Elicit Review: The AI Research Assistant That Actually Understands Academic Papers

Why We Tested Elicit

Every researcher knows the pain of literature reviews. You start with a question, spend hours scrolling through Google Scholar results, open dozens of tabs, skim abstracts, lose track of which papers said what, and eventually compile your findings in a spreadsheet that is outdated by the time you finish. The entire process is manual, repetitive, and error-prone — exactly the kind of work that AI should be able to improve.

Elicit, built by the research nonprofit Ought, claims to do precisely that. Unlike general-purpose chatbots that happen to know things about research papers, Elicit is purpose-built for academic workflows. We spent eight weeks using it across three active research projects — a biomedical systematic review, a social science meta-analysis, and an exploratory literature survey in computational linguistics — to evaluate whether it delivers on that promise.

Semantic Search That Finds What Keywords Miss

Elicit’s search engine is its foundation, and it works differently from the keyword matching you are accustomed to in PubMed or Google Scholar. Type a research question in natural language — not a Boolean string of keywords and operators — and Elicit returns papers that are conceptually relevant to your question, even if they do not contain your exact search terms.

In our systematic review on interventions for academic burnout, searching “does mindfulness training reduce burnout in graduate students” returned not only papers using those exact terms but also studies on contemplative practices, stress-reduction programmes, and wellbeing interventions in doctoral populations that we would have needed separate keyword searches to discover. The relevance ranking was strong, with the most methodologically rigorous and directly applicable papers appearing in the top results.

The trade-off is coverage. Elicit indexes approximately 125 million papers from Semantic Scholar’s corpus, which is substantial but notably smaller than Google Scholar’s estimated 400 million documents. In our testing, coverage was excellent for biomedical, computer science, and social science literature but showed gaps in humanities, regional journals, and non-English language publications. Researchers working in well-indexed STEM disciplines will find coverage more than adequate; those in niche or interdisciplinary fields should use Elicit alongside — not instead of — traditional database searches.

Structured Data Extraction

This is where Elicit genuinely separates itself from every other tool we have reviewed. When you find relevant papers, Elicit does not simply list them — it reads them and extracts structured data into customisable columns. By default, it pulls the study summary, methodology, sample size, and key outcomes. You can add custom columns for anything you need: intervention type, follow-up duration, effect sizes, reported limitations, funding sources, or measurement instruments.

In our burnout systematic review, we loaded 47 papers into an Elicit workspace and asked it to extract intervention type, sample size, study design, primary outcome measure, and reported effect size. The tool completed extraction across all 47 papers in under three minutes — work that took our research assistant approximately 14 hours to do manually for a previous review. Accuracy was impressive: we verified extractions against the original papers for a random sample of 15 studies and found the intervention type and sample size were correct in 14 of 15 cases. The one error involved a paper reporting multiple studies with different sample sizes, where Elicit extracted the total rather than the per-study figure.

Effect size extraction was less reliable. Elicit correctly identified reported Cohen’s d or odds ratios about 70% of the time but occasionally confused adjusted and unadjusted figures, or extracted confidence intervals when we asked for point estimates. We strongly recommend verifying quantitative extractions against source material — the tool saves enormous time on initial data collection but does not eliminate the need for human validation.

The Workspace as a Living Evidence Table

Elicit’s workspace interface deserves specific attention because it represents a fundamentally different approach to organising research. Rather than a chat window or a document, your workspace is a structured table where each row is a paper and each column is an extracted data point. You can sort, filter, group, and annotate this table exactly as you would a spreadsheet — but the data was populated by AI reading the actual papers rather than by you typing it manually.

For systematic reviewers, this is transformative. Your PRISMA flow diagram, inclusion criteria, and extracted evidence all live in one workspace that updates dynamically as you add or remove papers. We found ourselves using the workspace as our primary working document during the screening phase, adding a “Include?” column with yes/no/maybe values and notes explaining each decision. When it came time to write the methods section, the audit trail was already there.

The workspace also supports collaboration, though we found this feature still maturing. You can share workspaces with colleagues and they can view your extractions, but real-time co-editing is not yet seamless. For lab groups conducting collaborative reviews, this is usable but not yet on par with Google Sheets as a collaboration platform.

How It Handles Different Disciplines

We deliberately tested Elicit across disciplinary boundaries because many AI research tools perform well in biomedical literature but struggle elsewhere:

Biomedical research is where Elicit performs best. The structured nature of clinical trials, cohort studies, and meta-analyses maps naturally to column-based extraction. We found extraction accuracy highest for RCTs and observational studies with clear methods sections.

Social science research performed well for quantitative studies but showed limitations with mixed-methods and qualitative work. Elicit can extract survey instruments, sample demographics, and statistical findings effectively, but struggles to summarise qualitative themes or capture the nuance of interpretive findings.

Computational and technical research was handled competently for papers with clear experimental setups — datasets used, model architectures, reported metrics — but less reliably for theoretical papers without empirical results sections.

Humanities and arts remain a gap. Elicit’s extraction model is fundamentally designed around empirical research structures (hypothesis, methods, results), and it does not yet handle argumentative, historical, or critical scholarship well. Researchers in these fields will find the search functionality useful but the extraction features of limited value.

How It Compares

Consensus is Elicit’s closest competitor, and the choice between them depends on your workflow. Consensus excels at answering specific research questions with yes/no consensus indicators across the literature — it tells you what the research says. Elicit excels at helping you systematically work through the literature yourself — it helps you read, extract, and organise what papers say. For systematic reviewers, Elicit is the clear choice. For quick evidence checks, Consensus is faster.

ChatGPT Plus and Claude Pro can discuss research and even summarise papers you paste into them, but neither can search an academic database, extract structured data, or maintain a persistent research workspace. They are complements to Elicit, not replacements.

Scite.ai and Research Rabbit offer citation analysis and paper discovery but lack Elicit’s extraction capabilities. We recommend using them alongside Elicit for citation network exploration.

Pricing

Elicit offers a straightforward pricing structure:

• Free tier — 5,000 credits per month, which covers approximately 500 paper interactions (searches, extractions, and summaries each consume credits at different rates)
• Elicit Plus — $10/month with 12,000 credits, priority processing, and expanded export options
• Elicit Pro — Custom pricing for teams and institutions with unlimited credits and API access

The free tier is genuinely generous. In our testing, 5,000 credits was sufficient for managing an active literature review of moderate scope (50–80 papers) across a month. Only when we ran multiple concurrent projects with heavy extraction workloads did we hit the limit. At $10/month, the Plus tier is the best value proposition in the academic AI tool market — less than a single journal article processing charge, and less than half the price of ChatGPT Plus or Claude Pro.

For graduate students on limited budgets, the free tier alone justifies adding Elicit to your research workflow. The cost barrier is essentially zero for individual researchers.

Who It’s For

We recommend Elicit for:

• Systematic reviewers and meta-analysts who need to screen, extract, and synthesise data from dozens or hundreds of papers efficiently
• Graduate students conducting literature reviews for thesis proposals, qualifying exams, or dissertation chapters who want to work systematically rather than haphazardly
• Research teams running collaborative evidence synthesis projects who need a shared, structured workspace
• Quantitative researchers in STEM and social sciences whose papers have clear methods, results, and statistical findings that map well to structured extraction
• Anyone starting a new research project who wants to quickly survey what the literature says before committing to a research direction

It is less ideal for humanities scholars working with argumentative or interpretive texts, researchers whose primary literature is in non-English languages, or those who need real-time citation management integration with tools like Zotero or Mendeley (though CSV export provides a manual workaround).

Verdict

Elicit earns our Best Value badge and the highest overall score among the tools we have reviewed because it delivers genuine, measurable productivity gains for the specific task that consumes more academic time than almost any other: reading, understanding, and synthesising research literature. The structured extraction capabilities are unlike anything available in general-purpose AI tools, and the free tier is generous enough that every researcher should try it. We scored it a 9 across all three dimensions because the ease of use is excellent (natural language search, intuitive workspace), the academic value is exceptional (purpose-built for scholarly workflows), and the price-to-value ratio is outstanding (the most capable free tier in this category). The coverage gaps in humanities and non-English literature are real limitations, and extraction accuracy requires verification — but for quantitative research in well-indexed disciplines, Elicit is the single most impactful AI tool you can add to your workflow today.