Citing Elicit, search methodology, and using Elicit's content in your own work

When using Elicit for your research, we recommend citing it as you would any other source. Example citations are provided at the bottom of this page.

We do NOT recommend copying/pasting Elicit content verbatim into your work. To minimize hallucinations, Elicit's content stays as faithful to the source text as possible, so some summary text may be quoted as published. Elicit's content is intended to enhance and support your understanding of your research topic. It is not intended to be used as your own text.

Additionally, like any AI tool, Elicit is subject to AI writing detectors. Pasting Elicit content into your work may get your paper flagged as AI-generated, depending on the policies of your school or organization.

@software{elicit,
  author = {{Elicit}},
  title = {Elicit: The AI Research Assistant},  
  url = {https://elicit.com},
  year = {2023},
  date = {2023-01-24},
}

Elicit; Elicit: The AI Research Assistant; https://elicit.com; accessed xxxx/xx/xx

All Elicit citations should include the URL and the date. Check out example papers and articles where Elicit was cited here: Works about Elicit

Elicit's literature searches are conducted using an Elicit's LLM-powered semantic search system. This methodological approach proceeded in multiple stages: (1) initial semantic embedding-based retrieval using neural language models to identify candidate papers based on conceptual similarity rather than simple keyword matching; (2) relevancy scoring and ranking of papers within the candidate pool using a transformer-based relevance classification model trained to evaluate topical alignment; (3) selection of the top 500 papers by relevance score for further analysis; and (4) an explicit screening phase where papers were systematically evaluated against inclusion/exclusion criteria to determine final eligibility. This methodology leverages the capabilities of large language models to comprehend complex search requirements beyond traditional boolean keyword approaches, resulting in improved retrieval recall and precision. The re-ranking algorithm applied to the initial candidate pool employs contextual understanding of each paper's content to prioritize those with highest topical relevance, substantially increasing the concentration of pertinent literature in the analyzed corpus compared to conventional search methodologies.