Real example · Study, summaries & research
Tested Elicit against four published evidence reviews
What they did
Two UK health researchers ran Elicit's review mode on the questions of four published evidence syntheses (on vaping, physical activity, fertility treatment and breast reconstruction) and compared what it found with the original searches.
How it went
Elicit's sensitivity was 'averaging 39.5%' against '94.5%' for the original searches, but its precision was higher (41.8% vs 7.55%) and it found 1 to 4 relevant studies per review that the originals missed.
Worth knowing
Peer-reviewed test of one Elicit version; the authors say it is not yet sensitive enough to replace traditional searching but is useful for scoping and as an extra search, and they did not measure time saved.
Try it yourself with Elicit
Replace anything in [brackets] with your details.
I want to check how well an AI search tool finds the studies a published review included.
Research question: [question from a published review].
Find the most relevant studies on this question, and for each one list the design, sample size, population and main finding.
Then compare your list with the studies in [link or title of the published review]: show which ones you found, which you missed, and which extra ones you suggest.Read the original
Comparison of Elicit AI and Traditional Literature Searching in Evidence Syntheses Using Four Case Studies (Cochrane Evidence Synthesis and Methods) ↗
pmc.ncbi.nlm.nih.gov · · Peer-reviewed study
Attributed to the source linked below; not our own test.
Share Share on X Share on WhatsApp
