (primarily from Semantic Scholar and OpenAlex), reads their abstracts and in many cases full texts, and produces structured answers to your research question — not as a chatbot summary, but as a table where each row is a paper and each column is an extracted data point you specified.
The step-change Elicit introduces is the data extraction pattern. You ask a question, get back a table of 5–50 relevant papers, and then add columns like sample size, main finding, methodology, or limitations. Elicit reads each paper and fills the cells. You get something that would have taken a graduate research assistant a week in about fifteen minutes — and every cell links to the exact passage in the paper.
This changes what a literature review is. You are no longer manually typing a synthesis table from PDF after PDF; you are editing one that was pre-populated for you. That time savings is the product.
Setup
Account: sign up at elicit.com with an institutional or personal email. Free tier gives you a monthly credit budget that is sufficient for a real literature review; heavy users move to Plus ($10/mo) or Pro ($42/mo).
Scope: Elicit is strongest on biomedical, behavioral, and social-science literature. It works less well for humanities, where evaluation is qualitative and the “findings” pattern fits less naturally.
Walkthrough
Step 1: Start with a real research question
On the Elicit home page, type a research question — not a keyword search. Bad: adolescent depression. Good: What are the effects of smartphone use on adolescent depression? Elicit is tuned to parse the structure of a research question; the more it looks like one, the better the retrieval.
Step 2: Review the top 8 results
Elicit returns a ranked list of papers, each with a one-sentence summary of the finding. Skim the top eight. If they are relevant, proceed; if not, rephrase your question (try adding methodology: …in randomized trials).
Step 3: Add extraction columns
Click Add column. Pick from common templates (Sample size, Intervention, Outcome, Effect size) or write a custom column prompt like What limitations did the authors acknowledge? Elicit reads each paper and populates the cells.
Step 4: Inspect cells with skepticism
Click into any cell. Elicit shows you the source passage it drew the answer from. This is your chance to check — AI paper reading is imperfect. Spot-check roughly one in five cells on anything you will cite.
Step 5: Expand or narrow the set
If your initial eight papers are the wrong shape, use the filters to narrow by year, publication type, or keyword. Use Find related papers on a particularly relevant paper to discover adjacent work.
Step 6: Export to your writing tool
Elicit exports to CSV, Zotero, and RIS. If you are writing your literature review in Notion, Word, or Google Docs, the CSV export becomes the skeleton table of your Methods or Results section.
Your turn
Basic: A focused mini-review
Pick a research question from your domain that you think you understand. Run it through Elicit. Add three extraction columns: sample size, main finding, and a methodological concern of your choice. Open each paper's summary and decide whether it would have changed your review if you had missed it.
Goal: realize that even a question you thought you understood has layers Elicit surfaces that you would have missed manually.
Advanced: A 20-paper structured review
Pick a question you genuinely need an answer to for work or study. Build an Elicit workspace with at least 20 papers and five extraction columns (one of them custom, specific to your domain). Spot-check twenty cells. Export the table to CSV.
Write a 400-word synthesis paragraph grounded in the table. In a reflection note at the bottom, answer: (1) which columns were most valuable, and (2) which cells did Elicit get wrong — and what would have happened if you had trusted them without checking?
This exercise is the working template of every future literature review you will do.
Pitfalls and pro tips
Access matters. Elicit reads abstracts for free, but paywalled full texts are extracted less reliably. Connect your institutional access if you have it; the quality difference is significant.
Summaries flatten nuance. Extraction cells aim for concise answers and can smooth over methodological caveats. For any paper you will cite, read the abstract in full (not just Elicit's summary cell).
Citation drift. Elicit occasionally attributes an idea to a paper where the paper was citing that idea from somewhere else. For famous claims, always trace back to the primary source.
How it compares
Elicit competes most directly with Consensus (Day 4), Scispace (Day 28), Scite, and ResearchRabbit. Roughly: Elicit is strongest at structured extraction across many papers; Consensus is strongest for yes/no empirical questions (does X cause Y?); Scispace is strongest at reading one paper deeply; ResearchRabbit is strongest for discovering adjacent work via citation graphs. A real research workflow uses at least two of these in sequence.
When to use — and when not to
Use Elicit when you need to synthesize findings across many papers — anything with the shape of a systematic review, a literature scan, or a meta-analysis pre-screen.
Do not use Elicit when you already know the three papers that matter (just read them), when your field is mostly qualitative or theoretical (Elicit's extraction pattern fits badly), or when you need to reason about the argument of one paper in depth (Scispace on Day 28 is better).