共有:

Elicit · Accuracy

Elicit Reaches 95–99% Accuracy on Systematic Literature Reviews

Six months ago 80–85% was the AI screening ceiling. Elicit has now crossed the bar for clinical trial systematic reviews, with FDA workflow conversations likely next.

AI Navigate Editorial·2026.06.27·6 min read

Accuracy metrics (994 Cochrane reviews) 95% Search recall 99% Full-text screening 96% Data extraction Prior AI ceiling 80–85%
01
What Changed

AI Screening Crosses the Clinical Trial Bar

Six months ago 80–85% accuracy was considered the ceiling for AI-assisted literature screening.

PRISMA 2020-compliant and auditable; 95% search recall, 99% full-text screening, 96% extraction on 994 Cochrane reviews. For the first time, the accuracy threshold required for clinical trial systematic reviews has been cleared by an AI tool.

02

The Accuracy Numbers

95%
search recall
99%
full-text screening accuracy
96%
data extraction accuracy
994
Cochrane reviews validated against
03
Who Should Act

Which Organizations Should Move Now

01

Pharma R&D: pilot in clinical protocol evidence review

PRISMA 2020 compliance aligns with regulatory submission formats. Elicit can now be considered for evidence structuring in FDA-bound documentation.

02

Universities: automate first-pass screening for guideline-compliant reviews

Route initial screening through Elicit, then confirm manually. The switch can cut weekly effort significantly on literature-heavy research projects.

03

Evidence synthesis outsourcers: model the cost of insourcing

At 95–99% accuracy, outsourced literature screening becomes a candidate for bringing in-house. Run a cost model before the next contract renewal.