How to Run a Literature Review With Elicit in an Afternoon

November 8, 2026

Elicit searches academic literature and returns a table of papers, not a chat reply. Each row is a paper with extracted columns you choose — sample size, method, population, outcome. For a scoping pass over 200 papers, that table replaces two days of abstract skimming.

It’s built by Ought, launched in 2021, and runs on the Semantic Scholar corpus (about 125 million papers).

What is Elicit and what is it actually good at?

Elicit is an AI research assistant for literature review. You ask a research question in plain English, it returns candidate papers with abstracts, then lets you add extraction columns filled by a language model.

Strong at:

  • Broad scoping: “what does the literature say about X in population Y since 2019”
  • Extracting a consistent field across 50 papers (sample size, country, intervention type)
  • Finding papers you’d miss with the exact keyword you thought of first

Weak at:

  • Anything behind a paywall without abstracts in the index
  • Clinical or legal questions where missing one paper is unacceptable
  • Deep methodological critique — extraction is fast, judgment is yours

How do I structure a question so Elicit returns useful papers?

Specific beats clever.

Weak: “effects of social media”

Strong: “randomized controlled trials on social media use and adolescent sleep duration published after 2020”

The strong version names design, variables, population and a date filter. Elicit handles those constraints and the result set drops from thousands to dozens.

Steps:

  1. Type the question, run the search.
  2. Filter by year, study type, and whether full text is available.
  3. Scan the top 20 titles — if half are off-topic, tighten the question rather than scrolling.
  4. Add extraction columns only after the paper set looks right.

Skipping step 3 is the common mistake. People build elaborate column sets over a bad result set and then blame the tool.

Which extraction columns can I trust?

Columns are model-filled. Reliability varies by field:

Column Reliability
Publication year, journal High — metadata
Sample size Medium — usually right, occasional confusion with subgroups
Country / setting Medium
Main finding (one sentence) Low-medium — check the abstract
Effect size Low — frequently missing or misread

Rule: treat extracted text as a starting point for reading the abstract, never as a citation source. If a number will appear in your writeup, open the paper.

Elicit does show a confidence indicator per cell in the table view, and cells it couldn’t fill are marked empty rather than guessed. When a column is mostly empty, the model didn’t find the data — that’s a signal, not an error.

What does Elicit cost and what’s free?

As of late 2026: Free gives a limited number of monthly credits — enough for one small review. Plus is around $12/month with more credits and exports. Pro is around $49/month for heavier use, larger extractions, and reporting features.

Credits are consumed per search and per extraction column run, so a 100-paper table with six columns is expensive in credits. For a single thesis chapter, Plus usually covers it.

Where does Elicit not replace the real process?

A formal systematic review needs a documented search string, dual screening, and PRISMA reporting. Elicit can accelerate screening, but the protocol still has to be written by a human and the records still have to be reproducible.

Also: it won’t find grey literature, industry reports or very recent preprints reliably. Check Google Scholar and preprint servers as a separate pass.

Quick Summary

  • Elicit turns a research question into a table of papers with extracted columns, drawing on ~125M Semantic Scholar records.
  • Ask with design, population and date constraints; check the top 20 titles before adding columns.
  • Trust metadata columns, verify effect sizes and sample sizes against abstracts.
  • Free tier fits one small review; Plus is about $12/month, Pro around $49.
  • Not a substitute for PRISMA-compliant systematic review, and weak on grey literature.
Tags: