What is Elicit?
Elicit is an AI research assistant built by Ought, a San Francisco-based AI research lab founded in 2017. Unlike general-purpose search engines or AI chatbots, Elicit is purpose-built for academic research. It searches across a database of over 200 million academic papers from Semantic Scholar and other open-access repositories, and uses large language models to extract specific claims, findings, methodologies, and data points from those papers — not just return a list of titles and abstracts. When you ask Elicit a research question, it does not just give you an AI-generated answer. It surfaces actual papers that address your question, extracts the relevant findings from each paper with page-number citations, and organizes the evidence so you can evaluate it yourself.
Elicit is used by researchers at Harvard, MIT, Stanford, Oxford, Cambridge, and hundreds of other universities. It has become particularly popular among PhD students conducting literature reviews and professors who need to quickly survey the evidence landscape on a new research question. As of mid-2026, Elicit reports over 500,000 active researchers on the platform. The tool was developed with funding from the Open Philanthropy Project and has been recognized for its focus on improving the quality and efficiency of academic research workflows rather than replacing researcher judgment.
How it differs from Google Scholar, PubMed, and ChatGPT: Google Scholar and PubMed are search engines — they find papers matching your keywords and return metadata (titles, abstracts, citation counts). ChatGPT can discuss research topics but frequently hallucinates citations and cannot reliably extract specific data points from specific papers. Elicit sits between these approaches: it finds real papers (like a search engine) and extracts structured information from them (like an AI assistant), but every claim it makes is tied to a specific paper and page that you can verify. This makes it particularly valuable for systematic reviews, meta-analyses, and any research context where citation accuracy is non-negotiable.
Key Features
Research Question Search
Ask Elicit a natural-language research question — for example, "What is the effect of mindfulness meditation on working memory in adults?" — and it searches across 200M+ papers to find the most relevant studies. Elicit understands the semantics of your question, not just keyword matching. It returns a table of papers with columns that you customize: study design, sample size, intervention type, effect size, key findings, limitations, and more. Each extracted data point links to the specific page in the paper where the information appears. This structured approach transforms literature review from a reading task into a data evaluation task — you still need to read critically, but you start from an organized evidence table rather than a pile of PDFs.
Data Extraction from Papers
Upload a PDF of a specific paper or select papers from your search results, and Elicit extracts structured data points: population studied, intervention details, outcome measures, statistical results, effect sizes, p-values, sample characteristics, and funding sources. Define custom extraction columns for your specific research needs — Elicit will attempt to extract that information from every paper in your set. For a systematic review of 80 papers, this can reduce the initial data extraction phase from weeks of manual spreadsheet work to a few hours of verifying and correcting AI-extracted data. Elicit's extraction accuracy varies by domain and data type: numerical results (effect sizes, sample sizes) are extracted with higher accuracy than nuanced methodological descriptions. Always verify extracted data against the original paper.
Abstract Screening for Systematic Reviews
Elicit can help screen large sets of abstracts against your inclusion and exclusion criteria — a typically tedious and time-consuming step of systematic reviews. Upload your screening criteria as natural-language rules. Elicit evaluates each abstract and provides a relevance score and a brief justification. You then review Elicit's decisions — accepting, rejecting, or correcting them. This semi-automated screening workflow can reduce abstract screening time by 50-70% in initial pilots, though it is not a replacement for human screening. The key advantage: Elicit's screening is consistent — it applies the same criteria to every abstract, whereas human screeners can drift over hundreds of abstracts due to fatigue.
Citation Network Exploration
For any paper in your results, Elicit shows its citation network — papers it cites and papers that cite it. This allows researchers to trace the evolution of an idea forward (who built on this paper?) and backward (what evidence does this paper rest on?). Combined with Elicit's extraction capabilities, you can quickly assess whether subsequent studies replicated or contradicted a paper's findings. This feature is particularly useful for identifying seminal papers in a field and understanding how a research conversation has developed over time.
Customizable Extraction Tables
Beyond predefined extraction columns, Elicit allows researchers to define completely custom data extraction fields. For a meta-analysis on the effects of a specific educational intervention, you could define columns like: "Grade level of participants," "Duration of intervention in weeks," "Effect size (Cohen's d)," "Control group type," "Attrition rate," and "Reported limitations." Elicit attempts to extract these data points from every paper in your set. The structured output can be exported as CSV for statistical analysis in R, Stata, SPSS, or Python. This feature is what separates Elicit from simpler AI summarization tools — it is designed for the data-intensive workflows of systematic reviews and meta-analyses, not just casual literature browsing.
Paper Concept Mapping
Elicit groups search results by concept, methodology, or finding — helping researchers quickly identify clusters of papers that address similar questions or use similar approaches. For a broad research question like "What interventions reduce student dropout rates?", Elicit might group results into clusters like "financial incentives," "mentoring programs," "curriculum reform," and "early warning systems" — with papers organized under each cluster. This conceptual organization can reveal gaps in the literature: if one cluster has only 3 papers while another has 50, that disparity might indicate an under-researched area worth investigating. The concept mapping is AI-generated and should be treated as suggestive, not definitive — but it provides a useful starting point for structuring a literature review.
How Professors Use Elicit: A Literature Review Workflow
Formulate Your Research Question
Write your research question in plain English — the more specific, the better Elicit's results. Instead of "effects of exercise on cognition," write "What is the effect of aerobic exercise interventions of 12 weeks or longer on executive function in adults aged 60-75 without cognitive impairment?" Elicit performs best with well-scoped questions that include the population, intervention, comparison, and outcome (the PICO framework familiar to researchers). You can also phrase questions as comparisons: "How does the effectiveness of CBT compare to SSRI medication for moderate depression in adolescents?" Elicit's AI parses the question structure and weights papers accordingly, prioritizing those that directly address your comparison rather than papers that only discuss one side of it.
Configure Your Extraction Columns
Before reviewing results, set up the columns you want Elicit to extract from each paper. The default columns include: paper title, authors, year, journal, abstract summary, and key findings. Add custom columns for your specific analysis: for a meta-analysis on interventions, you might add "Sample size," "Effect size (Cohen's d)," "Intervention duration," "Control condition type," "Measurement instrument used," and "Reported limitations." For a qualitative synthesis, you might add "Theoretical framework," "Research methodology," "Key themes identified," and "Population characteristics." The more precisely you define your extraction columns upfront, the more useful Elicit's output table will be. Spend time on this step — it determines whether your final evidence table is analysis-ready or requires significant rework.
Review and Verify Extracted Data
Elicit returns a table of papers with your requested columns populated. Now comes the most important step that separates rigorous research from over-reliance on AI: verify the extracted data. Click through to each paper (Elicit provides direct links to the PDF or publisher page) and check that Elicit's extracted information is accurate. Pay special attention to numerical data — sample sizes, effect sizes, statistical tests, p-values. Elicit's accuracy for numerical extraction is estimated at 85-95% (based on Ought's published benchmarks), which means 5-15% of data points may contain errors. For a systematic review or meta-analysis, that error rate is too high to accept without verification. Rate each paper's relevance on a 1-5 scale within Elicit and add notes about extraction accuracy. Papers with verified data can be exported to your analysis software; papers with questionable extraction should be manually coded. The verification step typically takes 2-4 hours for a set of 50 papers — significantly less than manual extraction from scratch, but not zero.
Export and Analyze
Export your verified extraction table as CSV or Excel. The exported file includes all your extracted columns, your relevance ratings, and your verification notes. Import this into your statistical software — R, Stata, SPSS, Python (pandas) — for meta-analysis, thematic synthesis, or whatever analytical approach your research requires. For meta-analyses, the CSV format is directly compatible with the metafor package in R and the meta suite in Stata. For qualitative syntheses, the extracted themes and methodologies can be coded in qualitative analysis software like NVivo or ATLAS.ti. Elicit does not perform the analysis — it prepares the data for analysis. The intellectual work of interpreting findings, assessing study quality, identifying publication bias, and synthesizing conclusions remains the researcher's responsibility. Elicit's role is to dramatically reduce the mechanical work of finding, reading, and extracting data from papers, so you can spend more time on the analytical work that requires your expertise.
Elicit Pricing & Plans (2026)
| Plan | Price | What You Get |
|---|---|---|
| Basic | Free | 10,000 credits per month (approximately 10-15 searches with full extraction), access to 200M+ papers, basic extraction columns, export to CSV. Good for occasional literature searches and testing the platform. |
| Plus | $10/month ($8/month annual) | 50,000 credits per month, priority access during high-traffic periods, all extraction column types, ability to upload and extract from your own PDFs, concept mapping, and advanced filters. Suitable for PhD students and active researchers conducting regular literature reviews. |
| Pro | $50/month ($40/month annual) | Unlimited credits, systematic review workflow tools (abstract screening, PRISMA diagram generation), team collaboration (shared extraction tables, shared paper libraries), priority support. Designed for research groups, professors, and researchers conducting systematic reviews or meta-analyses. |
Pricing verified June 2026. Credits are consumed per action (search, extract, summarize). A typical literature review search with extraction of 50 papers consumes approximately 5,000-8,000 credits. The Pro plan is recommended for researchers conducting a full systematic review — the abstract screening and PRISMA tools alone justify the upgrade over Plus.
Elicit vs Competitors: How It Compares
Elicit competes in the growing AI-for-research space. Each tool has distinct strengths — understanding these differences helps researchers choose the right tool for each stage of their workflow.
| Feature | Elicit | Consensus | SciSpace | ResearchRabbit |
|---|---|---|---|---|
| Core Function | Research question → structured evidence table with extracted data | Research question → synthesized answer with consensus meter | Paper reading + AI explanations + citation extraction | Citation network exploration + paper recommendations |
| Paper Database | 200M+ papers (Semantic Scholar + open access) | 200M+ papers (Semantic Scholar) | 270M+ papers | Semantic Scholar database |
| Data Extraction | ✅ Custom columns, numerical extraction, systematic review workflows | ⚠️ Basic extraction of key findings only | ✅ Extracts methods, results, limitations from PDFs | ❌ No data extraction |
| Systematic Review Support | ✅ Abstract screening, PRISMA diagram, extraction tables | ❌ Not designed for systematic reviews | ⚠️ Limited systematic review features | ❌ Not designed for systematic reviews |
| Answer Synthesis | ⚠️ Provides evidence table; researcher synthesizes | ✅ AI-generated consensus answer with citations | ✅ AI-generated paper summaries and explanations | ❌ Does not synthesize |
| Citation Network | ✅ Shows cited-by and references within extraction workflow | ❌ Limited citation exploration | ✅ Citation network with paper relationships | ✅ Deep citation mapping, personalized recommendations |
| Custom PDF Upload | ✅ Upload your own PDFs for extraction (Plus/Pro) | ❌ Cannot upload PDFs | ✅ Upload and chat with PDFs | ❌ Cannot upload PDFs |
| Free Tier | ✅ 10,000 credits/month | ✅ 20 searches/month | ✅ Limited free tier | ✅ Fully free (academic project) |
| Paid Plans | $10–$50/month | $8.99–$12.99/month | $12–$20/month | Free |
| Best For | Systematic reviews, meta-analyses, and researchers who need structured data extraction from large paper sets | Quick evidence checks — "what does the literature say about X?" — and getting a directional answer fast | Reading and understanding individual papers deeply, exploring citation networks around a paper, and getting AI explanations of complex content | Discovering new papers in a field through citation networks, tracking what influential authors publish, and building a personalized research feed |
Comparison verified June 2026. Many researchers use multiple tools: Elicit for systematic review extraction, Consensus for quick evidence checks, SciSpace for deep reading, and ResearchRabbit for ongoing literature discovery. These tools are complementary rather than direct substitutes.
Pros & Cons
Pros
- Transforms literature review from reading to data evaluation: Instead of reading 100 papers linearly and taking notes, Elicit presents a structured evidence table with extracted data points. This shifts the researcher's cognitive load from "finding information in papers" to "evaluating the quality of information" — which is where expertise adds the most value.
- Every claim links to a specific paper and page: Unlike ChatGPT or other LLMs, Elicit does not generate knowledge — it extracts from real papers. Every data point, claim, and finding in Elicit's output is linked to the source paper with page numbers. This is essential for academic integrity and is what makes Elicit usable in formal research contexts.
- Systematic review workflow tools save weeks of work: The abstract screening tool, customizable extraction tables, and PRISMA diagram generation address the most time-consuming steps of systematic reviews. Researchers report 50-70% time reduction in the extraction and screening phases.
- Free tier is genuinely useful: 10,000 credits per month is sufficient for several literature searches with extraction of 30-50 papers. PhD students can conduct a preliminary literature review without paying. This generous free tier reflects Ought's mission-driven origins.
- Concept mapping reveals research gaps: The automatic grouping of papers by concept and methodology helps researchers quickly identify which areas are well-studied and which are under-researched — valuable for framing the significance of a new study or identifying dissertation topics.
Cons
- Extraction accuracy is not 100% — verification is mandatory: Elicit's data extraction accuracy for numerical values is estimated at 85-95%. For a systematic review, every extracted data point must be verified against the source paper. Elicit saves time but does not eliminate the need for careful human review. Researchers who skip verification risk incorporating erroneous data into their analyses.
- Limited to papers in Semantic Scholar's database: Elicit's coverage is determined by what is indexed in Semantic Scholar and open-access repositories. Papers behind paywalls, older papers not digitized, and grey literature (conference proceedings, preprints, dissertations) may be underrepresented. For comprehensive systematic reviews, Elicit should be supplemented with searches in PubMed, Web of Science, Scopus, and subject-specific databases.
- Struggles with highly specialized or niche domains: Elicit's AI models perform best on research questions in well-represented fields: psychology, medicine, economics, education, computer science. Highly specialized domains with unique terminology, small research communities, or non-English-language literature may yield lower-quality extractions and more missed papers.
- Not a writing tool: Elicit helps you find and organize evidence — it does not help you write the literature review. Researchers still need to synthesize extracted data into a coherent narrative, assess study quality, and construct arguments. For AI-assisted writing, tools like SciSpace or Jenni AI complement Elicit's extraction capabilities.
- Credit system can be confusing for new users: Different actions consume different amounts of credits, and it is not always transparent how many credits a complex extraction will use. Researchers on the free tier may unexpectedly run out of credits mid-workflow. The Plus plan's 50,000 credits are sufficient for most individual researchers, but groups conducting large reviews should budget for Pro.
Frequently Asked Questions
Is Elicit suitable for systematic reviews intended for publication?
Elicit can be a valuable tool within a systematic review workflow, but it should not be the sole search or extraction method for a publishable systematic review. Best practice based on current (2026) methodological guidance: (1) Use Elicit as a supplementary search tool alongside traditional database searches in PubMed, Web of Science, Scopus, and discipline-specific databases. Elicit's coverage of grey literature, non-English papers, and older publications is incomplete. (2) Use Elicit's extraction features to accelerate data collection, but verify every extracted data point against the original paper. Document your verification process — how many data points were corrected, what types of errors were most common. This documentation is important for methodological transparency. (3) Elicit's PRISMA diagram generation is useful for drafting the flow diagram, but verify the numbers against your manual screening records. (4) When writing your methods section, describe Elicit as a tool used within your workflow, not as the methodology itself. For example: "Elicit (Ought, 2026) was used to assist with initial abstract screening and data extraction. All extracted data were independently verified by two reviewers against the original papers." The PRISMA 2020 statement and Cochrane Handbook do not yet include specific AI tool guidance, but methodology experts generally agree that AI tools can be used as assistants within systematic reviews as long as human verification and methodological transparency are maintained.
How does Elicit compare to just using ChatGPT with the web browsing plugin?
This is the most common question researchers ask when evaluating Elicit. The key differences: (1) Elicit searches a curated database of 200M+ academic papers from Semantic Scholar. ChatGPT with browsing searches the general web, which includes academic papers but also blog posts, news articles, and non-scholarly sources — with no distinction between them. (2) Elicit extracts structured data into columns that you define (sample size, effect size, methodology, etc.) and links every data point to a specific paper and page. ChatGPT can summarize papers but does not produce structured, verifiable extraction tables. (3) Elicit does not generate synthetic answers — it presents extracted data for you to interpret. ChatGPT synthesizes answers, which can introduce hallucinated claims when discussing research. (4) Elicit's systematic review workflow (abstract screening, PRISMA, extraction tables) has no equivalent in ChatGPT. The bottom line: if you need a quick overview of a research area and are comfortable with the risk of hallucination, ChatGPT is faster. If you need verifiable, structured, citation-backed evidence for formal research, Elicit is the appropriate tool. Many researchers use both: ChatGPT for brainstorming research questions and exploring ideas, Elicit for the formal literature search and data extraction.
Can I use Elicit for a dissertation literature review?
Yes, and many PhD students do exactly this. Elicit is particularly well-suited for the dissertation literature review phase because it helps with: (1) Initial scoping — asking broad research questions to understand the landscape of literature on your topic and identify the key papers, authors, and debates. (2) Structured extraction — defining the data points you want to capture from each paper and building an evidence table that becomes the foundation of your literature review chapter. (3) Gap identification — Elicit's concept mapping helps you see which areas of your topic have been extensively studied and which have not, supporting your argument that your dissertation addresses a genuine research gap. A practical workflow: spend 2-3 weeks using Elicit to search, extract, and organize the literature on your topic. Export the evidence table. Then spend the following weeks writing your literature review using the evidence table as your reference — reading the most important papers in full, evaluating the quality of studies, and constructing your narrative. Elicit handles the "finding and organizing" part so you can focus on the "analyzing and synthesizing" part. One caution: your dissertation committee will expect you to have actually read the key papers in your field, not just extracted data from them. Elicit is a complement to deep reading, not a replacement for it. Plan to read the 20-30 most important papers in your literature review in full, even if Elicit has extracted their data points.
What fields does Elicit work best for?
Elicit performs best in fields with large bodies of empirical research, standardized reporting practices, and well-indexed literature. Strongest performance: psychology, medicine and public health, economics, education, computer science, sociology, political science, and environmental science. These fields benefit from standardized reporting of effect sizes, sample sizes, and methodologies that Elicit's extraction models are trained to recognize. Moderate performance: business and management, linguistics, anthropology, and law — these fields have empirical research but less standardized reporting formats, so extraction accuracy may be lower. Weaker performance: humanities fields with primarily interpretive rather than empirical research traditions (literature, philosophy, history, art history), fields with heavy reliance on non-English literature, and highly specialized technical fields with unique terminology not well represented in Elicit's training data. If you work in a field where research primarily takes the form of books rather than journal articles, Elicit will be less useful — its database is paper-focused. Researchers can test Elicit's performance in their field for free by running a few searches on the Basic plan before committing to a paid subscription.
Does Elicit support non-English language papers?
Elicit's database includes papers in multiple languages, but its search and extraction capabilities are primarily optimized for English. As of mid-2026, Elicit can: (1) Search for papers with non-English titles and abstracts if they are indexed in Semantic Scholar. (2) Extract structured data from papers in major European languages (Spanish, French, German, Portuguese) with moderate accuracy — approximately 70-80% of English-language extraction accuracy based on user reports. (3) The interface and extraction column labels are in English, and extracted data is returned in English regardless of the source paper's language. Elicit's coverage of research published in Chinese, Japanese, Korean, Arabic, and other non-Latin-script languages is limited — both because these papers are less comprehensively indexed in Semantic Scholar and because Elicit's extraction models are not trained on these languages. For researchers conducting systematic reviews that require comprehensive coverage of non-English literature, Elicit should be supplemented with database-specific searches in the relevant languages and manual extraction from non-English papers. Ought has stated that improving non-English language support is a development priority but has not announced a specific timeline.