The workflow
Import your search results
Create a project and upload the RIS file from your database search. Projects and import →
Write your criteria
Set what must be included or excluded, using keywords or Boolean rules. Criteria →
Screen
Make include, exclude or maybe decisions one article at a time. Screening →
Run Huxley AI
Let the AI screen the remaining records against your criteria, with a reason and confidence for each. AI screening →
Review and export
Check AI decisions, override where you disagree, and export the result as RIS. Review and export →
Decision types
Every decision records who made it: manual (you) or Huxley (the AI), so you can always filter to check the AI’s work.
Using Huxley well
Screen a sample yourself first
Screen a sample yourself first
Screen 50–100 records by hand before running the AI. It shows you whether your criteria are clear, and gives you something to check the AI’s decisions against.
Write criteria the AI can apply
Write criteria the AI can apply
Be specific: “randomised controlled trials in adults aged 18+” works better than “good quality studies”. The AI only sees the title and abstract, so criteria that need the full text belong at the next stage.
Treat the AI as a second screener, not the only one
Treat the AI as a second screener, not the only one
Review every AI exclusion with low confidence, and spot-check a sample of the confident ones. Report in your methods that AI-assisted screening was used and how it was checked.
Lean towards Maybe
Lean towards Maybe
Turn on Maybe decisions for the AI. Sending a borderline paper to full-text review costs a few minutes. Wrongly excluding a relevant paper weakens the review.

