It’s Digital Health Week! Running from November 17 – 23, Digital Health Week is an annual event stewarded by Canada Health Infoway exploring current challenges, solutions, and developments in digital health. In recognition of the week, we’re spotlighting an AI tool that you may have already encountered in your projects or conversations with other academics: Elicit.
Elicit is an artificial intelligence product developed by Ought, an AI research lab based in San Francisco, California. Elicit is a public-benefit corporation/AI tool, marketed as an “AI research assistant”. The tool is designed to automate the process of knowledge synthesis and assist in conducting literature reviews. It launched in early 2025 with its “start a systematic review” feature.
How does Elicit work?
Elicit works with Semantic Scholar, a free AI-powered academic search tool (comprising both a search engine and a database of academic papers). Elicit uses machine learning models like GPT to comb through ~125 million papers stored in Semantic Scholar’s database, returning results for articles relevant to the search query or question, with an AI-generated summary of key points across the articles retrieved, as well as an abstract summary for each individual article. The tool is capable of scanning both body text and diagrams/tables to source these key ideas.
The tool can support building a systematic review by giving users the ability to import papers from the Elicit search function and from previous systematic reviews, as well as PDFs from your desktop files.
In terms of pricing, Elicit offers plans ranging from $12 to $79 per month, plus custom enterprise options. There is also a free “basic” plan promoted for “casual exploration”, but this may not provide the functionality needed to support an in-depth review.
What is it good for?
As of this writing, Elicit is mainly relevant to health sciences librarians working with health professionals.
The tool is likely useful for getting a general sense of the scope of existing literature, or for locating “seed” articles — foundational or influential articles in your target subject area that can serve as examples of literature that meets your inclusion criteria. Seed articles can help point you in the direction of more high-quality literature, and help refine your overall search.
As these use cases indicate, Elicit may be ideal for tasks that are relevant very early in the initial planning and scoping phases of research.
What are the risks?
Because Elicit is an AI-powered tool and thus inherits the risks of bias and inaccuracy that are baked into many machine learning and large-language models, caution and critical thinking should always be applied to any results it produces.
A noted issue with most generative AI models is hallucination, which is the tendency for AI to invent evidence and sources to support a user query, where no such evidence or sources exist.
Elicit uses a technique called Retrieval Augmented Generation (RAG), which augments the generation of key-points summaries by sourcing information from an indexed database (in this case, Semantic Scholar). RAG helps to reduce the instance of hallucinations by narrowing the field of sources that the AI can use in generating responses. That said, RAG may not entirely cure hallucinations. AI tools are designed to supply responses to queries, and training an AI model to appropriately indicate uncertainty in a response is a very tricky thing. As such, responses offered by Elicit should always be evaluated for accuracy and alignment with the content of the sources.
The bottom line
Elicit may be a helpful tool to save time in the early scoping phases of a literature review. Like any AI tool, it should only be used in concert with human validation and critical thought.