Elicit: AI Research Assistant and Its Paper Sources

Elicit is an AI research assistant. It searches academic papers and clinical trials, extracts data from them and writes reports with citations. Its paper corpus comes from Semantic Scholar, OpenAlex and PubMed. If you want a paper in Elicit, make sure that one of these databases lists it.

Paper sources

Elicit’s help article Elicit’s source for papers gives these facts:

• Elicit searches over 138 million academic papers from Semantic Scholar, PubMed and OpenAlex, in all academic fields.
• The sources overlap. Elicit removes incomplete listings from its results.
• Elicit updates the corpus every week. A new paper can take a short time to appear after it reaches the source databases.
• Users can also search ClinicalTrials.gov as a separate corpus.
• The corpus includes preprints from arXiv, bioRxiv, medRxiv, SSRN and other preprint servers. It does not include books or dissertations.

Elicit reads the full text of a paper when the paper is open access. It also reads the full text when the user has the Elicit browser extension and a journal subscription. In other cases Elicit uses the title and abstract.

PDFs that users upload stay private to their accounts. Elicit says that it does not add them to its corpus.

How the search works

Elicit’s methodology note describes four stages:

1. Embedding-based retrieval finds candidate papers by meaning, not only by keywords.
2. A relevance model scores and ranks the candidates.
3. The top 500 papers by score go to the next stage.
4. Elicit screens these papers against the inclusion and exclusion criteria of the task.

Research Agent and the web

The Research Agent can go beyond the paper corpus. Elicit says that it can search the broader web, ClinicalTrials.gov, regulatory documents such as FDA and EMA filings, press releases and product labels. It can also use the user’s Library and up to 20 uploaded files. Each claim in its output links to the paper or page that it came from.

Elicit does not document a crawler or a user agent for these web searches. It also gives no robots.txt token.

What it means for site owners

Ordinary web pages do not enter the Elicit paper corpus. The Research Agent can read and cite public pages such as press releases. Elicit documents no way for a site to opt in or out.

For papers, the path goes through the source databases. The Semantic Scholar guide covers one of them.