Elicit organises research papers into evidence you can inspect

Elicit is an AI research platform for finding academic papers, extracting information and organising evidence into literature reports. It is most useful when the task is to understand what studies actually say, rather than collect general web answers. Its structured workflows can make the path from a research question to an evidence table easier to manage. They do not transfer responsibility for the search strategy, inclusion decisions or interpretation away from the researcher.

01 / ProductWhat Elicit does with research papers

Elicit’s Reports page¹ describes a workflow that finds and screens papers, assembles a summary and links claims back to passages. That source connection is the important feature to examine. A statement about a study is more useful when you can inspect the supporting text and determine whether it describes the population, method and outcome you intended to investigate.

The Systematic Reviews guide² adds explicit stages for gathering papers, abstract screening, optional full-text screening, extraction and reporting. A researcher can configure the question and criteria before the work runs. This is different from repeatedly asking a chatbot to improve an answer: the workflow gives the inclusion and extraction decisions their own places to be inspected and revised.

According to Elicit’s corpus documentation⁴, its paper sources include Semantic Scholar, PubMed and OpenAlex, with different update schedules across research features. Coverage and full-text access are separate issues. Finding a citation does not establish that Elicit has the complete methods section, supplementary material or a paywalled table. Check the underlying document whenever your conclusion depends on something beyond the abstract.

02 / AudienceWhich research jobs fit Elicit?

Elicit is a plausible fit for literature discovery, scoping a new research question and extracting comparable fields from a set of papers. A product team might investigate whether a training technique improves retention; a researcher might organise outcomes and methods across an existing study collection. The common feature is a question whose answer depends on published evidence rather than a company’s current website or an uncited opinion.

For a broad web briefing about companies, product announcements or public policy, Perplexity represents a different research interface. Elicit’s attraction is the paper-centred workflow, particularly when you need to track which studies were included and why. A general search answer and a literature synthesis can both contain citations while providing very different coverage and auditability.

It is a less direct fit for questions that cannot be expressed through the available literature, or for projects where your institution requires a specific database search and controlled screening process. Elicit can still assist part of such a workflow, but the research protocol remains the authority. For a custom application that needs retrieval as a programmable component, Exa’s APIs illustrate a separate build-versus-buy choice.

03 / WorkflowA worked workflow for an evidence table

Consider an illustrative research task: compare published studies of spaced practice in professional training. This is a proposed workflow, not research we performed in Elicit. Define the eligible population, training setting, comparison and outcome before asking for a summary. Decide whether you want a quick orientation or a formally documented review. Those choices determine what evidence needs to be retained and which steps require independent checks.

Begin by writing inclusion criteria that another researcher could apply consistently. A study of school pupils may discuss the same learning technique but fall outside a question about workplace training. A paper that measures satisfaction may not answer a question about delayed retention. These are substantive research differences, not merely alternative keywords. If the criteria are vague, an apparently complete evidence table can combine incompatible studies.

Use the Systematic Reviews setup² to configure the gathering and screening stages. Inspect the proposed search and supplement it with known relevant papers where your protocol permits. Keep a record of those additions. A tool that retrieves several expected papers may still miss an important part of the literature, so the presence of familiar references should be treated as a starting check rather than proof of completeness.

Elicit’s full-text screening documentation³ explains why a second pass can be necessary: the complete paper may reveal eligibility details absent from the abstract. It describes Fast and Thorough depths with different time and usage implications. Choose the depth for the research purpose, then inspect borderline exclusions. An incorrectly excluded study will not be rescued by a beautifully written final report.

Define extraction columns that preserve meaning: study setting, intervention, comparison, outcome definition, follow-up and the location of supporting text. Keep missing information distinct from a negative result. If a paper does not report follow-up, the table should not imply that follow-up was absent. Likewise, a model’s interpretation of a passage should remain distinguishable from the wording reported by the authors.

Before writing conclusions, export and inspect the evidence outside the interface. The export guide⁷ distinguishes report downloads, table exports and reference-manager formats, with access depending on the workflow and plan. Keep the screening decisions and extraction table alongside the final narrative. They allow a colleague to challenge the synthesis without having to reconstruct every interaction with the assistant.

04 / PricingElicit pricing depends on your audience and workflow

The pricing page⁶ has both an audience selector and a billing selector. The table below records its Industry view, checked in the browser on 15 September 2026. Amounts are USD per user. Academic offers differ, so a low price shown in another view should not be treated as an industry quote.

The same page positions Pro around systematic research and Scale around collaboration and higher usage. Treat those as workflow choices, not just larger quotas. Write down whether you need collaborative editing, figure extraction, screening exports or account administration before selecting a subscription. A plan that supports the initial search may not include the later stage required to deliver your project.

Research cost also includes reruns and validation time. Revising an eligibility criterion after extraction can require substantial rework, even if the software performs it quickly. For an evaluation, track the time spent defining criteria, resolving disagreements and fixing extracted fields. The practical budget is the cost of a defensible final synthesis, not only the cost of producing the first report.

05 / DistinctionsThe distinction is inspectable evidence, not answer length

A long literature summary can feel authoritative while hiding how its papers were selected. Elicit’s structured approach is interesting because it exposes more of that selection and extraction process. Our interpretation is that the strongest value comes when a researcher uses those intermediate outputs actively: revising a criterion, checking a supporting passage or noticing that an outcome was defined differently across studies.

The same distinction explains how to compare alternatives. A reference manager helps organise citations and documents. A database provides retrieval over a defined collection. A general assistant can explain or draft text. Elicit combines parts of discovery and evidence synthesis, but a project may still need the other tools. Evaluate which step causes the real bottleneck before replacing a familiar research stack.

For founders studying the product, the useful design lesson is to make consequential intermediate decisions visible. In a literature workflow, inclusion, exclusion and extraction shape the final answer. Showing only the finished prose would conceal the places where an expert can add the most value. This principle extends to other AI products whose outputs depend on a chain of judgments.

06 / QuestionsWhere human judgment still matters

Does a citation establish that a finding is trustworthy?

No. Elicit’s own limitations page⁵ acknowledges possible mistakes in nuance and numerical interpretation, and distinguishes summarising a paper from judging its quality. A study can be accurately summarised yet provide weak evidence for your decision. Review design, relevance and the strength of the underlying evidence separately from the accuracy of the extraction.

How should you evaluate accuracy?

The evaluation guide⁸ recommends comparing Elicit with an existing project or a sample of work and examining disagreements. Our suggested acceptance exercise is to include difficult cases deliberately: a result reported only in a figure, several outcomes with similar names, and a paper missing an important field. Resolve differences against the paper itself. Do not assume either the human entry or the model output is automatically correct.

We have not run a benchmark, replicated a systematic review or checked Elicit against a private paper collection. This article analyses the public workflow and commercial material. For work intended for publication, document how AI contributed and follow the research requirements of the institution, journal or project. Keep the original papers and decision trail so that later corrections can be traced to their source.

07 / DecisionChoose the amount of research control you need