What they built.
Oxford PharmaGenesis researchers used Elicit to extract comparable information from a large set of research papers. Instead of reading each paper without a common structure, the researchers defined the questions they wanted answered and used source-linked extraction to organize the findings. People then validated those answers and performed the analysis.
Elicit's customer story names the researchers and describes one rapid review with 40 questions across nearly 500 papers. It separately describes a Monday-to-Thursday first draft for another assignment. Those are vendor-reported accounts of distinct work, not a combined or independently verified speed benchmark.
A named research team's workflow is documented by Elicit. It is not an independent accuracy or speed benchmark.
Read the researchers' workflowHow the tools work together
The papers provide the evidence, the researchers define the extraction fields and Elicit places relevant findings into a consistent structure with links to supporting material. The structure makes it easier to compare papers asking the same questions. Human validation is essential because an extracted value can be incomplete, misinterpreted or unsupported by the passage it points to.
Workflow at a glance
Define the questions
Gather the papers and specify which fields need to be comparable.
Extract into a consistent structure
Use Elicit's parsing and extraction with source-linked support.
Validate and analyze
Researchers check the extracted material and perform the analysis.
What the evidence supports.
The story describes one rapid review with 40 questions across nearly 500 papers. A Monday-to-Thursday first draft belongs to a separate assignment and is not combined with that scope here.
Elicit customer story SourceHow you could adapt the idea
The general idea can help any careful comparison of a document collection: define the fields first, keep the source passages and review ambiguous entries. The medical research example itself requires domain expertise and is not a self-diagnosis workflow. A reader should adapt the organization pattern without assuming AI can replace scientific judgment or reliably complete an entire systematic review.
Effort, cost and limitations
- Effort
- Specialist workflow requiring domain expertise, a defined extraction scheme and human validation.
- Cost
- No reproducible total cost is disclosed for the described work.
This is a research implementation, not medical advice or evidence that AI can replace scientific judgment or reliably complete every review.
Go to the original work.
Named researchers, examples and workflowSource-backed editorial account; PrisAIm did not run or independently benchmark this implementation.